Alan Buxton’s Post

"Artificial intelligence chatbots tend to agree with the opinions of the person using them, even to the point that they nod along to objectively false statements. Research shows that this problem gets worse as language models increase in size, adding weight to concerns that AI outputs cannot be trusted ... "It’s a great example of how large language models are not truth-tracking, they’re not tied to truth. They’re designed to fool us and to kind of seduce us, in a way. If you’re using it for anything in which the truth matters, it starts to get tricky. I think it’s evidence that we have to be very cautious and take the risk that these models expose us to very, very seriously.”" Interesting dilemma - if you agree with the model, you may be subscribing to its hallucination. If you disagree, it may be subscribing to yours 🤔

GIVE AI AN INNER MONOLOGUE It's an interesting experiment covered here => https://lnkd.in/dxqNFtjV It's very similar to Tree of Thought, where multiple outputs are generated and the best one that fits the context of the situation is chosen. The obvious problem with that is token costs. If every query generates four potential outputs, your token costs jump 4x. I stumbled upon a method that I'm extremely happy with, alternatively. A method so simple It's easy to overlook at first. Easy not to even pay any mind to. But while working on my Personality Vector, I prompted my knowledgwbase AI bot to "identify the overall topic and the sentiment of the query. Then search the knowledgebase to determine the general feelings they convey on the topic. Then, based on the viewpoint built by the feelings on the topic from the knowledgebase, answer the query." That's, essentially, self reflection. See, one of the limits of AI is, when you make a statement, it'll tend to agree with you. That's because it has no values, no positions, no feelings on a topic. But if you give it a vector store to serve as a brain, those positions and values can be represented in recall. Anyway, I was extremely happy with the outputs I got and I think building external brains with core emotions and values represented in a RAG database is an inexpensive way to increase output quality. Cheers!

Ever got a response from a chatbot that's weird and wrong? It might have been an LLM hallucination – when the language model gets delusional and gives us answers that sound right but are false and fabricated. Dive in as I explain why LLM hallucinations happen and what we can do to avoid them. Also, stay tuned for my next video where I'll talk about world-famous cases when LLMs hallucinated and caused some big trouble 👀 #llms #hallucinations #llmhallucinations #aihallucinations #finetuning #generativeai #genai #ai #responsibleai

For all my friends exploring artificial intelligence, Large Language Models, and chatbots, I'm sharing this 1 hour 30 minute enlightening talk by Mr. Gustav Söderström at Spotify. This talk helps you understand how AI chatbots function, and it's well worth the full 1:30 to educate yourself. It's like lifting the hood of a car to see how the engine operates, understanding what make the car move when you press the gas pedal. You'll gain insight into what makes the AI tick. No more need you wonder why an AI chatbot is said to "hallucinate." Now you'll know it's a result of mathematics, vectors, supervised fine-tuning (SFT), reinforced learning with human feedback (RLHF) - and how the "temperature control is set". This academic talk will make you feel like you understand how it all happens, even without a tech background or strong math skills. Note 1: You can also easier draw your own conclusions about how human biases in RLHF and SFT will affect an AI's potential bias. It also is good to listen to, to more understand how much work it is to train a LLM in a new language, to fully understand the nuances of a new language for the AI. Breaking the language barriers will probably need new technological breakthroughs and a lot of human hours in preparing data to train the LLM:s to excel in many of the 6.000 languages we use on our planet. Note 2: Thank you Mr Oscar Söderlund for keeping sending links to good information to your old father in my ongoing quest to try to understand and make use of AI chatbots as a patient advocate for rare aortic diseases.

Are you talking about AI. This video is a great capture of this ‘moment in time’ and what it looks like as we move forward.

Time: "No, Today’s AI Isn’t Sentient. Here’s How We Know" (Fei-Fei Li & John Etchemendy) (May 22, 2024) "Artificial general intelligence (AGI) is the term used to describe an artificial agent that is at least as intelligent as a human in all the many ways a human displays (or can display) intelligence. It’s what we used to call artificial intelligence, until we started creating programs and devices that were undeniably “intelligent,” but in limited domains—playing chess, translating language, vacuuming our living rooms. The felt need to add the “G” came from the proliferation of systems powered by AI, but focused on a single or very small number of tasks. Deep Blue, IBM’s impressive early chess playing program, could beat world champion Garry Kasparov, but would not have the sense to stop playing if the room burst into flames. Now, general intelligence is a bit of a myth, at least if we flatter ourselves that we have it. We can find plenty of examples of intelligent behavior in the animal world that achieve results far better than we could achieve on similar tasks. Our intelligence is not fully general, but general enough to get done what we want to get done in most environments we find ourselves in. If we’re hungry, we can hunt a mastodon or find a local Kroger’s; when the room catches on fire, we look for the exit. One of the essential characteristics of general intelligence is “sentience,” the ability to have subjective experiences—to feel what it’s like, say, to experience hunger, to taste an apple, or to see red. Sentience is a crucial step on the road to general intelligence. With the release of ChatGPT in Nov 2022, the era of LLMs began. This instantly sparked a vigorous debate about whether these algorithms might in fact be sentient. The implications of the possible sentience of LLM-based AI has not only set off a media frenzy, but also profoundly impacted some of the world-wide policy efforts to regulate AI. The most prominent position is that the emergence of “sentient AI” could be extremely dangerous for human-kind, possibly bringing about an “extinction-level” or “existential” crisis. After all, a sentient AI might develop its own hopes and desires, with no guarantee they wouldn’t clash with ours. This short piece started as a WhatsApp group chat to debunk the argument that LLMs might have achieved sentience. Our main point here is to argue against the most common defense offered by the “sentient AI” camp, which rests on LLMs’ ability to report having “subjective experiences.” "1. Why some people believe AI has achieved sentience 2. Why they’re wrong" (keep reading) "Li and Etchemendy are co-founders of the Institute for Human-Centered Artificial Intelligence at Stanford University. Li is a professor of Computer Science, and 2023 TIME100 AI honoree. Etchemendy is a professor of Philosophy and former provost of Stanford." Time: https://lnkd.in/gu8fm3GE #ai #airesearch #aireality #aicloud #acloudai #aihype #aidebunking

🧠 Memory in LLMs: Enhancing Conversational AI 🤖 Memory plays a pivotal role in modern Large Language Models (LLMs). It's revolutionizing LLM frameworks by expanding their contextual understanding. These models now retrieve concepts from past interactions, making conversations more nuanced. In essence, LLM memory is like a treasure trove of context-target pairs, fueling better predictions for the next token. 💡 The exciting part? There are numerous avenues to incorporate memory into LLMs, each with unique effects. Stay tuned as we delve deeper into the fascinating world of AI memory! 🚀 #AI #LLM #ConversationalAI #machinelearning

There’s so much talk about #AI. How fast it’s developed, how brilliant it is, how easy to use it, how it would be integrated into every facade of our lives, how inevitable it is, how complicated it is, how scary and beautiful it is all at once… How we're going to deal with the fusion of AI and our humanity is a different discussion which requires #philosophy. But not the academic, convoluted, mystified version. Enjoy my latest musings on our understanding of us 👇