Iqimcaltech Retreat
βThe annual @IQIM_Caltech retreat at Lake Arrowhead is an opportunity to spend the weekend with some of my favorite scientists. It is always an uplifting and illuminating experience.β
What the smart people are recommending. 7786 books, tools, and products endorsed by the thinkers absorb.md tracks. Ranked by how many times each has been recommended across compiled podcasts, papers, posts, and tweets.
βThe annual @IQIM_Caltech retreat at Lake Arrowhead is an opportunity to spend the weekend with some of my favorite scientists. It is always an uplifting and illuminating experience.β
βWell, let's get to that. I mean, so, so yesterday, you know, and I'm glad to say this was the cover of Nature, Nature Journal yesterday.β
βFor some context on this repository, GPT, and language modeling it might be helpful to watch my [Zero To Hero series](https://karpathy.ai/zero-to-hero.html).β
βPlease watch: https://t.co/hgURPFQITX πβ
βThe model is available now on @GoogleAIStudio and the Gemini API. Find out more β https://deepmind.google/blog/gemini-robotics-er-1-6/?utm_source=x&utm_medium=&utm_campaign=&utm_content=β
βThe model is available now on @GoogleAIStudio and the Gemini API.β
βCounterpoint: https://www.niemanlab.org/2026/04/do-links-hurt-news-publishers-on-twitter-our-analysis-suggests-yes/β
βEnterprise: Rolling out in preview on Vertex AIβ
βI would buy Groβ
βThe three options right now for you and it's different in Europe because I know there's limitations but that's Claude 3 Opus that gp4 which you can also get access to for free in a limited form througβ¦β
βAnd we can put those together in a hybrid simulation engine that we open source called Robocassa. And it is an engine that leverages LOM's diffusion models text to 3D to procedurally generate infiniteβ¦β
βThere was an algorithm that was theorized by a guy named Peter Shor. I think I talked about this on a prior episode in 1994 called Shor's algorithm. Today, it's kind of the commonly well-known model fβ¦β
βIf you want to understand how to build LLM applications, I'd strongly recommend this course. It's a great course if you want to understand how to use LLMs, adapt them, fine tune them, really worth takβ¦β
βIt is a rewrite of [minGPT](https://github.com/karpathy/minGPT) that prioritizes teeth over education.β
β- [numpy](https://numpy.org/install/) <3β
β- `datasets` for huggingface datasets <3 (if you want to download + preprocess OpenWebText)β
β- `tiktoken` for OpenAI's fast BPE code <3β
β- `wandb` for optional logging <3β
β- `tqdm` for progress bars <3β
βI recommend getting the bleeding edge PyTorch nightly ([select it here](https://pytorch.org/get-started/locally/) when installing) as it is currently quite likely to make your code more efficient.β
βThis downloads and tokenizes the [OpenWebText](https://huggingface.co/datasets/openwebtext) dataset.β
βI think the best prediction for where the world is headed and this is not a endorsement or necessarily like this is where I think the world's headed because I think part of it is \nwill be slightly inβ¦β
βToday, it's kind of the commonly well-known model for how you could do this, and it's a You can watch a YouTube video on it. There's some YouTube videos that explain it pretty clearly. Takes a bit of β¦β
βThis book that I've been pawing through by the way, The Magic of M.C. Escher, is something that I got on a delightful visit to the Escher Museum in The Hague, which, if you're ever in the Netherlands,β¦β
βOne thing is maybe kind of random, but like I get really fired up to see like mad science experiments like the uh Deepseek OCR that came out the other day. Did you Did you see it? It's It's wild whereβ¦β
βwe've been publishing at the general of laward we've been P publishing prompts that turn the AI into a tutor into a mentor that uh into a student you have to explain stuff toβ
βIf you haven't had a chance to check out the Gemma models, I highly encourage you to do that.β
βSpecifically, the [GPT video](https://www.youtube.com/watch?v=kCc8FmEb1nY) is popular if you have some prior language modeling context.β
βAnd then, a couple years ago, I think in 2023, there was another computer scientist named Oded Regev from NYU who published another paper that showed a faster different approach to Shor's algorithm.β
βThis book that I've been pawing through by the way, The Magic of M.C. Escher, is something that I got on a delightful visit to the Escher Museum in The Hague, which, if you're ever in the Netherlands,β¦β
βif it helps we've at the J of we have a bunch of free YouTube videos on teaching there's a free corsera courseβ
βWe're doing Hard Fork Live again! ... June 10, SF, tickets just went live: https://www.nytimes.com/events/hardforkliveβ
βRequires the latest iOS update from the App Store.β
βI saw another thing on hacker news the other day where um you know uh text diffusion \nuh where someone made a text diffusion model by instead of doing go saying dnoising he would take like a single Bβ¦β
βAlso, if you want to see a fun application of this formula for high-dimensional sphere volumes, you might enjoy a video I did with Numberphile a couple of years back that includes a puzzle that incorpβ¦β
βEveryone: Rolling out to @Google Vidsβ
βRead more: https://www.perplexity.ai/hub/blog/plaid-integration-provides-full-view-of-personal-financesβ
βRegister: https://www.perplexity.ai/computer/a/the-billion-dollar-build-ZWzIFW.FTaKdLtufMa0yhwβ
βMistral Large v2 is now compatible with `mistral-finetune`!β
βthe open zone map that we can put on screen. The open zone map shows hundreds and hundreds hundreds of special economic zones globally, right?β
βThe code and model are open-source and accessible via Spinal Cord Toolbox v7.0.β
βWe also introduce a lifelong learning framework to automatically monitor the morphometric drift as the model is updated using additional datasets.β
βHere we present RareCollab, an agentic diagnostic framework that pairs a stable quantitative Diagnostic Engine with Large Language Model (LLM)-based specialist modules that produce high-resolution, inβ¦β
βThis paper introduces KunLunBaizeRAG, a reinforcement learning-driven reasoning framework designed to enhance the reasoning capabilities of large language models (LLMs) in complex multi-hop question-aβ¦β
βClaude Opus 4.7 is now the default orchestration model powering Computer.β
βIt's also available for Max subscribers on Perplexity web, iOS, and Android.β
βUsing glm-5 as a daily driver for a lot of thingsβ
βI want to teach uh this this semester only with um perplexity like biology especially right but that got changed entirely where All both the teaching assistants and the students were in one interface β¦β
βI first got to know his work through his blog which can be found at one useful thing. orgβ
βI think a good place to start will be Michelson-Morley and how special relativity is discovered, if it's different from the story that you get off of YouTube videos. I will prompt you that way, and thβ¦β
βLast week I shared our recent arXiv paper detailing the worldβs largest entangled state, with 120 qubits at 56% fidelity with a shot-retention rate of 28%, run on Heron R2 ibm_aachen.β
βSo go ahead and get clone.β
βAnd you can also use LMS to author the XML files and then you combine them into this exponential number of procedurally generated rooms and putting all of them together. This is a project, an open souβ¦β
βhe was getting a lot of value out of the chatbt desktop integration with his terminal and that it was a very simple thing. He was just like, if there's an error in my terminal, I just ask chatbt like β¦β
βSubtle is the Lord. Also from Imre Lakatos, The Methodology of Scientific Research Programmes.β
βAnd I couldn't do it without uh Back Blaze and Render, two of our great sponsors. And I I'll thank them a bunch of times, but uh and also Jetro, who's helping me uh launch Foundry University in Japan,β¦β
βbut like the u the gro 2.5 um open source model is actually very good. Um, and I think we'd probably be and and we'll continue to open source our modelsβ
βIn fact, we did the math and published it published the math, but nobody looked at it. Uh um it's on the Tesla websiteβ
βAnd I couldn't do it without uh Back Blaze and Render, two of our great sponsors. And I I'll thank them a bunch of times, but uh and also Jetro, who's helping me uh launch Foundry University in Japan,β¦β
βWatch todayβs live pitches from 9β10:30 AM PST: https://pplx.ai/pitch/finalsβ
β# Cambrian-S: Towards Spatial Supersensing in Videoβ
βThe Institute for Quantum Information (and Matter) @Caltech is celebrating its 25th anniversary. It has been a great run so far, and quantum information science is more fun now than ever!β
βThat's Gbrain, the Y Combinator CEO's personal AI knowledge brain, now fully open-sourced for everyone to use and build on. GitHub link in the description.β
βthe best tool out there in tutoring is Khan Academy's kigo um which while flawed is still the best approach out there right now to doing tutoring with AI and like anyone can subscribe for 20 bucks a mβ¦β
βThis is not in the technical report but this is from the paper actually that kind of originates our tokenization approach. \n Paper name is called one tokenizer to rule them all.β
βin terms of the training data for the tokenizer we use fine web 2 which you can access \n publicly in hugging faceβ
βThe corresponding paper is actually the art of asking you can find in in the archive that actually has research on this particular focus.β
βWe use their command a translate which is state-of-the-art translation model from coher that works in 23 languagesβ
βAll of them in command line translate but it includes rulebased \n filtering again difficulty filtering which come up again this difficulties as an important metric and also we looked at some some of β¦β
βAnd in fact, there's this great book called um >> gosh, it was like three New Deals. You might have you seen this one? Okay, you'll like this one.β
βFor for the merging recipe. We have another paper here a simmer merge that is actually not using only one merging technique \n but kind of select the merging techniques based on some of the metrics.β
βthen Quen is the model from Alibabaβ
β# Seemingly Conscious AI is Coming _Blog post by Mustafa Suleyman_β
βBut there's actually an alternative phrasing and this is uh I think John Stokes's coinage or others. Go broke, go woke.β
βToday we're releasing Personal Computer.β
βBeyond Annual Planning: Adapting to AI with the Six-Quarter Walk https://twitter.com/i/broadcasts/1ynKOloBqOEGRβ
βNamex is a simple utility to separate the implementation of your Python package and its public API.β
βMake sure your codebase is correctly structured. Your file structure should look like this (here we use the `keras_tuner` package as an example)β
βThe Messy Middle of AI Culture Change https://twitter.com/i/broadcasts/1BdxYqdeZoLxXβ
βAnd then our third product is obviously the full scale error corrected quantum computer. That's the big game for us.β
βThis paper is concerned with Quantum Picturalism, a novel visual mathematical language for quantum physics. Originally developed over two decades ago to explore the foundational structure of quantum tβ¦β
βThis brings us directly to a project called Superpowers. Right? It was created by a very clever developer who goes by Obra. Superpowers adds real software engineering discipline directly to claude codβ¦β
βThis brings us to a project called Hermes Agent. It was built by the brilliant team over at Noose Research. Hermes is a completely self-improving AI agent framework.β
βIf you're vibe coding anything with a UI, install this first.β
βJust use deepagents and the memory built in there!β
βIt's a collection of 20 plus skills that enforce a strict development methodology. Test-driven development, debugging frameworks, a full plan-to-execute pipeline.β
βWe present Legilimens, a continuous learning system for the mobile edge's System-on-Chip GPUs.β
βOne is in the machine learning space and that is to use essentially analog systems where we basically take classical data through what we call a quantum feature generator. It allows us to take that daβ¦β
βOur second product is in the analog simulation space where we're literally looking at how to mimic certain molecules. This is an area that we think will explode. It's early days for analog simulation,β¦β
β_ArXiv paper co-authored by Andrew Ng_β
βI am excited to share that the mathematics for machine learning and data science specialization from deep learning AI is available.β
βI know you have too many suggestions, but consider this anyway. Clearly in your territory.β
βIf you're building anything that involves iterating on results, study this repo. The pattern is actually probably more valuable than the tool itself.β
βIf you install one MCP server from this entire video, make sure it is this one.β
βVitalik already quoted some of the wonderful stuff that he released in his article in uh, April this year. And I want to focus on that, the importance of privacy.β
βHi listeners and welcome back to No Priors.β
βI really followed Charlamagne as the blueprint. Like I looked at Charlamagne, what he did with Wendy.β
βYou feed it your PRD, and it breaks it down into structured tasks with dependencies, complexity scores, and subtasks. Then Claude executes them one by one in order.β
βThe Urgency of Interpretability: Why it's crucial that we understand how AI models workβ
βMade by Microsoft, this gives Claude the ability to control a browser, click buttons, fill forms, navigate pages, take screenshots. Your AI agent can now use the web like a human.β