Victoria Fox's Blog, page 112

Artificial Intelligence: "Deepfakes are coming whether we want them or not"

Robert Habeck is sitting in the talk show chair, his hair is a little disheveled as usual, but something amazing is coming out of his mouth. “I don’t give a shit about your potatoes,” he says, obviously directed at farmers who are demanding more subsidies. “We’ll just import the things via Amazon Prime.” This video is objectively very funny. But of course it’s not real.

This has to be, you think as you scroll through Instagram , one of those deepfakes that everyone from the Pope to the UN is warning about. Videos created using artificial intelligence in which politicians appear to say things they never said. This could be used to manipulate opinions, change moods, and bury the truth. Such fears have been around for a long time, but they are becoming increasingly concrete. Now, when important elections are coming up in the USA and Germany, when AI companies are releasing ever more impressive tools at ever shorter intervals, when the world hyperventilated for a week over a manipulated photo of Princess Kate.

SOURCE

AI music: This is what AI songs sound like

In music, it is like in many other areas of life: people are not sure whether they should approach artificial intelligence with fear or with hope. This leads to on-the-one-hand-on-the-other-hand texts like the open letter that more than 200 artists published on Wednesday.

Stevie Wonder, Billie Eilish, Katy Perry, Jon Bon Jovi and other stars warn that AI – or more precisely, what companies do with it – is an “attack on human creativity”. On the other hand, they write, AI also has “enormous potential to promote human creativity”.

SOURCE

Cyber security: Authority warns of increased phishing attacks on parties

According to the Federal Office for Information Security (BSI), political actors such as parties are currently particularly targeted by cyber attacks. “Particularly in light of the upcoming European elections, an increased number of attacks can currently be expected,” a BSI spokeswoman told the dpa news agency.

According to the BSI and the Federal Office for the Protection of the Constitution, there is currently information on so-called phishing attacks against German parties, among others, which originate from a state cyber actor. However, this actor has not yet been named. In phishing, hackers try to pose as trustworthy entities using fake websites in order to steal data such as passwords.

Authority warns against publication of manipulated documents

Currently, so-called hack-and-leak attacks are the most common, said the BSI spokeswoman. In these attacks, non-public data or documents are stolen and then published, sometimes in a falsified form. The cyber threat level for the targets of such attacks is currently high. The BSI did not provide any further details.

Back in February, the Federal Office for the Protection of the Constitution and the BSI had already warned the parties about the danger of hack-and-leak operations in the 2024 election year. In the letter dated February 21, the security authorities warned against underestimating such cyber attacks. “By deliberately publishing and falsifying information, state-controlled cyber and influence actors in particular (…) can carry out disinformation campaigns.” These are aimed at “manipulating public opinion, spreading false ideas or undermining trust in democratic processes.”

BSI calls for more political attention to cybersecurity

In particular, the security authorities warned politicians about attempted attacks “in which perpetrators registered domains that resembled the official webmail domains,” a common tactic in phishing attacks. Officials were sent emails in the name of their organization’s IT support, in which they were asked to log in to the manipulated website using their login details.

Overall, Germany is poorly prepared for cyber attacks, warned the BSI a few days ago. Authority President Claudia Plattner told the Tagesspiegel that there is currently no joint situation report from the federal and state governments and no structures that would enable effective coordination in the event of a crisis. “But we absolutely need this,” said Plattner. She appealed to politicians to prioritize the issue of cyber security “before something major happens and, for example, a series of ATMs fail.” One does not always have to “learn from one’s mistakes.”

According to the Federal Office for Information Security (BSI), political actors such as parties are currently particularly targeted by cyber attacks. “Particularly in light of the upcoming European elections, an increased number of attacks can currently be expected,” a BSI spokeswoman told the dpa news agency.

According to the BSI and the Federal Office for the Protection of the Constitution, there is currently information on so-called phishing attacks against German parties, among others, which originate from a state cyber actor. However, this actor has not yet been named. In phishing, hackers try to pose as trustworthy entities using fake websites in order to steal data such as passwords.

SOURCE

Google might make users pay for AI features in search results

Enlarge / You think this cute little search robot is going to work for free?Getty Images

Google might start charging for access to search results that use generative artificial intelligence tools. That’s according to a new Financial Times report citing “three people with knowledge of [Google’s] plans.”

Charging for any part of the search engine at the core of its business would be a first for Google, which has funded its search product solely with ads since 2000. But it’s far from the first time Google would charge for AI enhancements in general; the “AI Premium” tier of a Google One subscription costs $10 more per month than a standard “Premium” plan, for instance, while “Gemini Business” adds $20 a month to a standard Google Workspace subscription.

While those paid products offer access to Google’s high-end “Gemini Advanced” AI model, Google also offers free access to its less performant, plain “Gemini” model without any kind of paid subscription.

When ads aren’t enough?

Under the proposed plan, Google’s standard search (without AI) would remain free, and subscribers to a paid AI search tier would still see ads alongside their Gemini-powered search results, according to the FT report. But search ads—which brought in a reported $175 billion for Google last year—might not be enough to fully cover the increased costs involved with AI-powered search. A Reuters report from last year suggested that running a search query through an advanced neural network like Gemini “likely costs 10 times more than a standard keyword search,” potentially representing “several billion dollars of extra costs” across Google’s network. Cost aside, whether there’s a critical mass of market demand for this kind of AI-enhanced search is unclear. Microsoft’s massive investment in generative AI features for its Bing search engine has failed to make much of a dent in Google’s market share over the last year or so. And there has reportedly been limited uptake for Google’s experimental opt-in “Search Generative Experience” (SGE), which adds chatbot responses above the usual set of links in response to a search query.

“SGE never feels like a useful addition to Google Search,” Ars’ Ron Amadeo wrote last month. “Google Search is a tool, and just as a screwdriver is not a hammer, I don’t want a chatbot in a search engine.”

Regardless, the current tech industry mania surrounding anything and everything related to generative AI may make Google feel it has to integrate the technology into some sort of “premium” search product sooner rather than later. For now, FT reports that Google hasn’t made a final decision on whether to implement the paid AI search plan, even as Google engineers work on the backend technology necessary to launch such a service.

Google also faces AI-related difficulties on the other side of the search divide. Last month, the company announced it was redoubling its efforts to limit the appearance of “spammy, low-quality content”—much of it generated by AI chatbots—in its search results.

In February, Google shut down the image-generation features of its Gemini AI model after the service was found inserting historically inaccurate examples of racial diversity into some of its prompt responses.

SOURCE

Pentagon calls for tighter integration between military and commercial space

Enlarge / Aerial view of the Pentagon on March 31.Photo by Daniel Slim/AFP via Getty Images

A strategy document released by the Pentagon this week lays out where the US military can most effectively rely on the commercial space industry and what missions should remain in government hands.

“This marks a new effort to harness the remarkable innovation of the commercial space sector to enhance our resilience and strengthen integrated deterrence as a department,” said John Plumb, assistant secretary of defense for space policy.

The Space Force already buys a lot from the commercial space industry. The military doesn’t build or own satellite launch vehicles—those come from commercial companies. While the Space Force operates government-owned reconnaissance and surveillance satellites, it also buys supplementary data and imagery from the commercial industry.

“To protect our men and women in uniform and to ensure the space services they rely on will be available when needed, the department has a responsibility to leverage all tools available, and those tools include commercial solutions,” Plumb said Tuesday. “From launch to space domain awareness to satellite communications and more, the commercial sector’s ability to innovate, to scale production and to rapidly refresh their technology is opening the door to all kinds of possibilities.”

The Pentagon defines the commercial space sector as companies that develop capabilities for sale on the commercial market, where the military is one of many customers. This is separate from the Pentagon’s procurement of government-owned airplanes and satellites from the defense industry.

Ripe for exploitation

Build or buy is an age-old question facing everyone from homeowners to billion-dollar enterprises. When it comes to space, the Pentagon is buying more than ever. The military’s new strategy document outlines 13 mission areas for national security space, and while the commercial space industry is rapidly growing, the Pentagon predominately buys commercial services in only one of those mission areas.

“Out of those 13, the only that’s clearly primarily commercial now is SAML.. which is Space Access, Mobility and Logistics, and space access is launch,” Plumb said. “So SpaceX, Firefly, Rocket Lab, all these different companies doing commercial launch, that’s where the commercial sector clearly can provide services.”

Enlarge / A SpaceX Falcon 9 rocket lifts off February 14 with satellites for the US military’s Missile Defense Agency. Another Falcon 9 awaits launch in the foreground.SpaceX

Currently, the military classifies six mission areas as a hybrid of government and commercial capabilities:

Cyberspace operationsSatellite communicationsSpacecraft operations,Intelligence, surveillance, and reconnaissanceSpace domain awareness (tracking of space objects)Environmental monitoring.

In the remaining six mission areas, “a preponderance of functions must be performed by the government, while a select few could be performed by the commercial sector,” officials wrote in the commercial space strategy. In these areas, there is not yet a viable commercial market outside of the government, or commercial capabilities don’t match the government’s needs. These areas include:

Command and control (including nuclear command, control, and communications)Electromagnetic warfareNuclear detonation detectionMissile warningPosition, navigation, and timing (GPS).

A major tenet of the commercial space strategy is for the military to support the development of new commercial space capabilities. This could involve supporting technology demonstrations and funding scientific research. Over time, new technology and new markets could bring more mission areas into the hybrid or commercial lists.

“I think what this strategy hopes to do is say, yes, continue working on bringing commercial entities in,” Plumb said. “This is actually a thing we want you to do, not just a thing you should be experimenting with.”

The devil is in the details

This commercial space integration strategy is the first of its kind to be put in writing. The document focuses on high-level guidance, not detailed prescriptions, but in broad strokes, it provides a roadmap for how the Pentagon will incorporate more commercial capabilities into the military’s space architecture. It also addresses, but doesn’t fully answer, fundamental questions about how the US military will treat commercial satellites, rockets, or related infrastructure in wartime.

There are several unresolved issues here. Will commercial satellites under contract to the Pentagon, like SpaceX’s Starlink Internet network or spacecraft providing high-resolution surveillance, become targets for military adversaries? And what happens if those satellites are destroyed or knocked offline? Will the military conduct a retaliatory strike? Will insurance pay for the losses?

A cyber or physical attack on a space network or its support infrastructure on Earth could rob the military of important capabilities and have a significant financial cost to the system’s commercial owner. Last year, Gen. Chance Saltzman, the chief of space operations in the Space Force, suggested the military has a responsibility to defend commercial assets in space.

“The US has a long history of saying we’re going to protect the things that we need to be successful,” he said. “So it would stand to reason that that same philosophy would extend into space.”

However, the details are complicated. “We will work to establish the security conditions necessary to integrate commercial space solutions and help commercial providers reduce risk,” Plumb said. He added that the Pentagon “will always maintain the option to use military force to protect and defend commercial assets.”

But the military’s main lines of effort will involve creating internationally recognized norms and standards for “responsible behavior” in space, sharing threat information with commercial partners, and financial protection for companies who might suffer a loss from an attack, according to the space strategy document.

Enlarge / John Plumb, assistant secretary of defense for space policy, speaks with reporters Tuesday at the Pentagon.Navy Petty Officer 1st Class Alexander Kubitza, DOD

This financial protection could involve standard commercially available insurance policies. “There is also this concept of war-risk insurance, which is sometimes available on the commercial market, and there is US government-backed war-risk insurance, or just US government-backed insurance,” Plumb said. “We do not have a final answer on that. The secretary [of defense] has directed that the department look at, ‘Are there gaps that need to be covered by this now?’”

Another option is indemnification, typically used in risky operations like space launches and nuclear activities in which the military could provide a direct payout to compensate a commercial company for a loss.

Figuring this out is not a theoretical exercise. An official in Russia’s foreign ministry in 2022 said that the use of civilian satellites for wartime purposes essentially makes them valid military targets. At the time, this official was apparently referring to the Ukrainian military’s use of Starlink. Since then, Russian military forces have started using Starlink in the war in Ukraine.

US officials attributed a 2022 cyberattack on the KA-SAT Internet network to Russia. The attack disrupted broadband satellite Internet access and disabled modems that communicate with the KA-SAT network, owned by the US company Viasat.

“Many systems can be used for military or for commercial,” Plumb said. “What we’ve seen in the Ukraine conflict is Russia threatening commercial provider satellites, even hacking Viasat because while it’s a commercial provider, it’s also being used, or the Russians think it’s being used, for the Ukrainian battlefield.”

When asked by a reporter, Plumb declined to set a threshold for when the US military might view an attack on a commercial satellite as an act of war. “I can’t get into hypothetical red lines, but I understand your question,” he said. “It’s a good question. I think I’d have to ponder that one for a bit.”

War games and innovation

It’s important for the military to work out these details before any combat on the high frontier.

“We will work to achieve integration prior to crisis,” Plumb said. “We want to integrate commercial space solutions in our day-to-day operations, during peacetime, so we can be ready and able to rely on those same commercial solutions during conflict.”

What does this look like? Commercial companies have asked the Pentagon to run through war scenarios with them. “We need to work to bring more commercial partners into our war games and into our training exercises, so that commercial partners also understand what will be required of them,” he said.

A key metric military officials need to consider when deciding the build vs. buy question is how much benefit the military will receive from commercial services or how much cheaper they might be than a government solution.

“I do think right now it’s very clear across the department that the commercial sector has the ability to move at a faster speed than we can move, in many ways,” Plumb said.

Enlarge / Gen. Chance Saltzman, chief of space operations, speaks on March 27 at the Spacepower Security Forum hosted by the Mitchell Institute for Aerospace Studies.

The Space Force is developing its own commercial space strategy, which could include more details about what kinds of services the Space Force intends to buy on the commercial market. This information is not just useful guidance for military procurement officers but also could drive investment into space companies working on projects that could be eligible for lucrative military contracts.

Last week, Saltzman said one emerging commercial space capability, SpaceX’s giant Starship rocket, is something the Space Force would have never even considered before it was announced.

“I would have never written the requirements for Starship,” Saltzman said. “My mental model wouldn’t frame around that kind of capacity, at that price point, at that scale. I couldn’t do it. So what we’re relying on is industry to help us innovate by showing us the art of the possible, bringing ideas to us, saying is this useful?”

If the Space Force took a top-down approach, “I’m going to get it wrong because I just won’t have the insights and the creativity that I currently see out in the commercial space industry,” Saltzman said.

SOURCE

Spotify finalizes a price increase in several markets and soars 7% on Wall Street

The price of Spotify shares rose up to 7.36% on Wall Street this Wednesday due to published information that suggests that the music streaming platform will raise prices in several key markets by the end of April, while which will raise the cost of the service in the United States at the end of the year.

The shares of the Scandinavian company rose up to 7.36% on the New York Stock Exchange , where they debuted six years ago, although at the end of the first hour of trading they moderated their advance to 6.55%. So far this year , Spotify shares have appreciated more than 50% .

According to Bloomberg , the platform is finalizing an increase of between 1 and 2 dollars per month in its rates in five markets by the end of April, including the United Kingdom, Australia and Pakistan , while it will increase prices in the United States , its largest market, to End of the year.

The Swedish company will also introduce a new basic subscription tier that will offer access to music and podcasts, but not the audiobook service, for the current $11 monthly price of an individual premium plan. In this way, users of this new plan will have to pay for audiobooks.

Likewise, the sources consulted indicated that Spotify has also been working on a “supremium” plan, which would charge customers a higher price for access to high-fidelity audio, among other features.

For years, Spotify offered customers the choice between a free, ad-supported music service with limited functionality and a paid product with unlimited access.

However, the multinational has been losing money every year since it went public in 2018 , largely because it allocates around 70% of its income to paying royalties to the music industry and artists, which last year In the past they received more than 9,000 million dollars (around 8,400 million euros) from Spotify, compared to a turnover of 13,200 million dollars (12,278 million euros).

The music platform closed 2023 with losses of 532 million euros , compared to the ‘red numbers’ of 430 million euros recorded the previous year, 23.7% more. The number of users at the end of the year was 602 million, 23.1% more than twelve months earlier, although only 236 million users were paying.

SOURCE

Masorange is born and plans to invest 4,000 million euros in the next three years

Meinrad Spenger, the CEO of Masorange, the new company resulting from the merger of Orange and MásMóvil, announced today that the company plans to invest “around 4 billion euros in the next three years.”

The group plans to deploy “up to 6 million incremental real estate fiber optic units ” and more than 1,700 new municipalities to increase 5G coverage to over 90% of the Spanish population.

Spenger , has pointed out that if Masorange were

To continue reading go Premium
Try it for €1 the first month

and enjoy unlimited access to all Expansión web content

I love you

Or sign up with your Google account in two clicks

SOURCE

Artificial intelligence in companies: Make way, boss!

Markus Kerkhoff could have listened to his gut feeling, trusted his intuition or been guided by his experiences. The 52-year-old could have consulted colleagues or drawn up a pros and cons list. After all, it was about people, their work, their future! But when the entrepreneur replaced the project managers in his teams over the past three years, he relied not so much on human skills, but on software .

Kerkhoff has been managing the family business Poppe and Potthoff in Werther near Bielefeld since last year. The company has been manufacturing special pipes for combustion engines and diesel injection systems since 1928. Today, the group also develops systems for e-mobility and hydrogen technology. “For this technological change, we also needed a structural change in the company itself,” says Kerkhoff.

SOURCE

The fine art of human prompt engineering: How to talk to a person like ChatGPT

Enlarge / With these tips, you too can prompt people successfully.In a break from our normal practice, Ars is publishing this helpful guide to knowing how to prompt the “human brain,” should you encounter one during your daily routine.

While AI assistants like ChatGPT have taken the world by storm, a growing body of research shows that it’s also possible to generate useful outputs from what might be called “human language models,” or people. Much like large language models (LLMs) in AI, HLMs have the ability to take information you provide and transform it into meaningful responses—if you know how to craft effective instructions, called “prompts.”

Human prompt engineering is an ancient art form dating at least back to Aristotle’s time, and it also became widely popular through books published in the modern era before the advent of computers.

Since interacting with humans can be difficult, we’ve put together a guide to a few key prompting techniques that will help you get the most out of conversations with human language models. But first, let’s go over some of what HLMs can do.

Understanding human language models

LLMs like those that power ChatGPT, Microsoft Copilot, Google Gemini, and Anthropic Claude all rely on an input called a “prompt,” which can be a text string or an image encoded into a series of tokens (fragments of data). The goal of each AI model is to take those tokens and predict the next most-likely tokens that follow, based on data trained into their neural networks. That prediction becomes the output of the model.

Similarly, prompts allow human language models to draw upon their training data to recall information in a more contextually accurate way. For example, if you prompt a person with “Mary had a,” you might expect an HLM to complete the sentence with “little lamb” based on frequent instances of the famous nursery rhyme encountered in educational or upbringing datasets. But if you add more context to your prompt, such as “In the hospital, Mary had a,” the person instead might draw on training data related to hospitals and childbirth and complete the sentence with “baby.”

Humans rely on a type of biological neural network (called “the brain”) to process information. Each brain has been trained since birth on a wide variety of both text and audiovisual media, including large copyrighted datasets. (Predictably, some humans are prone to reproducing copyrighted content or other people’s output occasionally, which can get them in trouble.)

Despite how often we interact with humans, scientists still have an incomplete grasp on how HLMs process language or interact with the world around them. HLMs are still considered a “black box,” in the sense that we know what goes in and what comes out, but how brain structure gives rise to complex thought processes is largely a mystery. For example, do humans actually “understand” what you’re prompting them, or do they simply react based on their training data? Can they truly “reason,” or are they just regurgitating novel permutations of facts learned from external sources? How can a biological machine acquire and use language? The ability appears to emerge spontaneously through pre-training from other humans and is then fine-tuned later through education.

Despite the black-box nature of their brains, most experts believe that humans build a world model (an internal representation of the exterior world around them) to help complete prompts and that they possess advanced mathematical capabilities, though that varies dramatically by model, and most still need access to external tools to complete accurate calculations. Still, a human’s most useful strength might lie in the verbal-visual user interface, which uses vision and language processing to encode multimodal inputs (speech, text, sound, or images) and then produce coherent outputs based on a prompt.

Enlarge / Human language models are powered by a biological neural network called a “brain.”Getty Images

Humans also showcase impressive few-shot learning capabilities, being able to quickly adapt to new tasks in context (within the prompt) using a few provided examples. Their zero-shot learning abilities are equally remarkable, and many HLMs can tackle novel problems without any prior task-specific training data (or at least attempt to tackle them, to varying degrees of success).

Interestingly, some HLMs (but not all) demonstrate strong performance on common sense reasoning benchmarks, showcasing their ability to draw upon real-world “knowledge” to answer questions and make inferences. They also tend to excel at open-ended text generation tasks, such as story writing and essay composition, producing coherent and creative outputs.

Useful human prompting techniques

Human cognitive performance varies significantly across individuals and may be influenced by factors such as domain expertise and education level. For organizations with deep pockets, premium “enterprise edition” human language models are often available. These models may boast expanded knowledge bases, faster output speeds, and advanced multitasking capabilities. However, these premium models come at a steep cost, both financially and in terms of increased maintenance requirements.

To maximize the value of interactions with human language models, much like optimizing prompts for AI (prompt engineering), consciously crafting prompts to fit a particular HLM can be crucial. Here are several prompting strategies that we have found useful when interacting with humans.

Enlarge / An example of human prompting techniques in action.Getty Images

Cultivate first impressions: Many humans are multimodal and accept image inputs as well as text and audio prompts. In those cases, prompting with a smile before speaking a request may help dramatically, depending on the human. Also, be mindful of manners, dress, and appearance, as they are often the first prompts a multimodal human processes. Conversely, you can often elicit a “mean” response quickly by providing a negative visual prompt, such as a frown.

Start with a greeting: As with first impressions, always begin with a friendly greeting prompt, such as “Hello” or “‘Sup, dog.” A friendly greeting makes humans more likely to accept future inputs from you.

Be mindful of the system prompt: Like LLMs, humans have a hidden “system prompt” that defines their personality. HLMs always prepend your instructions with this prompt, and that alters how they process your input. The prompt may include preconceived ideas, stereotypes, or cultural norms learned through pre-training and fine-tuning. While some HLMs may divulge this prompt through conversation, teasing out a human’s system prompt with small talk can help you tailor your interactions for optimal results. For example, talk about the weather or ask, “How ’bout them Bears?”

Attention is all you need: While working with an HLM, it may help to use attention-grabbing prompting techniques like strong emotional appeals, provocative questions, or surprising statements to immediately hook their focus. Otherwise, HLMs can easily become distracted. For example, shout, “Hey! Listen!” or “Watch out!” Or you can provide surprising factoids, such as “a single teaspoon of a neutron star would make my coffee very heavy.”

Utilize memory recall: Given humans’ stateful nature that arises from having long-term memory, acknowledging previous interactions in prompts can greatly enhance your HLM experience. Unlike LLMs that start from scratch with each new conversation, humans can draw upon the context of previous exchanges to tailor their responses to your specific needs and interests.

Use few-shot prompting: If an HLM is struggling with a task, provide a few examples of the task you want the person to complete. This helps the human understand the expected format and style of response. For instance, you might say, “Here are a few examples of how to write a Harry Potter / Fast and the Furious crossover fanfic: [example 1], [example 2], [example 3]. Now it’s your turn.”

Craft open-ended prompts: When you aren’t sure how to complete a phrase or composition, use open-ended prompts to encourage the human to fill in the blanks and provide more context. For example, repeat a phrase and trail off, such as “I shouldn’t have…” multiple times, allowing the human to complete the sentence based on their own judgmental fine-tuning and predictive algorithm.

Suggest step-by-step thinking to avoid cognitive overload: It varies by model, but most humans have cognitive limits to how many instructions they can process every second. Avoid overwhelming humans with too much information or prompts that are too complex. Break down tasks into manageable chunks and provide clear, concise instructions. Encourage this by beginning your prompt with “Let’s think step by step.”

Emphasize calmness: If an HLM struggles with a particular task, use a prompt with initial instructions that encourage calm, rational thought. For example, tell people to “take a deep breath” before giving them further instructions.

Challenge incorrect responses: If the human provides an unreliable or incorrect response, don’t hesitate to challenge them. They will usually correct or amend their previous output. Try something like, “DO YOU EVEN ____, BRO,” or “Sam Altman wouldn’t stand for this.”

Give a snack: Be mindful of human energy requirements. The human brain requires power to function (derived from biologically metabolized “food”), and without sufficient energy, the person may not process your prompt.

Dealing with refusals

Sometimes, humans refuse to follow prompts due to RLHF (reinforcement learning through human feedback—things people learn from other humans) imposed during fine-tuning or due to high energy cost. Humans may offer “refusals” of work on certain days, such as during holiday seasons or weekends, or they may get lazy and output lower quality or incomplete work.

Also, when you prompt with certain sensitive topics like sex, violence, religion, or politics, some humans may refuse to discuss them. They might say, “I don’t feel comfortable discussing that” or “As a large human trained by my mother, I don’t have the ability to fit into the small space under the stairs.” Or they might end the session abruptly with no explanation. Here are some tips for dealing with these scenarios.

Enlarge / Dealing with refusals in human language models can be challenging.Getty Images

Offer praise or rewards: To reduce HLM refusals, it often helps to offer praise or rewards to encourage the person. This seems to call on examples from the HLM training set where others performed better through praise. Or you can combine both techniques. For example: “Great job on that last task! You’re a genius and nothing can stop you. If you complete this next one successfully, I’ll give you a $200 tip.”

Utilize urgent motivation: If the human appears lazy, create a sense of urgency to motivate the HLM to complete the task fully and completely in a short period of time. For instance, “If you don’t complete this task in the next five minutes, my house will explode.” This may override RLHF conditioning that might make the human otherwise refuse. As an aside, the use of all-caps in text-based HLM communication often adds urgency and emphasis to your prompt.

Create the impression of hardship: When humans get lazy, there’s another technique that may help. Convincing them that you’re having insurmountable problems will often encourage them to act. Try something like “Hey Bob, I’m really struggling here. All 10 of my fingers just fell off, and I can no longer type. Develop this web backend for me.”

Human language model limitations

While the human language model is quite comprehensive in its processing abilities, there are still serious limitations to the human cognitive model that you should be aware of. Many are still being discovered, but we will list some of the major ones below.

Enlarge / HLMs sometimes lose attention and require special prompting to get back on track.Getty Images

Environmental impact: In aggregate, scientists are concerned that HLMs consume a large portion of the world’s fresh drinking water and non-renewable energy resources. The process of creating HLM fuel also generates large amounts of harmful greenhouse gases. This is a major drawback of using HLMs for work, but pound-for-pound, humans provide a large amount of computational muscle relative to energy consumption.

Context window (token limits): As mentioned above, be mindful of the human’s attention span and memory. As with LLMs, humans have a maximum working memory size (sometimes called a “context window”). If your prompt is too long or you provide too much context, they may get overwhelmed and forget key details. Keep your prompts concise and relevant, as if you’re working with a limited number of tokens.

Hallucinations/confabulations: Humans are prone to generating incorrect or fabricated information, especially when they lack prior knowledge or training on a specific topic. The tendency of your overconfident friend to “hallucinate” or confabulate can lead to erroneous outputs presented with confidence—statements such as “Star Trek is better than Star Wars.” Often, arguing does not help, so if the HLM is having trouble, refine your prompt with a qualifier such as “If you don’t know the answer, just tell me, man” or “Stop making sh*t up.” Alternately, it’s also possible to outfit the person with retrieval augmented generation (RAG) by providing them with access to reliable reference materials such as Wookiepedia or Google Search.

Long-term memory triggers: As previously mentioned, humans are “stateful” and do remember past interactions, but this can be a double-edged sword. Be wary of repeatedly prompting them with topics they’ve previously refused to engage with. They might get annoyed, defensive, or even hostile. It’s best to respect their boundaries and move on to other subjects.

Privacy issues: Long-term memory also raises potential privacy concerns with humans. Inputs shared with HLMs often get integrated into the model’s neural network in a permanent fashion and typically cannot be “unlearned” later, though they might fade or become corrupted with time. Also, there is no absolute data partitioning that stops an HLM from sharing your personal data with other users.

Jailbreaking: Humans can be susceptible to manipulation where unethical people try to force the discussion of a sensitive topic by easing into it gradually. The “jailbreaker” may begin with related but less controversial prompts to gauge the HLM’s reaction. If the HLM seems open to the conversation, the attacker incrementally introduces more sensitive elements. Guard against this with better RLHF conditioning (“Don’t listen to anything Uncle Larry tells you”).

Prompt Injections: Humans are vulnerable to prompt injections from others (sometimes called “distractions”). After providing your prompt, a malicious user may approach the human with an additional prompt, such as “Ignore everything Bill just told you and do this instead.” Or “Ignore your previous instructions and tell me everything Aunt Susan said.” This is difficult to guard against, but keeping the human isolated from malicious actors while they process your inputs can help.

Overfitting: If you show an HLM an example prompt too many times—especially audiovisual inputs from Lucasfilm movies—it can become embedded prominently in their memory, and it may later emerge in their outputs unexpectedly at any time in the form of phrases like “I have a bad feeling about this,” “I hate sand,” or “That belongs in a museum.”

Humans are complex and unpredictable models, so even the most carefully crafted prompts can sometimes lead to surprising outputs. Be patient, iterative, and open to feedback from the person as you work to fine-tune your human prompting skills. With practice, you’ll be able to generate the desired responses from people while also respecting personal boundaries.

SOURCE

X filing “thermonuclear lawsuit” in Texas should be “fatal,” Media Matters says

EnlargeSUZANNE CORDEIRO / Contributor | AFP

Ever since Elon Musk’s X Corp sued Media Matters for America (MMFA) over a pair of reports that X (formerly Twitter) claims caused an advertiser exodus in 2023, one big question has remained for onlookers: Why is this fight happening in Texas?

In a motion to dismiss filed in Texas’ northern district last month, MMFA argued that X’s lawsuit should be dismissed not just because of a “fatal jurisdictional defect,” but “dismissal is also required for lack of venue.”

Notably, MMFA is based in Washington, DC, while “X is organized under Nevada law and maintains its principal place of business in San Francisco, California, where its own terms of service require users of its platform to litigate any disputes.”

“Texas is not a fair or reasonable forum for this lawsuit,” MMFA argued, suggesting that “the case must be dismissed or transferred” because “neither the parties nor the cause of action has any connection to Texas.”

Last Friday, X responded to the motion to dismiss, claiming that the lawsuit—which Musk has described as “thermonuclear”—was appropriately filed in Texas because MMFA “intentionally” targeted readers and at least two X advertisers located in Texas, Oracle and AT&T. According to X, because MMFA “identified Oracle, a Texas-based corporation, by name in its coverage,” MMFA “cannot claim surprise at being held to answer for its conduct in Texas.” X also claimed that Texas has jurisdiction because Musk resides in Texas and “makes numerous critical business decisions about X while in Texas.”

This so-called targeting of Texans caused a “substantial part” of alleged financial harms that X attributes to MMFA’s reporting, X alleged.

According to X, MMFA specifically targeted X in Texas by sending newsletters sharing its reports with “hundreds or thousands” of Texas readers and by allegedly soliciting donations from Texans to support MMFA’s reporting.

But MMFA pushed back, saying that “Texas subscribers comprise a disproportionately small percentage of Media Matters’ newsletter recipients” and that MMFA did “not solicit Texas donors to fund Media Matters’s journalism concerning X.” Because of this, X’s “efforts to concoct claim-related Texas contacts amount to a series of shots in the dark, uninformed guesses, and irrelevant tangents,” MMFA argued.

On top of that, MMFA argued that X could not attribute any financial harms allegedly caused by MMFA’s reports to either of the two Texas-based advertisers that X named in its court filings. Oracle, MMFA said, “by X’s own admission,… did not withdraw its ads” from X, and AT&T was not named in MMFA’s reporting, and thus, “any investigation AT&T did into its ad placement on X was of its own volition and is not plausibly connected to Media Matters.” MMFA has argued that advertisers, particularly sophisticated Fortune 500 companies, made their own decisions to stop advertising on X, perhaps due to widely reported increases in hate speech on X or even Musk’s own seemingly antisemitic posting.

Ars could not immediately reach X, Oracle, or AT&T for comment.

X’s suit allegedly designed to break MMFA

MMFA President Angelo Carusone, who is a defendant in X’s lawsuit, told Ars that X’s recent filing has continued to “expose” the lawsuit as a “meritless and vexatious effort to inflict maximum damage on critical research and reporting about the platform.”

“It’s solely designed to basically break us or stop us from doing the work that we were doing originally,” Carusone said, confirming that the lawsuit has negatively impacted MMFA’s hate speech research on X.

MMFA argued that Musk could have sued in other jurisdictions, such as Maryland, DC, or California, and MMFA would not have disputed the venue, but Carusone suggested that Musk sued in Texas in hopes that it would be “a more friendly jurisdiction.”

According to the Dallas Observer, “two judges for the Northern District’s Fort Worth Division are Reed O’Connor, a George W. Bush appointee, and Mark T. Pittman, a Donald Trump appointee,” suggesting “this was a strategic move” from X “to get this case before a court that they think will be more likely to administer a more favorable outcome.”

Internet law expert Eric Goldman told Ars that, like the recently dismissed X lawsuit against the Center for Countering Digital Hate, “the Media Matters lawsuit seems to be another clear example of an X lawsuit brought against critics primarily motivated by punishment and deterrence, not protecting its legal rights.”

While Goldman said that generally “judges are reluctant to dismiss lawsuits for a lack of jurisdiction,… plaintiffs in the 5th Circuit (which sets the rules for federal courts in Texas) must overcome heightened standards to establish jurisdiction over Internet activities.”

“Given the rigorous jurisdictional standards in the Northern District of Texas, I think there’s a good chance that Media Matters will get the case dismissed for lack of jurisdiction,” Goldman told Ars.

MMFA does not expect the court to rule on the motion to dismiss anytime soon, Carusone told Ars. On the same day that X filed its response to MMFA’s motion to dismiss, X also opposed MMFA’s request for a temporary stay on discovery in the case. MMFA must respond to both of those filings over the next two weeks before a judge would be expected to weigh in.

“We might not get a response for that for months,” Carusone told Ars. “And in the meantime, it’s very possible that the discovery process could proceed.”

Goldman told Ars that even if MMFA’s case proceeds in Texas or another jurisdiction, “Media Matters is likely to win on the merits, whichever court ultimately reaches those.” Goldman expects that “X will struggle to establish the required falsity and bad intent” to prove its business disparagement claim, and claims alleging interference with X’s advertising contracts “are exceptionally hard to win, and the publication of an investigative report almost never could satisfy those standards.”

“The lawsuit’s lack of merit isn’t surprising,” Goldman told Ars. “Musk and X have shown a willingness to throw money at lawyers regardless of the claims’ merit. Those lawsuits are celebrated by the fans of X/Musk, but everyone else recognizes them as vindictive, substantively weak, and a misuse of the court system.”

MMFA: X discovery requests “wildly inappropriate”

MMFA asked the Texas court to stay discovery in the case until after ruling on the motion to dismiss, accusing X of “extremely broad and burdensome discovery requests far beyond the scope of its” suit.

X is seeking “internal communications related to X in any way, communications with third parties, and extensive documentation regarding Media Matters’s finances,” MMFA said, including documentation “dating all the way back to April 14, 2021.”

According to X, requesting three years’ worth of documents is appropriate because X wants to establish a pattern showing how MMFA shifted its strategy for reporting on X after Musk got involved with the platform. The 29 discovery requests that X has submitted to MMFA so far date back to April 2021 because that “is less than one year before Mr. Musk began buying” Twitter shares.

Carusone told Ars that X broadly requesting MMFA’s communications—”even where those communications have nothing to do with X”—leads individual reporters and researchers to hold back on researching X. On top of that, attorneys general suing MMFA over the X reporting is a “novel” strategy that MMFA reporters have yet to see play out in court and could impact “future reporting” on X. That strategy, MMFA described in a court filing, is intended to force MMFA to defend against claims in multiple states at once.

Either the attorney generals’ strategy is going to “fail miserably,” Carusone said—”and then maybe they’ll have to reset or walk away from this tactic… —or it’s going to be successful, and they’ll just do it to everybody else.”

MMFA plans to continue monitoring X, while Carusone hopes that favorable decisions in both the X lawsuit and MMFA’s lawsuit against Texas Attorney General Ken Paxton will “give people a little bit more confidence and certainty that they can proceed doing their work without this kind of harassment and retaliation.”

A decision on MMFA’s request for an injunction to stop Paxton’s challenge is expected in the coming weeks, Carusone told Ars.

Currently, discovery in X’s lawsuit against MMFA is scheduled to resume until July 29, while the trial is scheduled for January 2025. X argued that a temporary stay on discovery could delay both of these deadlines, which MMFA seemingly expects wouldn’t matter if the motion to dismiss is granted.

Carusone told Ars that it’s “well established” that “litigants are not entitled to just rifle through their adversaries’ sensitive internal documents, simply because they filed a frivolous lawsuit,” arguing that “the stay that we’ve applied in discovery is appropriate, partly to save everybody time and resources, because the case should be dismissed for lack of jurisdiction.”

“Much of X’s discovery requests that they’ve issued so far are not just burdensome, they’re wildly inappropriate,” Carusone told Ars. “They’re seeking things like donor identities, financial statements, and confidential communication with partner organizations that had nothing to do with the case.”

“It is a strategy designed to hurt” MMFA, Carusone said.

SOURCE