Michael Fassnacht’s Post

#AI’s Black Boxes Just Got a Little Less Mysterious Researchers claim to have found clues about the inner workings of large language models, possibly helping to curb their potential threats. “One of the weirder, more unnerving things about today’s leading artificial intelligence systems is that nobody — not even the people who build them — really knows how the systems work. That’s because large language models, the type of AI systems that power #ChatGPT and other popular chatbots, are not programmed line by line by human engineers, as conventional computer programs are. Instead, these systems essentially learn on their own, by ingesting vast amounts of data and identifying patterns and relationships in language, then using that knowledge to predict the next words in a sequence. One consequence of building AI systems this way is that it’s difficult to reverse-engineer them or to fix problems by identifying specific bugs in the code. Right now, if a user types “Which American city has the best food?” and a chatbot responds with “Tokyo,” there’s no real way of understanding why the model made that error, or why the next person who asks may receive a different answer. […] But this week, a team of researchers at the AI company Anthropic announced what they called a major breakthrough — one they hope will give us the ability to understand more about how AI language models actually work, and to possibly prevent them from becoming harmful. The team summarized its findings in a blog post called “Mapping the Mind of a Large Language Model.” By Kevin Roose https://lnkd.in/dFnNaVF5 #research #ArtificialIntelligence #GenerativeAI #LLMs #technology #health #security #humanity

Excerpt: The media frenzy surrounding ChatGPT and other large language model artificial intelligence systems spans a range of themes, from the prosaic – large language models could replace conventional web search – to the concerning – AI will eliminate many jobs – and the overwrought – AI poses an extinction-level threat to humanity. All of these themes have a common denominator: large language models herald artificial intelligence that will supersede humanity. But large language models, for all their complexity, are actually really dumb. And despite the name “artificial intelligence,” they’re completely dependent on human knowledge and labor. They can’t reliably generate new knowledge, of course, but there’s more to it than that. ChatGPT can’t learn, improve or even stay up to date without humans giving it new content and telling it how to interpret that content, not to mention programming the model and building, maintaining and powering its hardware. To understand why, you first have to understand how ChatGPT and similar models work, and the role humans play in making them work.

This is a cringeworthy clip of Scott Aaronson engaging in the debate over how we understand artificial intelligence, particularly language models like ChatGPT. The video segment dissects the "religion of justaism," which Aaronson claims is a reductionist argument regarding ChatGPT. He argues that reducing ChatGPT to a mere "stochastic parrot," "next-token predictor," "function approximator," or "gigantic autocomplete" is not fair to language models. His counterargument is not why or how ChatGPT is not these things; instead, it is a reductionist argument. He says, "What these people never do, what it never occurs to them to do is to ask the next question: What are you justa, right? Aren't you justa bundle of neurons and synapses, right?" Aaronson, a master orator, proves the limitations of reductionism by mirroring its flaws. The reason why humans never ask themselves if they are a bundle of neurons when evaluating ChatGPT is because it is a non-sequitur. Humans are biological, while AI is an algorithm trained on human-generated data to perform specific tasks without consciousness or personal experience. Therefore, making such a comparison would be a kind of category mistake. Scott never seriously considers that biology makes any difference since views like computational functionalism view biology and silicon as identical. Rather than directly comparing humans and language models, Aaronson should have explained AI's unique capabilities or emergent properties beyond simple next-token prediction. These include its capacity to generate novel content, adapt to diverse inputs, or reason in narrow situations. Such a discussion would have provided a more robust foundation for critiquing reductionism or "justaisms" without inaccurately equating AI to human cognitive processes. Scott Aaronson's school of rejoinders is so tired. The person who makes an argument for AI by attacking humans is okay with pretending to know less about themselves to support artificial intelligence. Yet, arguments supporting language models do not have to come at the expense of humans. This is the inverse of anthropomorphism, where intelligent, capable people deliberately underestimate their abilities and fail to attribute humanlike capacities to other humans. Joseph Schumpeter once said that “the first thing a man will do for his ideal is lie,” which is anthropomorphism. A corollary is that after lying for an ideal, we will perform a character assassination, even if it’s against ourselves.

“Briefly summarize of this article.” The capabilities of large language models, which power chatbots like ChatGPT. On most tasks, performance improved predictably and smoothly as the models scaled up—the larger the model, the better it got.these behaviors are not only surprising but unpredictable, and that they should inform the evolving conversations around AI safety, potential, and risk. They called the abilities “emergent,” a word that describes collective behaviors that only appear once a system reaches a high level of complexity. And even if emergence in today’s LLMs can be explained away by different measuring tools, it’s likely that won’t be the case for tomorrow’s larger, more complicated LLMs. “When we grow LLMs to the next level, inevitably they will borrow knowledge from other tasks and other models,” said Xia “Ben” Hu, a computer scientist at Rice University.it is directly to ongoing efforts to predict how LLMs will behave. “These technologies are so broad and so applicable,” he said. “I would hope that the community uses this as a jumping-off point as a continued emphasis on how important it is to build a science of prediction for these things. How do we not get surprised by the next generation of models?”

These are rhetorical tactics that are really effective. This clip is a great illustration of why this doesn't work as an argument. People might say this is "just a rhetorical device," which has led some to think rhetoric is bad. However, as my colleague Brian Sullivan, a classically trained rhetorician, would tell you, rhetoric is a serious discipline with science and philosophy behind it and a long heritage, history, and tradition. But saying it is "just a rhetorical device" doesn't mean--therefore it is invalid. You can use rhetoric to deliver substance or you can use it to deceive and deliver "empty rhetoric." It doesn't mean there is no skill involved but to use it without delivering substance is skillful but devious and usually hides another agenda. To interpret "just a rhetorical device" as a negation of the skill involved or a dismissal of all rhetoric as bad is a logical fallacy just as his argument is one, too. So: "just a stochastic parrot" does not mean that it isn't useful or that it isn't an advance. Here is what it means when I say it: 1) It means that to use the alleged impact of that advance with no proof-of-concept as to its use or social value, and as if it has "emergent reasoning capability" is ill-advised. 2) It means to use that alleged advance as an excuse to scrape, steal, and or hoard data without consent or foreknowledge is questionable at best. 3) It means that selling as a fully tested, flushed out, and stable decisioning system without an expert human (or team of humans) in the loop of any kind is unethical. 4) It means that hiding how many low-wage workers you have laboring in a BPO to keep it running without revealing the level of supervision or the degree to which the model is stable and survives durability testing is dishonest. 5) It means that using anthropomorphic words around it repeatedly is dangerous. 6) It means that defunding and ignoring any competing models and driving those developing them out of business based on the alleged superiority of models that "need" copyrighted data at high volumes at "need" all the high speed graphic cards and AI chips, and eclipse smaller, leaner, more sustainable models could be considered... I don't know...monopolistic...or... Criminal? I don't know. But this is "just a social media post," and I am "just a female PhD" who has been in the arts, social sciences, humanities, and STEM in an interdisciplinary way for "just 25 years." It's all in the word "just" isn't. Funny. That word "just" is polysemic isn't it? Polyvocality is kind of a problem isn't it? Who among us has enough LLM expertise to understand why that is important? #ifykyk #demystifyai

"Recently, there has been considerable interest in large language models: machine learning systems which produce human-like text and dialogue. Applications of these systems have been plagued by persistent inaccuracies in their output; these are often called “AI hallucinations”. We argue that these falsehoods, and the overall activity of large language models, is better understood as bullshit in the sense explored by Frankfurt (On Bullshit, Princeton, 2005): the models are in an important way indifferent to the truth of their outputs. We distinguish two ways in which the models can be said to be bullshitters, and argue that they clearly meet at least one of these definitions. We further argue that describing AI misrepresentations as bullshit is both a more useful and more accurate way of predicting and discussing the behaviour of these systems" https://lnkd.in/drPq7MNp

When ChatGPT burst onto the scene late 22, it sent shockwaves through the tech world and beyond. While ML researchers had been quietly experimenting with LLMs for years, the general public was caught off guard by their sudden rise to prominence. Today, LLMs are virtually ubiquitous, powering numerous applications we use daily. However, their inner workings remain shrouded in mystery for most people. We've heard that LLMs "predict the next word" using vast textual datasets, but the details are often treated as an impenetrable black box. What makes LLMs so enigmatic? Unlike conventional software built on explicit human instructions, LLMs emerge from training deep neural networks on billions of words of natural language data. As a result, no one fully comprehends the complexity under the hood. While unraveling the complete picture will take decades of research, experts have begun to shed light on key aspects of how LLMs operate. This is one of the best guides I found that aims at making that knowledge accessible without dense technical jargon, exploring: - Word Vectors: The surprising way LLMs numerically represent and reason about language - Transformers: The core architecture powering models like ChatGPT - Training Process: Why phenomenally large datasets are crucial for high performance Enjoy the read 😀

In the dynamic realm of AI, the integration of Large Language Models (LLMs) with knowledge graphs is a ground-breaking advancement. This fusion not only enhances the accuracy and reliability of LLMs like ChatGPT but also addresses key challenges such as data hallucinations and language biases. Why Knowledge Graphs Matter: Knowledge graphs represent a transformative leap in AI. They organize data from various sources, creating a structured, transparent network of facts. This structure, when integrated with LLMs, empowers them to access verifiable information, improving their reasoning capabilities and reducing errors. Reflecting on the Impact: How will this synergy between knowledge graphs and LLMs reshape industries? Can we anticipate more reliable and transparent AI systems that understand and reason better than ever before? #AI #KnowledgeGraphs #LLMs #Technology #Innovation #business #FutureOfWork

ChatGPT, Copilot, Gemini are all politically biased. Is anybody surprised? Do we want it any other way? If you ask in a clever way, you can force AI chatbots to reveal a political opinion. ChatGPT appears to more leftist, Meta’s LLaMA is more right-wing. This will be used by debaters looking for something to yell at, which we can all ignore. https://lnkd.in/d9qVHE6t AI language models have bias on race, gender, and more. You have seen the stories, and maybe you also heard the noise when Gemini tried to fix it but over-compensated, https://lnkd.in/dXe7VFhU. Trusted AI requires trusted data. And "the Internet" is not trusted data. AI chatbots are trained on material out there on the Internet, then moderated by real people. There is no way that can end up being 100% neutral. Just accept the fact. The good question is: do we want it any other way? Think about asking a question to a person who has absolutely no opinion on anything at all. How useful is that? Even a neutral stand is also a stand. Did we land on the moon? Is the earth flat? Does Illuminati secretly rule the world? When bias cannot be avoided, the next best thing is to be aware it.

As governments scramble to regulate AI and stop the spread of misinformation, can more #AI help? That’s the question a team at the University of #Buffalo are asking. Researchers have used large language models (#LLMs) such as OpenAI’s #ChatGPT and #GoogleGemini to identify #deepfakes of human faces. Presenting their results at last week’s IEEE/CVF Conference On Computer Vision & Pattern Recognition, the team found that currently LLMs cannot match dedicated, state-of-the-art deepfake detection #algorithms. However, their innate ability to process language means they could become powerful tools for identifying fake content and improving human skills in this area. ‘What sets LLMs apart from existing detection methods is the ability to explain their findings in a way that’s comprehensible to humans, like identifying an incorrect shadow or a mismatched pair of earrings,’ said the study’s lead author, Siwei Lyu, PhD, SUNY Empire Innovation Professor in the Department of Computer Science at the University of Buffalo. ‘LLMs were not designed or trained for deepfake detection, but their semantic knowledge makes them well suited for it, so we expect to see more efforts toward this application,’ Lyu added. LLMs look at statistical patterns and relationships between words to generate automated responses. The most up-to-date versions are also able to analyse images, calling on databases of captioned photos to understand links between words and pictures. In testing, ChatGPT was found to identify synthetic artifacts in images 79.5% of the time, and 77.2% on StyleGAN-generated images. The programme was also able to explain how it came to its decisions on the authenticity in plain language. This can help users understand and learn themselves. Examples given in the study report include LLMs pointing out ‘the hair on the left side of the image slightly blurs’ and ‘the transition between the person and the background is a bit abrupt and lacks depth’. ‘Existing deepfake detection models will tell us the probability of an image being real or fake, but they will very rarely tell us why they came to this conclusion. And even if we look into the model’s underlying mechanisms, there will be features that we simply can’t understand,’ said Lyu. ‘Meanwhile, everything ChatGPT outputs is understandable to humans.’ According to Lyu and the team, ChatGPT and other LLMs have a ‘common sense understanding of reality’, such as the symmetry of a human face and the look of a real photograph, and use semantic knowledge to understand what they are looking at. In comparison, deepfake detection technology relies on large datasets which have been labelled real or deepfake.