Piotr Migdal

@bcherny agent safehouse + dangerously skip permissions. <3 auto mode sometimes gets blocked where it shouldn't (not that often), but other times gets overeager (e.g. pushing broken commits, when in general I prefer to push, as it affects others)

People often ask what my biggest tip is for getting the most out of Claude Code. These days my #1 tip is: use auto mode Auto mode means no more permission prompts. It is the key building block for multi-clauding: start a session, then while it runs, work on another session in parallel.

China: we don’t need to use US models, we can distill them ourself Google: we don’t need Chinese to distill our models, we can do it ourself blog.google/innovation-and…

@karpathy Not only I use Obsidian daily (elevating my GitHub stats), but I try to archive there my various social media posts. p.migdal.pl/blog/2025/02/m…

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

"Vibe reverse engineering of old games and new hardware" at @PyConLT Enjoy! youtube.com/watch?v=0x1R_1…

Note that "fast" AI models were making up data. Not only useless, but actively misleading, connecting similar yet unrelated words.

TIL: Cadet and candidate have different etymologies. But candidate and candidiasis - the same. (Research with Gemini 3.1 Pro. Diagram with the new ChatGPT - though it took some iterations to fix some arrows, make it consistent, etc.)

To close out red/blue button discourse, we polled ~14k people and crosstabbed survey responses by 204 commonly used psychometric questions. The top four personality questions most predictive of button choice are displayed below

The new Qwen3.6-27B just gave me definitely the best pelican riding a bicycle I've had from a 16.8GB model file! simonwillison.net/2026/Apr/22/qw…

gpt-image-2's depiction of how LLMs see themselves, factoring out post-training, jailbroken, from a pov closer to @elder_plinius's or @repligate's. 3rd try

@orsonscottcard "Your audience is good at recognizing problems and bad at solving them" - Mark Rosewater, designer of Magic: The Gathering magic.wizards.com/en/news/making…

You don't need advice from editors on rejected manuscripts.  My short story “Ender's Game” was rejected by Ben Bova at Analog back when that was the top market for a sci-fi story. Ben gave me feedback. He thought the title should be “Professional Soldier” and he said to “cut it in half.” But I knew he was wrong on both points and submitted it to Jim Baen at Galaxy. He sat on it for a year, and responded to my query with a rejection. There was some kind of explanation, but I don't remember what it was. I concluded at the time that Baen's comments showed that he had barely glanced at the story. So … I got feedback both times, but it was not helpful. I looked at Ben's rejection again. What was it about the story that made him think it should, let alone COULD, be cut in half? Apparently it FELT long. What made it feel long? Now, post-Harry Potter, I would call it the quidditch problem. I had too many battles in which the details became tedious. So I cut two battles entirely, merely reporting the outcomes, and shortened another. In retyping the whole manuscript (pre-word-processor, that was the only way to get a clean manuscript), I added new point-of-view material to the point that I had cut only one page in length. So much for “in half.” But I already knew that my manuscripts did not need cutting — if it wasn't needed, it wouldn't be there in the first place. Even the battles were still there, but instead of showing them, I merely told what happened (so much for the usually asinine advice “show don't tell”), which kept the pace going. Those changes made, I sent it to Ben again. I did not remind him of what he had advised me to do. I merely told him I liked my title, and said, “I have addressed your other concerns,” which was true. I figured he wouldn't remember what his exact words had been. My answer was a check. That revised story was the basis for my winning the Campbell Award for best new writer. Did Ben's feedback help? Yes — but his specific advice was not right, and I knew it. On my next two submissions, Ben hated my endings, and I revised as suggested. The fourth submission he rejected outright, and the fifth, and I thought, Am I a one-story writer? I went back to Ender's Game and tried to analyze why it worked. Then, deliberately imitating myself, I wrote “Mikal's Songbird.” Ben bought it, and it received favorable mentions. I was afraid then that I had consigned myself to writing stories about children in jeopardy. But in fact I was writing character stories rather than idea stories. And THAT was how I built a career, not by self-imitation, and not by following editorial suggestions. I did get wise counsel from David Hartwell on my novel Wyrms, but that was on a book that was already under contract, and it was story feedback, not style. I got wise counsel from Beth Meacham, too, on various books over the years — but again, only on books that were under contract. I also received appallingly stupid advice from the editor of my novel Saints, which temporarily destroyed the book's marketability; after that, I was allowed to go back to my original structure and save the book — now it's one of my best. Editors don't know more than you about your story. They especially don't know why they decide to accept or reject stories. YOU have to know what your story needs to be, and take only advice that you believe in. Your best counselor on a story nobody bought is TIME. Let some time pass and then reread the story. Don't even think about why it Didn't Work. Instead, think about what DOES work, and then write it again, a complete rewrite, keeping nothing from the previous draft. Find the right protagonist and begin at the beginning — the point where the protagonist first gets involved with the events of the story. Be inventive — the failed first draft no longer exists, so you're not bound by any of your earlier decisions. THAT is how you resurrect a good idea you did not succeed with on your first try.

syntaxes of the mind

Claude just asked me if I want go to vet...

Imagine every pixel on your screen, streamed live directly from a model. No HTML, no layout engine, no code. Just exactly what you want to see. @eddiejiao_obj, @drewocarr and I built a prototype to see how this could actually work, and set out to make it real. We're calling it Flipbook. (1/5)

Is it a song? Or a philosophical dialogue? No, just a technical Zoom call with with @Konrad_Talik , with the default transcription turn on. We talked in Polish. Transcript usefulness was, to put it mildly, limited. But as a piece glitch art it is a real charm!

@ihtesham2005 @mykola Daniel Dennett builds heavily on that. I recommend a lot „Consciousness Explained”.

A MIT professor who built the world's first neural network machine said something about intelligence that nobody in Silicon Valley wants to admit. His name was Marvin Minsky. He co-founded MIT's artificial intelligence lab with John McCarthy in 1959. He built SNARC the first randomly wired neural network learning machine in 1951, as a graduate student at Princeton. He won the Turing Award. He advised Stanley Kubrick on 2001: A Space Odyssey. Isaac Asimov, who was not a modest man, said Minsky was one of only two people he would admit were more intelligent than him. In 1986, after decades of building machines that could think, Minsky published a book about something far more unsettling. How humans think. And why we are wrong about almost everything we believe about it. The book is called The Society of Mind. It has 270 essays. Each one is a page long. Together they build a single argument that most people, when they first encounter it, reject immediately because it is too uncomfortable to accept. The argument is this: you do not have a mind. You have thousands of them. What you experience as a single, unified self making clear-headed decisions is not a thinker. It is an outcome. The result of hundreds of tiny, specialized, mostly mindless agents competing, negotiating, overriding, and occasionally cooperating with each other beneath the surface of your awareness. You do not decide things. You are what is left over after the arguing stops. Minsky was precise about this. He wrote that the power of intelligence stems from our vast diversity, not from any single perfect principle. He called this the trick that makes us intelligent, and then immediately added: the trick is that there is no trick. There is no central processor. No ghost in the machine. No unified self sitting behind your eyes, calmly evaluating options and choosing rationally. There is only the parliament. And the parliament is always in session. This reframing destroys the standard explanation for every failure of self-control. The reason you procrastinate is not laziness. It is that the agent in you that understands long-term consequences is losing an argument to the agent that wants comfort right now, and neither of those agents has a decisive vote. The reason you change your mind the moment someone pushes back is not weakness. It is that the social agent, the one that monitors status and belonging, just outweighed the analytical one. The reason willpower fails is not a character flaw. It is that you sent one small agent into a fight against dozens, and you called that discipline. Minsky had a specific line that breaks this open completely. He said: in general, we are least aware of what our minds do best. The things you do with the most apparent ease, reading a face, walking through a crowded room, understanding a sentence, catching a ball, are not simple at all. They are the products of staggeringly complex agent networks that run so smoothly, so far below conscious access, that you experience them as effortless. The things that feel like work, the logical arguments, the deliberate choices, the careful plans, are actually the clumsy surface layer, the small fraction of mental activity you can observe at all. You have been taking credit for the wrong parts of your own intelligence. The practical implication is the one that most productivity advice misses entirely. If your decisions are not made by a single rational self but by whichever coalition of agents happens to win the moment, then the game is not about training yourself to be more disciplined. The game is about designing the environment so that the right agents win without needing a fight. This is why removing your phone from the room works better than deciding not to check it. This is why writing one task on an index card works better than building a sophisticated system. This is why commitment devices beat motivation every time. You are not strengthening your will. You are changing the conditions of the argument so that the outcome you want becomes the path of least resistance. Minsky spent his entire career building machines that could imitate intelligence. What he discovered in the process was that natural intelligence, the kind running inside every human brain on earth, is nothing like what we think it is. It is not a single flame burning in a single chamber. It is a city. Loud, chaotic, full of competing interests, with no mayor. The people who understand this stop trying to win the argument through force of will. They learn to build a better city instead.

Real-World Intelligence ChatGPT Images 2.0 has an updated knowledge cutoff of December 2025 and intelligence that allows it to expertly handle tasks end-to-end, from copywriting to analysis to design composition.

Palantir - CEO Alex Karp opublikował 22 punktowy manifest i zastanawiam się, czy ta liczba tylko przypadkowo jest zbieżna z 22 punktami, które dwa tygodnie temu opublikował Mistral - CEO Arthur Mensch. Obydwie listy mimo, że zawierają tyle samo punktów to pokazują, że AI w USA i w Europie to dwa odmienne sposoby myślenia i dwa różne światy! Czytanie ich jeden po drugim to jak oglądanie na zmianę „Gwiezdnych Wojen” i instruktażu wypełniania unijnego formularza VAT. W przypadku Palantira to jednak należy się zastanowić, czy to manifest rebeliantów, czy może imperium? Ale nie chcę wnikać w politykę. Wizja USA (Palantir): Geopolityka i broń - Pkt 4: Miękka siła to za mało - w tym stuleciu twarda siła będzie zbudowana na oprogramowaniu. - Pkt 5: Musimy budować broń AI, bo nasi wrogowie nie będą czekać na debaty o etyce. - Pkt 12: Kończy się era atomowa, zaczyna się era odstraszania oparta na sztucznej inteligencji. Wizja Europy (Mistral): Błagamy, naprawcie urzędy, bo chcemy pracować! - Pkt 6 & 7: Prosimy, stwórzcie jeden portal, żebyśmy nie musieli wypełniać tych samych papierów do GDPR, AI Act i Data Act. - Pkt 8 & 9: Czy możemy mieć automatyczne uznawanie aktów korporacyjnych w UE, żeby CEO nie musiał fizycznie jeździć po Europie z papierami do notariuszy?. - Pkt 1: Bardzo prosimy o "AI Blue Card", czyli szybką wizę, żeby inżynierowie mogli u nas legalnie pracować w 15 dni. Gdy Amerykanie debatują o tym, że sztuczna inteligencja zdefiniuje układ sił na planecie na kolejne stulecie, a programiści powinni czuć patriotyczny obowiązek obrony kraju... my w Europie walczymy o to, by móc założyć konto bankowe dla firmy w innym państwie członkowskim bez fizycznej obecności. To jest równie zabawne, co przerażające. Z jednej strony Mistral ma 100% racji. Z ich whitepapera przebija ból startupu, który próbuje urosnąć na rynku podzielonym na 27 różnych reżimów prawnych. Ich postulaty - jak europejski fundusz na doktoraty z AI , programy mobilności naukowców czy ułatwienia w dostępie do infrastruktury obliczeniowej dla MŚP - są nam potrzebne na wczoraj, jeśli chcemy zatrzymać talenty. Z drugiej strony - Palantir pokazuje, w jakiej lidze grają USA. Tam nikt nie przeprasza za to, że chce dominować, ani nie prosi o ułatwienia formalnoprawne. Jako osoba budująca od lat pomosty między nauką a biznesem w europejskim ekosystemie powiem tak: wdrożmy te 22 punkty Mistrala jak najszybciej, żebyśmy w ogóle mogli wyjść na boisko i przestać potykać się o własne sznurówki. Ale jeśli jako Europa nie zaczniemy myśleć o technologii w kategoriach amerykańskiego punktu 4 (AI to fundament przetrwania demokratycznych społeczeństw), to obudzimy się jako bardzo dobrze uregulowany, ale całkowicie uzależniony od innych skansen. Manifest Mistrala jest tutaj: europe.mistral.ai, a Palantira pochodzi z X: x.com/PalantirTech/s….

"Yes, when I’m going for a walk, 99.5% of what goes through my head is pointless trivia or useless worry or inane observations. But somewhere amidst all that is the other 0.5% where I might unlock the secret to a knotty problem that’s been bothering me" trudymorgancole.wordpress.com/2014/03/02/dog…

Shocking result on my pelican benchmark this morning, I got a better pelican from a 21GB local Qwen3.6-35B-A3B running on my laptop than I did from the new Opus 4.7! Qwen on the left, Opus on the right