Stetson’s review of Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI > Likes and Comments
146 likes · Like
This reads like it was written by ChatGPT.
I frequently use AI tools to provide editorial support on my writing (as everyone should), but the language here is my original thought.
What exactly informs your claim here? I think there are political sentiments in my review that are usually pretty hard to get any AI tool to spit out.
cracking up at how chatGPT generated this reads. the irony delights me. talk about disappointingly shallow! 😂😂😂😂
Emily wrote: "cracking up at how chatGPT generated this reads. the irony delights me. talk about disappointingly shallow! 😂😂😂😂"
Which aspect of this do you believe is chatGPT generated or reads like genAI writing?
What is shallow about providing criticism of a book that is a sloppy assembly of previously written journalistic pieces that are whining that the big baddies in tech are the new "colonialists" and are building an evil empire?
Do you honestly believe that the developing world or "Global South" as Hao likes to say really has contributed anything of significance to the advancements in AI in recent years? If you do, I'd like you to invest in my new B2B SaaS startup....
🙃 discounting the environmental and labor contributions of the Global South… i know what you are
Lanae wrote: "🙃 discounting the environmental and labor contributions of the Global South… i know what you are"
What am I then? Generally, to be persuasive, one should be clear about what they mean. Insinuation is ultimately ineffective as a rhetorical strategy.
What did the "Global South" actually contribute here? Basic data annotation and RLHF is not eminently skilled work here. In many ways, it is a generous largesse these populations get to enjoy as they wouldn't otherwise be getting the investment. Hao, like you, fails to consider the alternative.
Thank you for the review Stetson, you've made a compelling case for this book is just not good.
As for the detractors on this thread, I don't think you guys are engaging with the points made in the review. All your critiques are "haha he used ChatGPT", "i know what you are." This adds nothing to the discussion. Please read the book before trying to be so dismissive.
Umar wrote: "Thank you for the review Stetson, you've made a compelling case for this book is just not good."
Thanks, Umar! Glad you appreciated the review.
Simon wrote: "Oh dear, an AI review ie fake."
There's nothing fake about the review. This is my opinion of Hao's book, which is mostly a loose collection of her reporting over the last few years.
The book is more of a polemic than actual journalism on OpenAI or genAI broadly. Additionally, the angle of the polemic is childish. The AI revolution has little to nothing to do with the "Global South" and is certainly not any form of "colonialism."
Unlike the hyenas in this thread, I’ve actually read the book. This is spot on. Very disappointing.
PCR wrote: "Unlike the hyenas in this thread, I’ve actually read the book. This is spot on. Very disappointing."
Thanks, PCR!
Thank you for this review. I simply could not agree more with the majority of the points you made. My biggest frustrations were:
1) Hao’s analysis & criticisms of the impacts of AI are not unique to AI in the slightest, they’re simply a critique of capitalism/ how the world (yes, unfortunately) operates...
2) Hao fails to adequately and impartially address the potential long-term upside of AI and whole (very persuasive, in my view) case for why there is so much investment going into it. It is truly unlike any other technology to have ever existed.. her tone around the technology comes off as transparently biased. Just really disappointing for someone with such an insider’s view to be lacking in their depth of analysis and objective insight.
Thanks, Tommy! Yes, even if she wanted in her heart of hearts to eviscerate Sam, openAI, and AGI the technology, it would have had a better chance if she just simply did incredible objective reporting.
She lost access because they sniffed out her agenda early and thus she didn't have much (unique) to add despite her experience in the AI beat and her chance at persuading people is low because her POV is overt and imo absurd.
Pretty much agree but I scored it a 4 because of the muckraking and abundant detail. https://www.goodreads.com/review/show...
Hi John, reasonable take! I didn't think Hao should get credit for double dipping. She did contribute real reporting on OpenAI but that reporting was available prior to the book. She should get credit for those things. The score of my review pertains just to the book's failure to deliver on its own terms and the unnecessary and sophomoric polemicism. Hao appears plenty capable as a reporter.
I work in AI and I actually share a lot of her baseline concerns, I think Sam Altman is one of the craziest and most dangerous people running around right now. But this book really got on my nerves with a lot of the obsessive hand wringing over like, equity and marginalized communities and colonialism and just sort of the whole leftist lens it was filtered through. I think when you analyze anything this broad and impactful through a heavily partisan lens with, like you said, an obvious disdain for the technology itself as opposed to maybe the way it's being handled, you're not giving an honest or complete view.
All great points, C. I'm less moved by the critique of Altman, though there are certainly reasons to be concerned about his character based on the reporting. If the book was just the reporting on OpenAI internal dynamics and Altman I would improve my rating of it.
Thanks, good review. Does "academic post Colonialism" always mean anti-capitalism?
No, she argues that AI development is "digital colonialism" exploiting and pillaging the "global South." I wouldn't say there is much in the way of a critique of capitalism other than within this post-colonial framing.
A “patchwork of grievances” is the perfect description and tracks with my experience. I had a growing sense that Hao has a huge, unnamed personal chip on her shoulder which motivated much of her approach.
It’s disappointing because I believe she touches on many important issues which deserve a fair hearing, but I can’t trust her impartiality.
Thanks for such an articulate and extensive review.
Ok wait, but what’s the justification for the scaling, exploitation and waste of resources? I fail to read how we don’t care about these things, what’s the great gain from AI which grants the sacrifices being made? I haven’t read the book yet but if you have a suggestion I’d be happy to read more books to understand you better.
Andrea Maria wrote: "Ok wait, but what’s the justification for the scaling, exploitation and waste of resources? I fail to read how we don’t care about these things, what’s the great gain from AI which grants the sacri..."
Hi Andrea, the justification for scaling and AI capex generally is that there is a strong expectation that the technology will lead to increased productivity across the economy. The extreme version of this vision is a post-scarcity future where human labor is not an essential input in the economy. This vision is unlikely to come to fruition but increase productivity will make everyone wealthier and healthier.
The tradeoff between AI capex and energy usage is not particularly concerning given that energy usage is going to increase if the economy is going to grow anyway. Additionally, it is possible, if not likely, that advances in AI technology may increase the efficiency of energy system or transition energy systems away from reliance of carbon sources. Much of this is already within reach given advances in solar, battery tech, and untapped nuclear sources. It is largely a failure of political will that keeps a country like the U.S. from running primarily on nuclear energy and battery tech.
There are many foundational works on general AI technology but most of them are not in book formats. They're in research articles and white papers. Here are two useful examples:
- "AI 2027" - white papers on the intelligence explosion
-"Attention is All You Need" - Transformer architecture paper
-AlphaFold Paper - "Highly accurate protein structure prediction with AlphaFold"
If you're looking for books that are for general readers but are more positive about AI then you can read:
-Co-Intelligence by Ethan Mollick (mostly about specific use cases for individuals)
-On the Edge by Nate Silver (not specifically about AGI but covers major tech players and industry)
-Superintelligence by Nick Bostrom (a philosophy book on the alignment problem)
There are also journalistic/commentary sources that report on the potential and current impact of AI:
"Understanding AI" by Timothy B. Lee on Substack
"The Cognitive Revolution" podcast from Erik Torenberg
so your main criticism of this book is that the author doesn't spend enough time praising AI? got it.
Rivara wrote: "so your main criticism of this book is that the author doesn't spend enough time praising AI? got it."
No, my primary criticism is that the central framing of the book is ridiculous (i.e. AI capex in data centers and labor outsourcing in the developing world =/= imperialism nor "data imperialism"). The lesser criticism of the singular pursuit of AGI rather than other types of AI is not trenchant. One, there is plenty of other major AI projects that are not focused on AGI (e.g. AlphaFold). Two, genAI has already seeing plenty of success even if the rate of progress is plateauing.
Additionally, the reporting itself leaves much to be desired. Hao doesn't focus on the actual newsworthy events at OpenAI in sufficient detail and instead wastes time trying to provoke the audience's ire concerning Sam Altman. This includes taking obviously unnecessary cheap shots like retreading Altman's troubled sister's accusations. The original reporting that Hao actually did is also just recycling stuff she's already published, whereas a book like this would be an opportunity to debut some actual news.
As I said in the review itself, I walked away from the book without really learning anything new about Altman's candor and its relationship with his brief ousting. Altman is portrayed as Machiavellian and somewhat two-faced or just a tell-people-what-they-want-hear kind of guy, but none of his individual actions looked especially shocking in the scope of what's par-for-course among tech CEO in an emerging sector.
I thought an additional weakness was the Hao provides very little conceptual and technical detail concerning the field of artificial intelligence. It is a capacious domain that is not particularly young. Many people are re-discovering it in the post-ChatGPT world, but there is a long intellectual history that is relevant to the very subjects she reports on. She provides the most 101 of 101 type coverage.
This is a deeply biased review. Let’s take one example. You claim she treats comparisons between AI companies and colonial empires as self-evident and avoids examinations. And yet she offers abundant detail about the harms done to AI contractors in the global south. Her accounts moreover align with similar investigations conducted by other AI critics in the field.
Your criticism that her investigation of Sam Altman is too thin is risible considering he and OpenAI refused to speak with her for the book.
Your attacks on her for not discussing on the entire history of AI development or its many contributors is just weird. She’s not writing an 800pg history book about everything that ever happened in AI. She made reasonable choices about how much industry backstory to include or not.
Tbh you sound like your wrote your review before you read the book. Why you feel the need to be w white knight for the most valuable private corporation on the planet is beyond me.
Andrew wrote: "This is a deeply biased review. Let’s take one example. You claim she treats comparisons between AI companies and colonial empires as self-evident and avoids examinations. And yet she offers abundant detail about the harms done to AI contractors in the global south. Her accounts moreover align with similar investigations conducted by other AI critics in the field."
Reviews are by definition biased; they're an expression of a personal opinion. What you mean to say is that the evidence for the argument on which my opinion sits is flawed in someway. However you believe it to be flawed is not well articulated. You simply assert (without evidence) that she has a bunch of evidence and that others (with the same worldview) agree with her...
This entirely misunderstand my critique as there is no world I could conceive in which there's any evidence for the idea that AI companies are functioning like the Dutch East India Trading Company. It is a motivated category error. AI companies don't have military forces, don't have the imprimatur of the U.S. government, and don't exercise any substantial coercive powers. Everything they do some other party has mutually agreed to.
Andrew wrote: "Your criticism that her investigation of Sam Altman is too thin is risible considering he and OpenAI refused to speak with her for the book.
Your attacks on her for not discussing on the entire history of AI development or its many contributors is just weird. She’s not writing an 800pg history book about everything that ever happened in AI. She made reasonable choices about how much industry backstory to include or not."
My views on how she treated sama and AI, the technology, generally are simply my opinion. I'm entitled to it as you are entitled to disagree. Plenty of people also see it the way I do.
Plus, If you think it's totally cool that Hao parades around Sam's mentally disturbed sister as if that's relevant to his stewardship of his company then you're probably a lot less impartial about these subject than I am.
Andrew wrote: ""Tbh you sound like your wrote your review before you read the book. Why you feel the need to be w white knight for the most valuable private corporation on the planet is beyond me."
By the way, not only did I read the book, but I also listened to many of Hao's interviews in which she also maintains her ridiculous ideological position, appearing like a sophomoric activist rather than an actual journalist.
Additionally, it is unclear why an American would be particularly opposed to OpenAI. It is a wildly successful company who will undeniably be of strategic and material value to the U.S. taxpayer. In fact, this reality may indeed entirely account for Hao's desperate and silly effort to politicize the company among Americans.
I'm sorry, but your review is just so narrow minded. I see that you expected a) an in-depth technical analysis of what OpenAI has worked on and b) an in-depth analysis of OpenAI as a company. Why not take a step back and have a look at what the author is trying to raise an alarm over: the recklessness of releasing AI to the general public without a proper structure or proper consultation with governments to understand and mitigate the impact over the entire population. She is trying to draw a parallel between how the company started and its initial ethos (do no harm) and what it ended up being, which is yolo-ing its way through, despite having the Facebook and the Internet as models of what could go wrong if people in charge release features without thinking about the consequences (e.g. spread of child pornography, copyright infringement, etc.).
This moment in AI evolution looks similar to the evolution of gene studies. If you remember the Dolly experiment, this was considered highly unethical in the medical field and similar experiments were banned in multiple countries, or at least funding human cloning experiments was not allowed. While the medical field has the Hippocratic oath to protect human beings, the IT industry does the equivalent of releasing experimental drugs to the general population without any sort of oversight. If you would have been more open minded while reading the book, this is one of the aspects of the AI industry that the author tries to highlight.
Maybe you should go back to reading papers about neural networks if that is your interest and the ethics of AI is something you don't consider important. Don't rate a book a 1 star because it didn't meet your preconceived expectations. It seems like you read the book like a professor grading a student's paper on a topic they are teaching. Maybe try getting out of your comfort zone and consider another point of view on the topic while reading a book.
My primary complaint about Hao's book was that she tried to force the business decisions made by sama/OpenAI and the field of genAI in general into a nonsensical frame (i.e. that scaling dogma downstream of AGI worship leads to exploitation of the resources of the Global South). The idea that AI capex in data centers, pre-training on open Internet data, and outsourcing RLHF is a colonial project is laughable. Great power competition is much more relevant to understanding and commenting on AI development than third-rate Said/Fanon poasting (what Hao is doing here).
My concerns about the book's superficial treatment of the technical aspects of genAI were a secondary critique. The same is true of the reporting on OpenAI itself. These lesser issues simply underscore how poor the overall work is. She re-heated her old reporting, added nothing actually new or useful, and threw in other random journalistic work she did.
I don't think my expectations for the work were that high. If she just included the sama OpenAI drama and some general stuff about the AI industry then it would be like a 2.5-3 star book, but she ruined it with an ideological thesis meant to get aging far left-wing luddites worked up.
If you are so favourable to chatgpt that you think everyone should use it, then you weren't prepared to read this book. You didn't even try to be objective.
Rohit wrote: "If you are so favourable to chatgpt that you think everyone should use it, then you weren't prepared to read this book. You didn't even try to be objective."
Is this an argument? Perhaps genAI could help you formulate one.
As one of my prior comments states, book reviews are inherently subjective, meaning they are definitionally an expression of one's biases. If you want to attack the review, you have to show their is a specific weakness in the reasoning I've presented not take issue that my tastes are different than your tastes. Why should my tastes be the same as yours?
“Reviews are by definition biased”… that made me laugh. 😆 Shame on you for not subscribing to a particular viewpoint! I’m about 3/4 of the way through and I’m finding a lot of the same problems with the book that you mentioned in your review. It just seems more like a hit piece on Sam Altman and OpenAI, rather than any real balanced view of Sam and his company. So many parts detailing threats that data centers and supercomputers have brought of exploitation to third world countries’ natural resources and Indigenous populations are speculative - talking about misdeeds that Microsoft and Google have committed, yes - but rather than acknowledge it as a potential problem Open AI may also replicate, she paints it as a foregone conclusion. Sure, there’s obviously an argument to be made for what Google and Microsoft have done in Chile, but OpenAI? Not quite.
And while I know it makes me sound like a horrible person to many to disregard sexual abuse allegations, I just can’t take Annie Altman’s allegations seriously. So we’re supposed to believe that their entire lives, Sam and Annie were incredibly close, she viewed him as her protector, etc.? Yet suddenly once OpenAI and ChatGPT become huge, Sam becomes a billionaire, she suddenly files a lawsuit saying he sexually abused her from the age of three?
If abuse did occur at that age, it would be extremely difficult to identify the perpetrator(s). Your memory at 3-5 years old, or whatever, isn’t really advanced enough to warrant the types of allegations she’s making, and furthermore, if trauma indeed did occur, people often fully believe something to be 100% factual that, in reality, never happened (this actually happened with me through the loss of a loved one - while I was there. I remembered things and insisted I’d seen them, the image was IN MY MIND - only to realize that these things I believed weren’t logically possible).
I don’t know how great her artistic skill is, but it doesn’t seem feasible for even the best artists to live in the Bay Area on income generated solely from their artwork (whatever type of artwork she was doing, it didn’t bother to mention it). So it seems very likely that she is being financially irresponsible if she’s living in the most expensive area of the U.S., investing her money into art, doing artwork for a living, yet claiming she never expected Sam or her family to provide for her. He buys her a house but then she’s mad that he stops payment on it once the sexual abuse allegations are printed for the entire world to see? 🤦🏼♀️ If it really did happen, that’s awful, but it just doesn’t add up for Sam to have been her main abuser, or have abused her at all. Ruining someone’s reputation with false or inaccurate claims is awful, too.
Megan wrote: "I’m finding a lot of the same problems with the book that you mentioned in your review. It just seems more like a hit piece on Sam Altman and OpenAI, rather than any real balanced view of Sam and his company."
Agreed, this is an angle I didn't spend much time highlighting in my review because I can't speak to the specific motivation behind Hao's perspective on Sam Altman and OpenAI. However, it is clear there is animus there for some reason... I could speculate on a few, but it would be a bit irresponsible so I've demurred from that course.
Megan wrote: " So many parts detailing threats that data centers and supercomputers have brought of exploitation to third world countries’ natural resources and Indigenous populations are speculative - talking about misdeeds that Microsoft and Google have committed, yes - but rather than acknowledge it as a potential problem Open AI may also replicate, she paints it as a foregone conclusion. Sure, there’s obviously an argument to be made for what Google and Microsoft have done in Chile, but OpenAI? Not quite."
I see capital expenditure in the developing world or outsourcing of RLHF positions to English-speaking foreigners as almost an unalloyed positive for those countries and people. It is investment and a cash influx that otherwise would not be there at all. Instead of Latin America and Africa, the tech companies are pivoting to the Middle East instead because of the self-defeating political opposition from people on the ground. It reminds me of the time AOC worked so hard to keep Amazon from setting up HQ#2 in NY - complete self-own.
Megan wrote: " I just can’t take Annie Altman’s allegations seriously."
I think Hao in her heart of hearts absolutely knows that Altman's allegations are part of an effort to continue to extract money from her brother. It is evident from even the information that Hao presents while trying to use the story to impugn Sam that Annie has a lot of mental health struggles and has already received a great deal of support from Sam and the rest of the Altman family.
I don’t see much of anything about regulations or unregulated practices or governance at all. Those ideas were pretty large takeaways for me, which have some pretty heavy moral implications. What are your thoughts?
I'm not an AI doomer. I think the x risk is infinitesimally small.
I'm not sure where in the book you see Hao dedicating much focus to governance or policy. There isn't even that much focus on the export controls on chips, which is usually the primary focus of AI-related regulatory discussions or philosophical musings about the alignment problem and intelligence explosion. She of course expresses a normative desire for pre-training input (copyright), environmental and financial regulations, which are all things if generally oppose. I don't really see any socially pressing issues there. It's also strange she's that concerned about it since she allegedly doesn't think the technology is going to go anywhere (basically the Marcus critique).
You'd have to be more specific about what you see as particularly urgent and morally grave.
You probably wrote this with chatgpt
Bobbydigital wrote: "You probably wrote this with chatgpt"
You can plug this into AI text checker if you like, Grammarly's tool says my review is only 4% AI generated.
I often use genAI to edit, usually in a paragraph by paragraph fashion, but clearly this review is overwhelmingly my wording and my perspective.
Clearly my review has touched a nerve with folk. If it was just AI slop that's triggering them, isn't that a reflection on them?
I had the same thought as Bobbydigital - the use of AI in your writing makes it hard to read. “I’ve demurred from that” and “infinitesimally small” are not real phrases people use. I’ll add an extra star to my review because at least the author wrote it.
You should read more of my writing. I often use the word "demure," and I'm a geneticist where the infinitesimal model is the basis for complex traits.
Clearly, the readability of my review is decent, and it's one of the most popular reviews of this book here at this site.
hi stetson,
you seem unable to actually engage with any of the criticism of the AI industry she presented, or its real impact on human lives.
Also it’s a bit misleading to say Hao is hostile to AI as a technology which is just categorically false. She critiques the current nature of the industry and those who are leading it, not the actual technology. She often highlights the actual goal of researchers (at openai!) and their seemingly pure intentions and discusses real world examples of sustainable AI development and usage (like the group in New Zealand)
Also she did not spend a lot of time on Ann Altman, that was literally such a small part of the book. not sure why you chose to mention that but not any of the much larger discussions in the book like the environmental toll, loss of transparency that contradicts the main ethos of openAI, etc.
You disregard the exploitative nature of the AI industry’s presence in the Global South by claiming these low wages are better than no wages. Which is a pretty childish point of view and shows you didn’t really engage with this book in any meaningful way.
These companies are normalizing and setting these standards, meanwhile their executives are getting paid millions. They have billions in funding and are being bankrolled by the biggest companies, but it’s okay for workers in developing nations to be paid scraps and wait for pings at midnight because that’s the only way they can afford to eat? Not to mention their actual job is being exposed to the most vile stuff the internet has created.
Which leads me to your point saying Hao just assumes the industry to act as an empire rather than analyzing it … which is a complete misread imo. Like the whole thing is her diving into why the industry and openAI specifically act as a colonial empire by analyzing their actions and infrastructure. She doesn’t make any claims without evidence.
Also your wording of how she expects us to be “enraged” by the responses from execs seems like a case of projection. that is not at all how it came across to me as a reader, it feels like this is a narrative that you’re putting onto her. Hao’s main discussions are the systemic failings of the company that thus make it run as a profit seeking empire fuelled by extraction of labour and ideas. None of this is assumed, she deeply discusses and interviews people who are victims and also those responsible for it.
It’s pretty clear going into this book you had already formed your opinions about the whole matter and the author itself. So it’s interesting that you don’t see the hypocrisy in you ending the review as if this book is some self affirming exercise when there exists no nuance in your review or opinion.
Meha,
I explicitly address the core claim made by the sloppy book. Hao argued genAI companies and capex are "data colonialism" that exploits the global south. The evidence she tenders is very thin: 1) remote workers in developing countries doing challenging work and 2) data center water use. The former is stupid as those workers are voluntarily participants and are literally better off for the opportunity to earn money. The latter is not factually supportable. There is a beautiful debunking of the water use claims (Google it) that specifically shows Hao is either lying or is wildly ignorant about these facts.
You're right Hao is suspiciously just hostile to America's premier AI firms and strategy. Hmmm... Makes ya wonder. Why could that be!?
You have the same incorrect economic model. Working for a wage is almost always materially and socially superior to unemployment. If the job is so bad, they can voluntarily not do it!! Being a mature economic actor means navigating tradeoffs.
Do you know what colonialism is? It is when a state sponsored entity goes (with guns) and conquers a foreign land and appropriates the resources as they see fit. This is not even metaphorically similar to capex in foreign countries and voluntary market exchange. If anything it's the precise opposite of colonialism.
I've listened to many interviews of Hao concerning her book and followed where those interviews happen, which itself is a signal. It is clear how she wants people to react. It's also plainly there as subtext or actual text in her book.
I had no idea what to expect going in other than I should get some details on Altman's leadership of openAI and get detailed reporting on why he was briefly deposed. Hao didn't add anything new to this story that wasn't already available. Those events still seem largely inexplicable and an amazing self-own by those involved. I also credit her for whatever original journalism she actually did on altman and openAI but there wasn't much original material in the book.
There isn't any hypocrisy embedded in having a strong opinion, especially one supported heavily by empirical reality and logical analysis. I'm not criticizing Hao for hypocrisy. I'm criticizing her for trying to pass off dorm room polemic as straight journalism. I'm also criticizing her for not doing her homework. It was lazy and sloppy given how much she gets wrong or facts she fails to produce as part of her overall thesis.
"Being a mature economic actor means navigating tradeoffs". What if the person doesn't have any good options? Maybe an AI data center job is the highest paying in a given area but still doesn't pay well AND you have to wrestle with the moral implications of working at a data center. The person could technically get a different job but how long should they have to look to find something better? Sometimes people end up in jobs that exploit their labor because it is a better decision than not eating. But is that really a meaningful choice? Or one that is coerced?
What counts as a "good option" is a highly subjective assessment made by an individual. The rest of what you say is just walking through hypothetical aspects of a trade-off.
These details aren't exactly relevant to the complaints Hao alleges though. She argued that AI companies outsourcing RLHF work to Kenya and English speaking developing countries to ensure models don't return NSFW content is exploitative or degrading for foreign workers. Her evidence consists of a few anecdotes of people having fairly pedestrian complaints. The reality is that these opportunities are great material opportunities that wouldn't otherwise exist. It's literally a choice between a job (material benefit) or no job (no material benefit). This can be evaluated fairly objectively since in almost all cases individuals in otherwise precarious conditions are always better off with a wage-earning job.
Hey,
Yes, having $1 is better than 0 dollars. But these companies are doing everything they can to extract as much profit from people in the Global South who have very limited options. They are choosing to pay them so little that they have to work as much as they possibly can because they know these workers have no other choice, they know they can get away with it. You do realize that is not a good thing, they are taking advantage of these workers and there’s a reason they have to outsource to countries that lack labour protections. this is an extremely harmful precedent to have set, like do you think the way the fast fashion industry for instance runs is good?
Also about your point about the water usage, did you not remember the book? Like literally as it starts coming out about the environmental impacts of data centers, like water usage, that’s when companies like openai started not releasing that information at all (funny because open is in their name … almost as if their actions consistently contradict what they set out to do.. hmm). And openai very much seems to be acting LIKE a colonial empire because (state sponsored to an extent), literally going to foreign countries where they know they can freely exploit the labour there, bulldoze the environment to build their data centres, in multiple examples taking water from areas that are already in a drought. taking these resources to enrich their stakeholders only (almost reminds u of colonial nations in past where money and power become concentrated in select industries/companies). Your opinion about Hao not adding anything more insightful about Altman and it being a rehash of her previous stuff is a valid subjective opinion, the only reason i’m commenting is because of the other points i wanted to address where there seems to be a disconnect where you’re not seeing what the cost of the convenience in your life is costing.
I would also like some clarification on why you think Hao would be against American companies in particular?
meha wrote: "Yes, having $1 is better than 0 dollars."
If you accept this as true, then you are accepting my criticism of Hao's framing of RHLF work outsourced to Kenya and other English-speaking developing nations.
meha wrote: "But these companies are doing everything they can to extract as much profit from people in the Global South who have very limited options."
There are assumption in this claim premised on economic illiteracy. There is any way to "extract" profit from an impoverished place. One could steal resources or force a population into slavery or perpetrate some other form of deceit (e.g selling a faulty product or service) on a vulnerable population, but hiring people for a market-going wage is not "extraction" nor exploitation.
meha wrote: "They are choosing to pay them so little that they have to work as much as they possibly can because they know these workers have no other choice, they know they can get away with it. You do realize that is not a good thing, they are taking advantage of these workers and there’s a reason they have to outsource to countries that lack labour protections."
This contradicts your prior claim that $1 is better than $0. American AI do not have to create jobs in developing countries. They can pay a higher wage in America or other developed anglophone countries if that was the only deal available to them. This would be less efficient. Efficiency is good for the economy as a whole. The developing world benefit with jobs that wouldn't otherwise exist. The American AI firms get RHLF work for an inexpensive wage.
If those developing countries managed to actually develop themselves then their population would be able to command higher wages. Even then there is always going to be some level of geographical arbitrage based on various material variables (i.e. different countries have different amount of differently aged people, different laws, different climates, etc).
meha wrote: "Also about your point about the water usage, did you not remember the book?"
I previously informed you that the water use claims in Hao's book are incorrect. You can check out an analysis by Andy Masley (titled "Empire of AI is wildly misleading about AI water use") to see those figures.
meha wrote: "openai very much seems to be acting LIKE a colonial empire because (state sponsored to an extent), literally going to foreign countries where they know they can freely exploit the labour there, bulldoze the environment to build their data centres, in multiple examples taking water from areas that are already in a drought." taking these resources to enrich their stakeholders only (almost reminds u of colonial nations in past where money and power become concentrated in select industries/companies).
In what way do you believe OpenAI is "state sponsored?" Do you think this is because they receive local tax abatement or because the federal government is a client? Typically, these are not things that would be understood as state sponsorship and are not consistent with historical forms of colonialisms like the Dutch or British East India Trading Company or even charter ventures in the new world.
You are characterizing any capital expenditure by any corporation into the economy of another country a form of colonialism.... This would mean that the entire world economy is colonialism and thus Hao's claim is so banal as to be meaningless...
meha wrote: "I would also like some clarification on why you think Hao would be against American companies in particular?"
I don't plan to spell out why this is the case because my claims would be speculation and until I can strongly substantiate these suspicions I wouldn't make a public claim.
My rhetorical question, which you are echoing presumably in earnest, is meant to get you to reflect: Cui bono?
tl;dr this is a terrible review from someone determined to malign this book
What is my motive for being "determined to malign this book" other than the book being very bad?
It appears to me that you characterize Hao’s conclusions as presuppositions. What you refer to as her slant, I would call her results.
Bob wrote: "It appears to me that you characterize Hao’s conclusions as presuppositions. What you refer to as her slant, I would call her results."
There is an easy test for whether your claim is true: Does she have a methodology to analyze the "data" she collects, where that data could falsify or fail to falsify her hypothesis?
Obviously, she does nothing of the kind, and she's just wildly inserting her own opinion (which is also premised on economically illiterate ideas) into what should just be straightforward journalism (as in "data" collection and curation only).
She also reveals a hostility toward the American generative AI industry as a whole. Her criticism extends to ALL the firms not just OpenAI.
It is all very ham-fisted. She could easily have done a more understated version of the same work.
back to top
message 1:
by
Yousuf
(new)
Jun 04, 2025 02:54PM
This reads like it was written by ChatGPT.
reply
|
flag
I frequently use AI tools to provide editorial support on my writing (as everyone should), but the language here is my original thought. What exactly informs your claim here? I think there are political sentiments in my review that are usually pretty hard to get any AI tool to spit out.
cracking up at how chatGPT generated this reads. the irony delights me. talk about disappointingly shallow! 😂😂😂😂
Emily wrote: "cracking up at how chatGPT generated this reads. the irony delights me. talk about disappointingly shallow! 😂😂😂😂"Which aspect of this do you believe is chatGPT generated or reads like genAI writing?
What is shallow about providing criticism of a book that is a sloppy assembly of previously written journalistic pieces that are whining that the big baddies in tech are the new "colonialists" and are building an evil empire?
Do you honestly believe that the developing world or "Global South" as Hao likes to say really has contributed anything of significance to the advancements in AI in recent years? If you do, I'd like you to invest in my new B2B SaaS startup....
🙃 discounting the environmental and labor contributions of the Global South… i know what you are
Lanae wrote: "🙃 discounting the environmental and labor contributions of the Global South… i know what you are"What am I then? Generally, to be persuasive, one should be clear about what they mean. Insinuation is ultimately ineffective as a rhetorical strategy.
What did the "Global South" actually contribute here? Basic data annotation and RLHF is not eminently skilled work here. In many ways, it is a generous largesse these populations get to enjoy as they wouldn't otherwise be getting the investment. Hao, like you, fails to consider the alternative.
Thank you for the review Stetson, you've made a compelling case for this book is just not good.As for the detractors on this thread, I don't think you guys are engaging with the points made in the review. All your critiques are "haha he used ChatGPT", "i know what you are." This adds nothing to the discussion. Please read the book before trying to be so dismissive.
Umar wrote: "Thank you for the review Stetson, you've made a compelling case for this book is just not good."Thanks, Umar! Glad you appreciated the review.
Simon wrote: "Oh dear, an AI review ie fake."There's nothing fake about the review. This is my opinion of Hao's book, which is mostly a loose collection of her reporting over the last few years.
The book is more of a polemic than actual journalism on OpenAI or genAI broadly. Additionally, the angle of the polemic is childish. The AI revolution has little to nothing to do with the "Global South" and is certainly not any form of "colonialism."
Unlike the hyenas in this thread, I’ve actually read the book. This is spot on. Very disappointing.
PCR wrote: "Unlike the hyenas in this thread, I’ve actually read the book. This is spot on. Very disappointing."Thanks, PCR!
Thank you for this review. I simply could not agree more with the majority of the points you made. My biggest frustrations were: 1) Hao’s analysis & criticisms of the impacts of AI are not unique to AI in the slightest, they’re simply a critique of capitalism/ how the world (yes, unfortunately) operates...
2) Hao fails to adequately and impartially address the potential long-term upside of AI and whole (very persuasive, in my view) case for why there is so much investment going into it. It is truly unlike any other technology to have ever existed.. her tone around the technology comes off as transparently biased. Just really disappointing for someone with such an insider’s view to be lacking in their depth of analysis and objective insight.
Thanks, Tommy! Yes, even if she wanted in her heart of hearts to eviscerate Sam, openAI, and AGI the technology, it would have had a better chance if she just simply did incredible objective reporting. She lost access because they sniffed out her agenda early and thus she didn't have much (unique) to add despite her experience in the AI beat and her chance at persuading people is low because her POV is overt and imo absurd.
Pretty much agree but I scored it a 4 because of the muckraking and abundant detail. https://www.goodreads.com/review/show...
Hi John, reasonable take! I didn't think Hao should get credit for double dipping. She did contribute real reporting on OpenAI but that reporting was available prior to the book. She should get credit for those things. The score of my review pertains just to the book's failure to deliver on its own terms and the unnecessary and sophomoric polemicism. Hao appears plenty capable as a reporter.
I work in AI and I actually share a lot of her baseline concerns, I think Sam Altman is one of the craziest and most dangerous people running around right now. But this book really got on my nerves with a lot of the obsessive hand wringing over like, equity and marginalized communities and colonialism and just sort of the whole leftist lens it was filtered through. I think when you analyze anything this broad and impactful through a heavily partisan lens with, like you said, an obvious disdain for the technology itself as opposed to maybe the way it's being handled, you're not giving an honest or complete view.
All great points, C. I'm less moved by the critique of Altman, though there are certainly reasons to be concerned about his character based on the reporting. If the book was just the reporting on OpenAI internal dynamics and Altman I would improve my rating of it.
Thanks, good review. Does "academic post Colonialism" always mean anti-capitalism?
No, she argues that AI development is "digital colonialism" exploiting and pillaging the "global South." I wouldn't say there is much in the way of a critique of capitalism other than within this post-colonial framing.
A “patchwork of grievances” is the perfect description and tracks with my experience. I had a growing sense that Hao has a huge, unnamed personal chip on her shoulder which motivated much of her approach. It’s disappointing because I believe she touches on many important issues which deserve a fair hearing, but I can’t trust her impartiality.
Thanks for such an articulate and extensive review.
Ok wait, but what’s the justification for the scaling, exploitation and waste of resources? I fail to read how we don’t care about these things, what’s the great gain from AI which grants the sacrifices being made? I haven’t read the book yet but if you have a suggestion I’d be happy to read more books to understand you better.
Andrea Maria wrote: "Ok wait, but what’s the justification for the scaling, exploitation and waste of resources? I fail to read how we don’t care about these things, what’s the great gain from AI which grants the sacri..."Hi Andrea, the justification for scaling and AI capex generally is that there is a strong expectation that the technology will lead to increased productivity across the economy. The extreme version of this vision is a post-scarcity future where human labor is not an essential input in the economy. This vision is unlikely to come to fruition but increase productivity will make everyone wealthier and healthier.
The tradeoff between AI capex and energy usage is not particularly concerning given that energy usage is going to increase if the economy is going to grow anyway. Additionally, it is possible, if not likely, that advances in AI technology may increase the efficiency of energy system or transition energy systems away from reliance of carbon sources. Much of this is already within reach given advances in solar, battery tech, and untapped nuclear sources. It is largely a failure of political will that keeps a country like the U.S. from running primarily on nuclear energy and battery tech.
There are many foundational works on general AI technology but most of them are not in book formats. They're in research articles and white papers. Here are two useful examples:
- "AI 2027" - white papers on the intelligence explosion
-"Attention is All You Need" - Transformer architecture paper
-AlphaFold Paper - "Highly accurate protein structure prediction with AlphaFold"
If you're looking for books that are for general readers but are more positive about AI then you can read:
-Co-Intelligence by Ethan Mollick (mostly about specific use cases for individuals)
-On the Edge by Nate Silver (not specifically about AGI but covers major tech players and industry)
-Superintelligence by Nick Bostrom (a philosophy book on the alignment problem)
There are also journalistic/commentary sources that report on the potential and current impact of AI:
"Understanding AI" by Timothy B. Lee on Substack
"The Cognitive Revolution" podcast from Erik Torenberg
so your main criticism of this book is that the author doesn't spend enough time praising AI? got it.
Rivara wrote: "so your main criticism of this book is that the author doesn't spend enough time praising AI? got it."No, my primary criticism is that the central framing of the book is ridiculous (i.e. AI capex in data centers and labor outsourcing in the developing world =/= imperialism nor "data imperialism"). The lesser criticism of the singular pursuit of AGI rather than other types of AI is not trenchant. One, there is plenty of other major AI projects that are not focused on AGI (e.g. AlphaFold). Two, genAI has already seeing plenty of success even if the rate of progress is plateauing.
Additionally, the reporting itself leaves much to be desired. Hao doesn't focus on the actual newsworthy events at OpenAI in sufficient detail and instead wastes time trying to provoke the audience's ire concerning Sam Altman. This includes taking obviously unnecessary cheap shots like retreading Altman's troubled sister's accusations. The original reporting that Hao actually did is also just recycling stuff she's already published, whereas a book like this would be an opportunity to debut some actual news.
As I said in the review itself, I walked away from the book without really learning anything new about Altman's candor and its relationship with his brief ousting. Altman is portrayed as Machiavellian and somewhat two-faced or just a tell-people-what-they-want-hear kind of guy, but none of his individual actions looked especially shocking in the scope of what's par-for-course among tech CEO in an emerging sector.
I thought an additional weakness was the Hao provides very little conceptual and technical detail concerning the field of artificial intelligence. It is a capacious domain that is not particularly young. Many people are re-discovering it in the post-ChatGPT world, but there is a long intellectual history that is relevant to the very subjects she reports on. She provides the most 101 of 101 type coverage.
This is a deeply biased review. Let’s take one example. You claim she treats comparisons between AI companies and colonial empires as self-evident and avoids examinations. And yet she offers abundant detail about the harms done to AI contractors in the global south. Her accounts moreover align with similar investigations conducted by other AI critics in the field. Your criticism that her investigation of Sam Altman is too thin is risible considering he and OpenAI refused to speak with her for the book.
Your attacks on her for not discussing on the entire history of AI development or its many contributors is just weird. She’s not writing an 800pg history book about everything that ever happened in AI. She made reasonable choices about how much industry backstory to include or not.
Tbh you sound like your wrote your review before you read the book. Why you feel the need to be w white knight for the most valuable private corporation on the planet is beyond me.
Andrew wrote: "This is a deeply biased review. Let’s take one example. You claim she treats comparisons between AI companies and colonial empires as self-evident and avoids examinations. And yet she offers abundant detail about the harms done to AI contractors in the global south. Her accounts moreover align with similar investigations conducted by other AI critics in the field."Reviews are by definition biased; they're an expression of a personal opinion. What you mean to say is that the evidence for the argument on which my opinion sits is flawed in someway. However you believe it to be flawed is not well articulated. You simply assert (without evidence) that she has a bunch of evidence and that others (with the same worldview) agree with her...
This entirely misunderstand my critique as there is no world I could conceive in which there's any evidence for the idea that AI companies are functioning like the Dutch East India Trading Company. It is a motivated category error. AI companies don't have military forces, don't have the imprimatur of the U.S. government, and don't exercise any substantial coercive powers. Everything they do some other party has mutually agreed to.
Andrew wrote: "Your criticism that her investigation of Sam Altman is too thin is risible considering he and OpenAI refused to speak with her for the book.
Your attacks on her for not discussing on the entire history of AI development or its many contributors is just weird. She’s not writing an 800pg history book about everything that ever happened in AI. She made reasonable choices about how much industry backstory to include or not."
My views on how she treated sama and AI, the technology, generally are simply my opinion. I'm entitled to it as you are entitled to disagree. Plenty of people also see it the way I do.
Plus, If you think it's totally cool that Hao parades around Sam's mentally disturbed sister as if that's relevant to his stewardship of his company then you're probably a lot less impartial about these subject than I am.
Andrew wrote: ""Tbh you sound like your wrote your review before you read the book. Why you feel the need to be w white knight for the most valuable private corporation on the planet is beyond me."
By the way, not only did I read the book, but I also listened to many of Hao's interviews in which she also maintains her ridiculous ideological position, appearing like a sophomoric activist rather than an actual journalist.
Additionally, it is unclear why an American would be particularly opposed to OpenAI. It is a wildly successful company who will undeniably be of strategic and material value to the U.S. taxpayer. In fact, this reality may indeed entirely account for Hao's desperate and silly effort to politicize the company among Americans.
I'm sorry, but your review is just so narrow minded. I see that you expected a) an in-depth technical analysis of what OpenAI has worked on and b) an in-depth analysis of OpenAI as a company. Why not take a step back and have a look at what the author is trying to raise an alarm over: the recklessness of releasing AI to the general public without a proper structure or proper consultation with governments to understand and mitigate the impact over the entire population. She is trying to draw a parallel between how the company started and its initial ethos (do no harm) and what it ended up being, which is yolo-ing its way through, despite having the Facebook and the Internet as models of what could go wrong if people in charge release features without thinking about the consequences (e.g. spread of child pornography, copyright infringement, etc.). This moment in AI evolution looks similar to the evolution of gene studies. If you remember the Dolly experiment, this was considered highly unethical in the medical field and similar experiments were banned in multiple countries, or at least funding human cloning experiments was not allowed. While the medical field has the Hippocratic oath to protect human beings, the IT industry does the equivalent of releasing experimental drugs to the general population without any sort of oversight. If you would have been more open minded while reading the book, this is one of the aspects of the AI industry that the author tries to highlight.
Maybe you should go back to reading papers about neural networks if that is your interest and the ethics of AI is something you don't consider important. Don't rate a book a 1 star because it didn't meet your preconceived expectations. It seems like you read the book like a professor grading a student's paper on a topic they are teaching. Maybe try getting out of your comfort zone and consider another point of view on the topic while reading a book.
My primary complaint about Hao's book was that she tried to force the business decisions made by sama/OpenAI and the field of genAI in general into a nonsensical frame (i.e. that scaling dogma downstream of AGI worship leads to exploitation of the resources of the Global South). The idea that AI capex in data centers, pre-training on open Internet data, and outsourcing RLHF is a colonial project is laughable. Great power competition is much more relevant to understanding and commenting on AI development than third-rate Said/Fanon poasting (what Hao is doing here).My concerns about the book's superficial treatment of the technical aspects of genAI were a secondary critique. The same is true of the reporting on OpenAI itself. These lesser issues simply underscore how poor the overall work is. She re-heated her old reporting, added nothing actually new or useful, and threw in other random journalistic work she did.
I don't think my expectations for the work were that high. If she just included the sama OpenAI drama and some general stuff about the AI industry then it would be like a 2.5-3 star book, but she ruined it with an ideological thesis meant to get aging far left-wing luddites worked up.
If you are so favourable to chatgpt that you think everyone should use it, then you weren't prepared to read this book. You didn't even try to be objective.
Rohit wrote: "If you are so favourable to chatgpt that you think everyone should use it, then you weren't prepared to read this book. You didn't even try to be objective."Is this an argument? Perhaps genAI could help you formulate one.
As one of my prior comments states, book reviews are inherently subjective, meaning they are definitionally an expression of one's biases. If you want to attack the review, you have to show their is a specific weakness in the reasoning I've presented not take issue that my tastes are different than your tastes. Why should my tastes be the same as yours?
“Reviews are by definition biased”… that made me laugh. 😆 Shame on you for not subscribing to a particular viewpoint! I’m about 3/4 of the way through and I’m finding a lot of the same problems with the book that you mentioned in your review. It just seems more like a hit piece on Sam Altman and OpenAI, rather than any real balanced view of Sam and his company. So many parts detailing threats that data centers and supercomputers have brought of exploitation to third world countries’ natural resources and Indigenous populations are speculative - talking about misdeeds that Microsoft and Google have committed, yes - but rather than acknowledge it as a potential problem Open AI may also replicate, she paints it as a foregone conclusion. Sure, there’s obviously an argument to be made for what Google and Microsoft have done in Chile, but OpenAI? Not quite.And while I know it makes me sound like a horrible person to many to disregard sexual abuse allegations, I just can’t take Annie Altman’s allegations seriously. So we’re supposed to believe that their entire lives, Sam and Annie were incredibly close, she viewed him as her protector, etc.? Yet suddenly once OpenAI and ChatGPT become huge, Sam becomes a billionaire, she suddenly files a lawsuit saying he sexually abused her from the age of three?
If abuse did occur at that age, it would be extremely difficult to identify the perpetrator(s). Your memory at 3-5 years old, or whatever, isn’t really advanced enough to warrant the types of allegations she’s making, and furthermore, if trauma indeed did occur, people often fully believe something to be 100% factual that, in reality, never happened (this actually happened with me through the loss of a loved one - while I was there. I remembered things and insisted I’d seen them, the image was IN MY MIND - only to realize that these things I believed weren’t logically possible).
I don’t know how great her artistic skill is, but it doesn’t seem feasible for even the best artists to live in the Bay Area on income generated solely from their artwork (whatever type of artwork she was doing, it didn’t bother to mention it). So it seems very likely that she is being financially irresponsible if she’s living in the most expensive area of the U.S., investing her money into art, doing artwork for a living, yet claiming she never expected Sam or her family to provide for her. He buys her a house but then she’s mad that he stops payment on it once the sexual abuse allegations are printed for the entire world to see? 🤦🏼♀️ If it really did happen, that’s awful, but it just doesn’t add up for Sam to have been her main abuser, or have abused her at all. Ruining someone’s reputation with false or inaccurate claims is awful, too.
Megan wrote: "I’m finding a lot of the same problems with the book that you mentioned in your review. It just seems more like a hit piece on Sam Altman and OpenAI, rather than any real balanced view of Sam and his company."Agreed, this is an angle I didn't spend much time highlighting in my review because I can't speak to the specific motivation behind Hao's perspective on Sam Altman and OpenAI. However, it is clear there is animus there for some reason... I could speculate on a few, but it would be a bit irresponsible so I've demurred from that course.
Megan wrote: " So many parts detailing threats that data centers and supercomputers have brought of exploitation to third world countries’ natural resources and Indigenous populations are speculative - talking about misdeeds that Microsoft and Google have committed, yes - but rather than acknowledge it as a potential problem Open AI may also replicate, she paints it as a foregone conclusion. Sure, there’s obviously an argument to be made for what Google and Microsoft have done in Chile, but OpenAI? Not quite."
I see capital expenditure in the developing world or outsourcing of RLHF positions to English-speaking foreigners as almost an unalloyed positive for those countries and people. It is investment and a cash influx that otherwise would not be there at all. Instead of Latin America and Africa, the tech companies are pivoting to the Middle East instead because of the self-defeating political opposition from people on the ground. It reminds me of the time AOC worked so hard to keep Amazon from setting up HQ#2 in NY - complete self-own.
Megan wrote: " I just can’t take Annie Altman’s allegations seriously."
I think Hao in her heart of hearts absolutely knows that Altman's allegations are part of an effort to continue to extract money from her brother. It is evident from even the information that Hao presents while trying to use the story to impugn Sam that Annie has a lot of mental health struggles and has already received a great deal of support from Sam and the rest of the Altman family.
I don’t see much of anything about regulations or unregulated practices or governance at all. Those ideas were pretty large takeaways for me, which have some pretty heavy moral implications. What are your thoughts?
I'm not an AI doomer. I think the x risk is infinitesimally small.I'm not sure where in the book you see Hao dedicating much focus to governance or policy. There isn't even that much focus on the export controls on chips, which is usually the primary focus of AI-related regulatory discussions or philosophical musings about the alignment problem and intelligence explosion. She of course expresses a normative desire for pre-training input (copyright), environmental and financial regulations, which are all things if generally oppose. I don't really see any socially pressing issues there. It's also strange she's that concerned about it since she allegedly doesn't think the technology is going to go anywhere (basically the Marcus critique).
You'd have to be more specific about what you see as particularly urgent and morally grave.
You probably wrote this with chatgpt
Bobbydigital wrote: "You probably wrote this with chatgpt"You can plug this into AI text checker if you like, Grammarly's tool says my review is only 4% AI generated.
I often use genAI to edit, usually in a paragraph by paragraph fashion, but clearly this review is overwhelmingly my wording and my perspective.
Clearly my review has touched a nerve with folk. If it was just AI slop that's triggering them, isn't that a reflection on them?
I had the same thought as Bobbydigital - the use of AI in your writing makes it hard to read. “I’ve demurred from that” and “infinitesimally small” are not real phrases people use. I’ll add an extra star to my review because at least the author wrote it.
You should read more of my writing. I often use the word "demure," and I'm a geneticist where the infinitesimal model is the basis for complex traits. Clearly, the readability of my review is decent, and it's one of the most popular reviews of this book here at this site.
hi stetson, you seem unable to actually engage with any of the criticism of the AI industry she presented, or its real impact on human lives.
Also it’s a bit misleading to say Hao is hostile to AI as a technology which is just categorically false. She critiques the current nature of the industry and those who are leading it, not the actual technology. She often highlights the actual goal of researchers (at openai!) and their seemingly pure intentions and discusses real world examples of sustainable AI development and usage (like the group in New Zealand)
Also she did not spend a lot of time on Ann Altman, that was literally such a small part of the book. not sure why you chose to mention that but not any of the much larger discussions in the book like the environmental toll, loss of transparency that contradicts the main ethos of openAI, etc.
You disregard the exploitative nature of the AI industry’s presence in the Global South by claiming these low wages are better than no wages. Which is a pretty childish point of view and shows you didn’t really engage with this book in any meaningful way.
These companies are normalizing and setting these standards, meanwhile their executives are getting paid millions. They have billions in funding and are being bankrolled by the biggest companies, but it’s okay for workers in developing nations to be paid scraps and wait for pings at midnight because that’s the only way they can afford to eat? Not to mention their actual job is being exposed to the most vile stuff the internet has created.
Which leads me to your point saying Hao just assumes the industry to act as an empire rather than analyzing it … which is a complete misread imo. Like the whole thing is her diving into why the industry and openAI specifically act as a colonial empire by analyzing their actions and infrastructure. She doesn’t make any claims without evidence.
Also your wording of how she expects us to be “enraged” by the responses from execs seems like a case of projection. that is not at all how it came across to me as a reader, it feels like this is a narrative that you’re putting onto her. Hao’s main discussions are the systemic failings of the company that thus make it run as a profit seeking empire fuelled by extraction of labour and ideas. None of this is assumed, she deeply discusses and interviews people who are victims and also those responsible for it.
It’s pretty clear going into this book you had already formed your opinions about the whole matter and the author itself. So it’s interesting that you don’t see the hypocrisy in you ending the review as if this book is some self affirming exercise when there exists no nuance in your review or opinion.
Meha,I explicitly address the core claim made by the sloppy book. Hao argued genAI companies and capex are "data colonialism" that exploits the global south. The evidence she tenders is very thin: 1) remote workers in developing countries doing challenging work and 2) data center water use. The former is stupid as those workers are voluntarily participants and are literally better off for the opportunity to earn money. The latter is not factually supportable. There is a beautiful debunking of the water use claims (Google it) that specifically shows Hao is either lying or is wildly ignorant about these facts.
You're right Hao is suspiciously just hostile to America's premier AI firms and strategy. Hmmm... Makes ya wonder. Why could that be!?
You have the same incorrect economic model. Working for a wage is almost always materially and socially superior to unemployment. If the job is so bad, they can voluntarily not do it!! Being a mature economic actor means navigating tradeoffs.
Do you know what colonialism is? It is when a state sponsored entity goes (with guns) and conquers a foreign land and appropriates the resources as they see fit. This is not even metaphorically similar to capex in foreign countries and voluntary market exchange. If anything it's the precise opposite of colonialism.
I've listened to many interviews of Hao concerning her book and followed where those interviews happen, which itself is a signal. It is clear how she wants people to react. It's also plainly there as subtext or actual text in her book.
I had no idea what to expect going in other than I should get some details on Altman's leadership of openAI and get detailed reporting on why he was briefly deposed. Hao didn't add anything new to this story that wasn't already available. Those events still seem largely inexplicable and an amazing self-own by those involved. I also credit her for whatever original journalism she actually did on altman and openAI but there wasn't much original material in the book.
There isn't any hypocrisy embedded in having a strong opinion, especially one supported heavily by empirical reality and logical analysis. I'm not criticizing Hao for hypocrisy. I'm criticizing her for trying to pass off dorm room polemic as straight journalism. I'm also criticizing her for not doing her homework. It was lazy and sloppy given how much she gets wrong or facts she fails to produce as part of her overall thesis.
"Being a mature economic actor means navigating tradeoffs". What if the person doesn't have any good options? Maybe an AI data center job is the highest paying in a given area but still doesn't pay well AND you have to wrestle with the moral implications of working at a data center. The person could technically get a different job but how long should they have to look to find something better? Sometimes people end up in jobs that exploit their labor because it is a better decision than not eating. But is that really a meaningful choice? Or one that is coerced?
What counts as a "good option" is a highly subjective assessment made by an individual. The rest of what you say is just walking through hypothetical aspects of a trade-off. These details aren't exactly relevant to the complaints Hao alleges though. She argued that AI companies outsourcing RLHF work to Kenya and English speaking developing countries to ensure models don't return NSFW content is exploitative or degrading for foreign workers. Her evidence consists of a few anecdotes of people having fairly pedestrian complaints. The reality is that these opportunities are great material opportunities that wouldn't otherwise exist. It's literally a choice between a job (material benefit) or no job (no material benefit). This can be evaluated fairly objectively since in almost all cases individuals in otherwise precarious conditions are always better off with a wage-earning job.
Hey, Yes, having $1 is better than 0 dollars. But these companies are doing everything they can to extract as much profit from people in the Global South who have very limited options. They are choosing to pay them so little that they have to work as much as they possibly can because they know these workers have no other choice, they know they can get away with it. You do realize that is not a good thing, they are taking advantage of these workers and there’s a reason they have to outsource to countries that lack labour protections. this is an extremely harmful precedent to have set, like do you think the way the fast fashion industry for instance runs is good?
Also about your point about the water usage, did you not remember the book? Like literally as it starts coming out about the environmental impacts of data centers, like water usage, that’s when companies like openai started not releasing that information at all (funny because open is in their name … almost as if their actions consistently contradict what they set out to do.. hmm). And openai very much seems to be acting LIKE a colonial empire because (state sponsored to an extent), literally going to foreign countries where they know they can freely exploit the labour there, bulldoze the environment to build their data centres, in multiple examples taking water from areas that are already in a drought. taking these resources to enrich their stakeholders only (almost reminds u of colonial nations in past where money and power become concentrated in select industries/companies). Your opinion about Hao not adding anything more insightful about Altman and it being a rehash of her previous stuff is a valid subjective opinion, the only reason i’m commenting is because of the other points i wanted to address where there seems to be a disconnect where you’re not seeing what the cost of the convenience in your life is costing.
I would also like some clarification on why you think Hao would be against American companies in particular?
meha wrote: "Yes, having $1 is better than 0 dollars."If you accept this as true, then you are accepting my criticism of Hao's framing of RHLF work outsourced to Kenya and other English-speaking developing nations.
meha wrote: "But these companies are doing everything they can to extract as much profit from people in the Global South who have very limited options."
There are assumption in this claim premised on economic illiteracy. There is any way to "extract" profit from an impoverished place. One could steal resources or force a population into slavery or perpetrate some other form of deceit (e.g selling a faulty product or service) on a vulnerable population, but hiring people for a market-going wage is not "extraction" nor exploitation.
meha wrote: "They are choosing to pay them so little that they have to work as much as they possibly can because they know these workers have no other choice, they know they can get away with it. You do realize that is not a good thing, they are taking advantage of these workers and there’s a reason they have to outsource to countries that lack labour protections."
This contradicts your prior claim that $1 is better than $0. American AI do not have to create jobs in developing countries. They can pay a higher wage in America or other developed anglophone countries if that was the only deal available to them. This would be less efficient. Efficiency is good for the economy as a whole. The developing world benefit with jobs that wouldn't otherwise exist. The American AI firms get RHLF work for an inexpensive wage.
If those developing countries managed to actually develop themselves then their population would be able to command higher wages. Even then there is always going to be some level of geographical arbitrage based on various material variables (i.e. different countries have different amount of differently aged people, different laws, different climates, etc).
meha wrote: "Also about your point about the water usage, did you not remember the book?"
I previously informed you that the water use claims in Hao's book are incorrect. You can check out an analysis by Andy Masley (titled "Empire of AI is wildly misleading about AI water use") to see those figures.
meha wrote: "openai very much seems to be acting LIKE a colonial empire because (state sponsored to an extent), literally going to foreign countries where they know they can freely exploit the labour there, bulldoze the environment to build their data centres, in multiple examples taking water from areas that are already in a drought." taking these resources to enrich their stakeholders only (almost reminds u of colonial nations in past where money and power become concentrated in select industries/companies).
In what way do you believe OpenAI is "state sponsored?" Do you think this is because they receive local tax abatement or because the federal government is a client? Typically, these are not things that would be understood as state sponsorship and are not consistent with historical forms of colonialisms like the Dutch or British East India Trading Company or even charter ventures in the new world.
You are characterizing any capital expenditure by any corporation into the economy of another country a form of colonialism.... This would mean that the entire world economy is colonialism and thus Hao's claim is so banal as to be meaningless...
meha wrote: "I would also like some clarification on why you think Hao would be against American companies in particular?"
I don't plan to spell out why this is the case because my claims would be speculation and until I can strongly substantiate these suspicions I wouldn't make a public claim.
My rhetorical question, which you are echoing presumably in earnest, is meant to get you to reflect: Cui bono?
tl;dr this is a terrible review from someone determined to malign this book
What is my motive for being "determined to malign this book" other than the book being very bad?
It appears to me that you characterize Hao’s conclusions as presuppositions. What you refer to as her slant, I would call her results.
Bob wrote: "It appears to me that you characterize Hao’s conclusions as presuppositions. What you refer to as her slant, I would call her results."There is an easy test for whether your claim is true: Does she have a methodology to analyze the "data" she collects, where that data could falsify or fail to falsify her hypothesis?
Obviously, she does nothing of the kind, and she's just wildly inserting her own opinion (which is also premised on economically illiterate ideas) into what should just be straightforward journalism (as in "data" collection and curation only).
She also reveals a hostility toward the American generative AI industry as a whole. Her criticism extends to ALL the firms not just OpenAI.
It is all very ham-fisted. She could easily have done a more understated version of the same work.
