Stanford AI Lab Blog

I'm not sure if you're trolling or not, but I would suggest you start taking some Robotics AI courses (http://ai.stanford.edu) and reading some books before taking this project on. I'm not sure how far you'd have to go but there's a big different between building a robot that moves and building a robot that can mobilize itself and perform actions in response to stimulation. Also, engineering spec is quite vague. If you boss really asked you to do something like, think twice about taking it on. Loose, vague, and otherwise overreaching projects tend to not end well for those involved. I think my peeps will feel similarly and pass on this one. But I'd be happy to introduce you should you want to come out to a hack night and ask them yourself. :) Best of luck to you.

I Googled for: "Daniel Levy ai" First hit: <<Daniel Levy - Stanford AI Lab: Stanford Artificial Intelligence Laboratory https://ai.stanford.edu › ~danilevy Bio. I currently lead the Optimization team at OpenAI. Before that, I was a Ph.D. student in Computer Science at Stanford University advised by John Duchi.>>

Wrong: it's the Stanford Artificial Intelligence Laboratory (founded 1962 by John McCarthy -- who co-authored the document that first defined AI as a field, invented LISP, time sharing, and garbage collection, heavily influenced the design of ALGOL, and so on: one of the pioneers of computer science): https://ai.stanford.edu/ https://en.wikipedia.org/wiki/John_McCarthy_(computer_scientist)

Not beats experiential education and building an early network if intellectual peers. I'd recommend to a 14-year old an AI camp being offered by the Stanford AI Lab: http://ai.stanford.edu/sailors/

> […] fed large chunks of an article multiple times to an LLM […] So they had to prompt? An LLM? I got this argument before and still don’t get what it’s trying to say. These models do not output anything unless prompted, that’s not any kind of gotcha. On the code outputting front there is a lot of relevant evidence beyond the NYC lawsuit [0]. [0] https://arxiv.org/html/2601.02671?amp=&amp= and https://arxiv.org/abs/2506.12286 and https://ai.stanford.edu/blog/verbatim-memorization/

via wikipedia: "If a program behaves as if it were multiplying, most of us would say that it is, in fact, multiplying. For all I know, Searle may only be behaving as if he were thinking deeply about these matters. But, even though I disagree with him, his simulation is pretty good, so I’m willing to credit him with real thought."[1] [1] A Short Rebuttal to Searle, https://ai.stanford.edu/%7Enilsson/OnlinePubs-Nils/General%20Essays/Ot… I like this quite a bit, and I also wish that Prof. Nilsson were still with us to comment on current LLMs and other systems.

Yes it is; it's just not updating the weights. https://ai.stanford.edu/blog/understanding-incontext/

Thanks. "In-context learning" was the phrase I was looking for. Also, I found these 2 links pretty good too. 1. http://ai.stanford.edu/blog/understanding-incontext/ 2. http://ai.stanford.edu/blog/in-context-learning/ I'm still not completely convinced. Probably need to dwell on the topic longer.

You might be better off looking at MIT OCW search, and selecting video lectures, looking at standford youtube, checking out the 2019 videos https://ai.stanford.edu/stanford-ai-courses/ Most of these look like they're just the slides and syllabus, correct me if I'm wrong.

We talk exactly about clustering pictures in our blog post! https://ai.stanford.edu/blog/banditpam/

A more theoretical but increasingly popular approach to identifying behavior like this is interchange intervention: https://ai.stanford.edu/blog/causal-abstraction/

> Remarkably, conditioning the model on such an “example-based specification” effectively enables the model to adapt on-the-fly to novel tasks whose distributions of inputs vary significantly from the training distribution. The idea that simply minimizing the negative log loss of a single next-word-prediction objective implies this apparent optimization of many more arbitrary tasks – amounting to a paradigm of “learning” during inference time – is a powerful one, and one that raises many questions. http://ai.stanford.edu/blog/in-context-learning/

If retrieval based NLP [0] becomes a thing, then trillion plus parameters models will likely be less of a thing; as very likely, most of these tens to hundreds of billions of parameters are likely over-fitting (better word: memorized) on training data [text corpus] as seen in the case of GPT-3. [0] https://ai.stanford.edu/blog/retrieval-based-NLP/

I think this is a bit too cynical/negative. It's not true that neural net approaches don't work, they just have different strengths and weaknesses (strength being generalization/flexibility/ease of use, weakness being long term coherence, interpretability, reliability). There are plenty of examples of traditional methods being combined with neural net methods (eg https://ai.stanford.edu/blog/qagnn/ , https://arxiv.org/abs/1907.04799 , NeRF), and I think that is likely to happen in this context as well to some extent.

A bonus for complexity can be given according to something like this: https://ai.stanford.edu/blog/topologylayer/

This is the answer. Unfortunately there is a plethora of dubious machine learning and AI tutorial clickbait material appearing on the internet to sell ads. Here are some resources I've found which don't suck if you actually want to learn this stuff: https://ai.stanford.edu/courses/ <- Stanford's AI course materials. https://karpathy.ai/zero-to-hero.html <- Karpathy's "neural networks - zero to hero". The other ones (eg the transformers ones on youtube also seem excellent to me after an initial skim although I haven't got to actually working through them yet). https://ocw.mit.edu/search/?q=artificial%20intelligence and https://ocw.mit.edu/search/?q=machine%20learning (MIT's relevant opencourseware) https://probml.github.io/pml-book/book2.html (Draft book "Probabilistic machine learning: Advanced topics" by Kevin Murphy) <- looks seriously excellent although I've really only taken a cursory dip into it so far. He's also got 2 other books at https://probml.github.io/pml-book/ which are more introductory in nature which I expect are probably great too.

Because Nethack and video games remove most of the reasons that GOFAI failed; for example, everything hard about perception from pixels is removed when you have a little parsed text grid of cells presented to you or are hooked into the game engine and query objects directly. (The real world is not made of a clean little grid of discrete high-level objects like 'dragon' or 'black jelly'.) Moravec's paradox again: humans find perception easy but things like Sokoban hard, while GOFAI approaches find Sokoban so trivial that it's common to use it (or Sudoku) as a toy problem introduction to constraint solving. Nethack literally has levels which are just Sokoban and which are important to solve; GOFAI can push the boulders around perfectly, even though it would be unable to recognize a photograph of 'a boulder' or use the word 'boulder' in a story. Then you have the extensive hand-engineering of expert knowledge which goes into ascension agents and the symbolic winners, above and beyond merely plugging in a Sokoban solver. There are increasing experiments in making DRL agents exploit or initialize from pretrained language models (https://arxiv.org/abs/2005.07648#google https://arxiv.org/abs/2201.12122 https://arxiv.org/abs/2204.01691#google https://ai.stanford.edu/blog/DrRepair/ https://arxiv.org/abs/2005.07648#google https://arxiv.org/abs/2009.03393 https://arxiv.org/abs/2204.00598 come to mind) or reading manuals (https://arxiv.org/abs/1401.5390 all the way back in 2012!), and…