Large Language Models Quotes

“The lack of transparency regarding training data sources and the methods used can be problematic. For example, algorithmic filtering of training data can skew representations in subtle ways. Attempts to remove overt toxicity by keyword filtering can disproportionately exclude positive portrayals of marginalized groups. Responsible data curation requires first acknowledging and then addressing these complex tradeoffs through input from impacted communities.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Many presume that integrating more advanced automation will directly translate into productivity gains. But research reveals that lower-performing algorithms often elicit greater human effort and diligence. When automation makes obvious mistakes, people stay attentive to compensate. Yet flawless performance prompts blind reliance, causing costly disengagement. Workers overly dependent on accurate automation sleepwalk through responsibilities rather than apply their own judgment.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“It is critical to recognize the limitations of LLMs from a consumer perspective. LLMs only possess statistical knowledge about word patterns, not true comprehension of ideas, facts, or emotions. Their fluency can create an illusion of human-like understanding, but rigorous testing reveals brittleness. Just because a LLM can generate coherent text about medicine or law doesn’t mean it grasps those professional domains. It does not. Responsible evaluation is essential to avoid overestimating capabilities.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Every piece of data ingested by a model plays a role in determining its behavior. The fairness, transparency, and representativeness of the data reflect directly in the LLMs' outputs. Ignoring ethical considerations in data sourcing can inadvertently perpetuate harmful stereotypes, misinformation, or gaps in knowledge. It can also infringe on the rights of data creators.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“LLMs represent some of the most promising yet ethically fraught technologies ever conceived. Their development plots a razor’s edge between utopian and dystopian potentials depending on our choices moving forward.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Automation promises to execute certain tasks with superhuman speed and precision. But its brittle limitations reveal themselves when the unexpected arises. Studies consistently show that, as overseers, humans make for fickle partners to algorithms. Charged with monitoring for rare failures, boredom and passivity render human supervision unreliable.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Open source philosophies once promised to democratize access to cutting-edge technologies radically. Yet for AI, the eventual outcome of the high-stakes battle between open and closed systems remains highly uncertain.
Powerful incentives pull major corporate powers to co-opt open source efforts for greater profit and control, however subtly such dynamics might unfold. Yet independent open communities intrinsically chafe against restrictions and centralized control over capacity to innovate. Both sides are digging in for a long fight.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“One can, to be sure, program a digital machine in such a way as to be able to carry on a conversation with it, as if with an intelligent partner. The machine will employ, as the need arises, the pronoun “I” and all its grammatical inflections. This, however, is a hoax! The machine will still be closer to a billion chattering parrots—howsoever brilliantly trained the parrots be—than to the simplest, most stupid man. It mimics the behavior of a man on the purely linguistic plane and nothing more.”
― Stanisław Lem, A Perfect Vacuum

“Imagine an alternate universe in which people don’t have words for different forms of transportation—only the collective noun “vehicle.” They use that word to refer to cars, buses, bikes, spacecraft, and all other ways of getting from place A to place B. Conversations in this world are confusing. There are furious debates about whether or not vehicles are environmentally friendly, even though no one realizes that one side of the debate is talking about bikes and the other side is talking about trucks. There is a breakthrough in rocketry, but the media focuses on how vehicles have gotten faster—so people call their car dealer (oops, vehicle dealer) to ask when faster models will be available. Meanwhile, fraudsters have capitalized on the fact that consumers don’t know what to believe when it comes to vehicle technology, so scams are rampant in the vehicle sector.

Now replace the word “vehicle” with “artificial intelligence,” and we have a pretty good description of the world we live in.

Artificial intelligence, AI for short, is an umbrella term for a set of loosely related technologies. ChatGPT has little in common with, say, software that banks use to evaluate loan applicants. Both are referred to as AI, but in all the ways that matter—how they work, what they’re used for and by whom, and how they fail—they couldn’t be more different.”
― Arvind Narayanan, AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference

“[All] modern chatbots are actually trained simply to predict the next word in a sequence of words. They generate text by repeatedly producing one word at a time. For technical reasons, they generate a “token” at a time, tokens being chunks of words that are shorter than words but longer than individual letters. They string these tokens together to generate text.

When a chatbot begins to respond to you, it has no coherent picture of the overall response it’s about to produce. It instead performs an absurdly large number of calculations to determine what the first word in the response should be. After it has output—say, a hundred words—it decides what word would make the most sense given your prompt together with the first hundred words that it has generated so far.

This is, of course, a way of producing text that’s utterly unlike human speech. Even when we understand perfectly well how and why a chatbot works, it can remain mind-boggling that it works at all.

Again, we cannot stress enough how computationally expensive all this is. To generate a single token—part of a word—ChatGPT has to perform roughly a trillion arithmetic operations. If you asked it to generate a poem that ended up having about a thousand tokens (i.e., a few hundred words), it would have required about a quadrillion calculations—a million billion.”
― Arvind Narayanan, AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference

“Unleashing Reliable Insights from Generative AI by Disentangling Language Fluency and Knowledge Acquisition

Generative AI carries immense potential but also comes with significant risks. One of these risks of Generative AI lies in its limited ability to identify misinformation and inaccuracies within the contextual framework.

This deficiency can lead to mistakenly associating correlation with causation, reliance on incomplete or inaccurate data, and a lack of awareness regarding sensitive dependencies between information sets.

With society’s increasing fascination with and dependence on Generative AI, there is a concern that the unintended consequence that it will have an unhealthy influence on shaping societal views on politics, culture, and science.

Humans acquire language and communication skills from a diverse range of sources, including raw, unfiltered, and unstructured content. However, when it comes to knowledge acquisition, humans typically rely on transparent, trusted, and structured sources.

In contrast, large language models (LLMs) such as ChatGPT draw from an array of opaque, unattested sources of raw, unfiltered, and unstructured content for language and communication training. LLMs treat this information as the absolute source of truth used in their responses.

While this approach has demonstrated effectiveness in generating natural language, it also introduces inconsistencies and deficiencies in response integrity.

While Generative AI can provide information it does not inherently yield knowledge.

To unlock the true value of generative AI, it is crucial to disaggregate the process of language fluency training from the acquisition of knowledge used in responses. This disaggregation enables LLMs to not only generate coherent and fluent language but also deliver accurate and reliable information.

However, in a culture that obsesses over information from self-proclaimed influencers and prioritizes virality over transparency and accuracy, distinguishing reliable information from misinformation and knowledge from ignorance has become increasingly challenging. This presents a significant obstacle for AI algorithms striving to provide accurate and trustworthy responses.

Generative AI shows great promise, but addressing the issue of ensuring information integrity is crucial for ensuring accurate and reliable responses. By disaggregating language fluency training from knowledge acquisition, large language models can offer valuable insights.

However, overcoming the prevailing challenges of identifying reliable information and distinguishing knowledge from ignorance remains a critical endeavour for advancing AI algorithms. It is essential to acknowledge that resolving this is an immediate challenge that needs open dialogue that includes a broad set of disciplines, not just technologists

Technology alone cannot provide a complete solution.”
― Tom Golway

“As LLMs burgeon and permeate diverse sectors, the mandate for transparency, facilitated by all-encompassing documentation, becomes even more pressing.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Benchmarks should aid rather than substitute multifaceted, human-centric assessment focused on benefiting diverse populations. We must see behind the leaderboard, upholding wisdom over metrics. Tools like model cards and datasheets support responsible benchmark practices. But comprehensive governance requires collaboration at all levels of society.”
― I. Almeida, Introduction to Large Language Models for Business Leaders: Responsible AI Strategy Beyond Fear and Hype

“Warning to the Spellbound
(Sonnet from the future)

In our times we wrote our own literature,
In our times we wrote our own music.
In our times we wrote our own code,
In our times we wrote our own poetry.

Ours was the last human generation,
where humans shaped their own society.
The day you traded comfort for originality,
you forfeited the right to life and liberty.

Today you are nothing, you mean thing,
you are no more significant than woodworm.
You are just puppets to large gibberish models,
backboneless victims of algorithm addiction.

If you can still hear my voice, AI is still adolescent,
Once in control, it'll erase all records of humanness.
We can't yet treat human bias, 'n here comes AI bias,
Abandon all non-vital tech, return to simpler ways.”
― Abhijit Naskar, Brit Actually: Nursery Rhymes of Reparations

“AI is the white colonizer of the modern world, headed to destroy everything that is sweet, original and meaningful about human life. Unless you clip its wings while there is time, like the British empire, AI empire will bring back the dark ages, not light.”
― Abhijit Naskar, Brit Actually: Nursery Rhymes of Reparations

“Welcome to the age of AI, where algorithms grow bigger, and minds get smaller.”
― Abhijit Naskar, The Humanitarian Dictator

“Once trained, the LLM is ready for inference. Now given some sequence of, say, 100 words, it predicts the most likely 101st word. (Note that the LLM doesn’t know or care about the meaning of those 100 words: To the LLM, they are just a sequence of text.) The predicted word is appended to the input, forming 101 input words, and the LLM then predicts the 102nd word. And so it goes, until the LLM outputs an end-of-text token, stopping the inference. That’s it!

An LLM is an example of generative AI. It has learned an extremely complex, ultra-high-dimensional probability distribution over words, and it is capable of sampling from this distribution, conditioned on the input sequence of words. There are other types of generative AI, but the basic idea behind them is the same: They learn the probability distribution over data and then sample from the distribution, either randomly or conditioned on some input, and produce an output that looks like the training data.”
― Anil Ananthaswamy, Why Machines Learn: The Elegant Math Behind Modern AI