So much this...
Read Atlas Shrugged so you can have burned into your mind "show don't tell" as Rand delivers unreadable multi-page speeches of her philosophy over and over.
Replying to @RyanHoliday
Why Atlas Shrugged? Because the entire premise of the book, “You guys don’t appreciate me so I’m taking my ball and going home,” is an exercise in ego and petulance.
68
Brook Miller retweeted
Wrote up about my personal journey from AI skeptic to someone who finds a lot of value in it daily. My goal is to share a more measured approach to finding value in AI rather than the typical overly dramatic, hyped bait out there. mitchellh.com/writing/my-ai-…
Brook Miller retweeted
💫 Agentic Coding Trends Report by @AnthropicAI
→ AI compresses SDLC into shorter loops
→ Multi-agent systems replace single-agent flows
→ Agents execute multi-day build cycles
→ Human review focuses on edge cases... 👇
resources.anthropic.com/hubf…
Brook Miller retweeted
I try to not talk about politics. I generally believe the best way I can serve the world is as a non-partisan expert, and my genuine beliefs are quite moderate. So the bar is very high for me to comment.
But recent events – a federal agent killing an ICU nurse for seemingly no reason and with no provocation – shock the conscience.
My deep loyalty is to the principles of classical liberal democracy: freedom of speech, the rule of law, the dignity of the human person. I immigrated to the United States – and eventually cofounded Anthropic here – believing it was a pillar of these principles.
I feel very sad today.
after working with claude all morning on a detailed plan...
can you run each of the steps in the plan as a sub agent where each gets a a fork in the conversation from this point while you monitor and don't pollute your context with the details
thanks @trq212 for your great tweets on getting the most from claude
1
72
"Whether you are a scientist or a policy professional, a linguist or a diplomat, a literary critic or a musician, or just a curious person, I am confident these tools [coding agents] have something to offer you."
- @deanwball on Agent Coding Tools
Around Q3 last year using the agent coding tools became a fundamentally different experience. Instead of asking the tool to implement a particular feature / function in an opinionated way, I'd ask "how should we..."
Claude code would propose a plan, I'd tweak the plan, Claude would work for 15-30 minutes and then on to the next idea / task. Less writing code and more like collaboration.
I think these tools will change how a lot of work is done but I've struggled to explain why to people who aren't familiar with coding.
Dean's article does a good job particularly on the implications.
If you get to a point in the middle where you're thinking about what to order for lunch, skip to the implications section at the end. 🙂
hyperdimensional.co/p/among-…
today: the basics of coding agents for a non-technical audience, and 13 conjectures on what they mean for business, policy, and culture.
76
This is a great summary of the value people are finding in agent based tools. (Claude Code, Codex, etc...)
I'm not sure they're still particularly approachable to non-programmers but the juice is worth the squeeze.
A part of the workflow he doesn't discuss is using git for version control (to track what the agent changed and go back when it messes up).
Check it out: hyperdimensional.co/p/among-…
today: the basics of coding agents for a non-technical audience, and 13 conjectures on what they mean for business, policy, and culture.
74
Brook Miller retweeted
Replying to @ekuyda
This is for you mom, @ekuyda (and for everyone ready to run a local agent without living in the matrix): everything.intellectronica.n…
I need @matt_levine to write about how $TSLA gets killed when Space X lists.
Suddenly there are 2 ways for retail to invest in Elon's projects so they compare the DCF of each and arbitrage to both of them trading at 10% of current market cap.
70
Brook Miller retweeted
Everyone's obsessing over hypothetical future AI. Here's a thought experiment: what if progress stopped today?
The economic impact of current capabilities — GPT-5, Gemini 3, Claude 4.5 — continuing to be used in inventive ways? Enormous. Already transformative.
"There is no AI shaped hole in most big companies."
@kevinroose made such a great point on today's Hard Fork episode.
I've been working with a few companies on Gen AI strategy / plans and fundamentally big companies are really well run.
The obvious problems are being solved (that's why they got to be big companies).
nytimes.com/2026/01/02/podca…
24
I updated my data cleaner claude skill to use uv to run the scripts.
Increases portability especially if you're using claude code across multiple environments or want to share skills through github.
Thanks @intellectronica for the great tip.
PS: The newsletter is very high signal and recommend you subscribe if you're trying to improve your AI coding usage.
elite-ai-assisted-coding.dev…
If you're curious to try the skill: github.com/brook-miller/clau…
1
1
53
Bread and circuses
Chamath: We need to follow the example of Industrialist leaders like Carnegie & Ford to fix AI's massive perception problem.
@chamath "Enough of the stupid haircuts, dumb watches, ugly clothes, ostentatious displays of wealth.
We've all done it. I've been guilty of it.
It has to stop. Absolutely stop.
And instead, we need to start to use a percentage of the balance sheets of these companies in order to benefit as many Americans as possible. That is the absolute minimum."
1
60
Built my first @claudeai Skill: data-cleaner!
Using Claude code to create analysis notebooks is a super power.
A lot of the notebooks spend time cleaning data, normalizing, fixing values etc... so I prefer to do the cleaning in a separate file.
Claude code can do that too but it ends up writing a bunch .py files to check values etc.... with Claude data cleaner skill I created a set of re-useable functions that Claude uses to analyze and create a single data cleaning function.
Example prompt:
Claude use the data-cleaner skill to write a clean-up.py for this path/to/csv. Run it and save the final output to: path/to/cleaned.csv.
You may have to update the skill.md for your path to python but the skill let's Claude clean-up data for analysis without touching a bunch of boilerplate.
github.com/brook-miller/clau…
1
1
71
Brook Miller retweeted
My pleasure to come on Dwarkesh last week, I thought the questions and conversation were really good.
I re-watched the pod just now too. First of all, yes I know, and I'm sorry that I speak so fast :). It's to my detriment because sometimes my speaking thread out-executes my thinking thread, so I think I botched a few explanations due to that, and sometimes I was also nervous that I'm going too much on a tangent or too deep into something relatively spurious. Anyway, a few notes/pointers:
AGI timelines. My comments on AGI timelines looks to be the most trending part of the early response. This is the "decade of agents" is a reference to this earlier tweet nitter.cf/karpathy/status/188254… Basically my AI timelines are about 5-10X pessimistic w.r.t. what you'll find in your neighborhood SF AI house party or on your twitter timeline, but still quite optimistic w.r.t. a rising tide of AI deniers and skeptics. The apparent conflict is not: imo we simultaneously 1) saw a huge amount of progress in recent years with LLMs while 2) there is still a lot of work remaining (grunt work, integration work, sensors and actuators to the physical world, societal work, safety and security work (jailbreaks, poisoning, etc.)) and also research to get done before we have an entity that you'd prefer to hire over a person for an arbitrary job in the world. I think that overall, 10 years should otherwise be a very bullish timeline for AGI, it's only in contrast to present hype that it doesn't feel that way.
Animals vs Ghosts. My earlier writeup on Sutton's podcast nitter.cf/karpathy/status/197343… . I am suspicious that there is a single simple algorithm you can let loose on the world and it learns everything from scratch. If someone builds such a thing, I will be wrong and it will be the most incredible breakthrough in AI. In my mind, animals are not an example of this at all - they are prepackaged with a ton of intelligence by evolution and the learning they do is quite minimal overall (example: Zebra at birth). Putting our engineering hats on, we're not going to redo evolution. But with LLMs we have stumbled by an alternative approach to "prepackage" a ton of intelligence in a neural network - not by evolution, but by predicting the next token over the internet. This approach leads to a different kind of entity in the intelligence space. Distinct from animals, more like ghosts or spirits. But we can (and should) make them more animal like over time and in some ways that's what a lot of frontier work is about.
On RL. I've critiqued RL a few times already, e.g. nitter.cf/karpathy/status/194443… . First, you're "sucking supervision through a straw", so I think the signal/flop is very bad. RL is also very noisy because a completion might have lots of errors that might get encourages (if you happen to stumble to the right answer), and conversely brilliant insight tokens that might get discouraged (if you happen to screw up later). Process supervision and LLM judges have issues too. I think we'll see alternative learning paradigms. I am long "agentic interaction" but short "reinforcement learning" nitter.cf/karpathy/status/196080…. I've seen a number of papers pop up recently that are imo barking up the right tree along the lines of what I called "system prompt learning" nitter.cf/karpathy/status/192136… , but I think there is also a gap between ideas on arxiv and actual, at scale implementation at an LLM frontier lab that works in a general way. I am overall quite optimistic that we'll see good progress on this dimension of remaining work quite soon, and e.g. I'd even say ChatGPT memory and so on are primordial deployed examples of new learning paradigms.
Cognitive core. My earlier post on "cognitive core": nitter.cf/karpathy/status/193862… , the idea of stripping down LLMs, of making it harder for them to memorize, or actively stripping away their memory, to make them better at generalization. Otherwise they lean too hard on what they've memorized. Humans can't memorize so easily, which now looks more like a feature than a bug by contrast. Maybe the inability to memorize is a kind of regularization. Also my post from a while back on how the trend in model size is "backwards" and why "the models have to first get larger before they can get smaller" nitter.cf/karpathy/status/181403…
Time travel to Yann LeCun 1989. This is the post that I did a very hasty/bad job of describing on the pod: nitter.cf/karpathy/status/150339… . Basically - how much could you improve Yann LeCun's results with the knowledge of 33 years of algorithmic progress? How constrained were the results by each of algorithms, data, and compute? Case study there of.
nanochat. My end-to-end implementation of the ChatGPT training/inference pipeline (the bare essentials) nitter.cf/karpathy/status/197775…
On LLM agents. My critique of the industry is more in overshooting the tooling w.r.t. present capability. I live in what I view as an intermediate world where I want to collaborate with LLMs and where our pros/cons are matched up. The industry lives in a future where fully autonomous entities collaborate in parallel to write all the code and humans are useless. For example, I don't want an Agent that goes off for 20 minutes and comes back with 1,000 lines of code. I certainly don't feel ready to supervise a team of 10 of them. I'd like to go in chunks that I can keep in my head, where an LLM explains the code that it is writing. I'd like it to prove to me that what it did is correct, I want it to pull the API docs and show me that it used things correctly. I want it to make fewer assumptions and ask/collaborate with me when not sure about something. I want to learn along the way and become better as a programmer, not just get served mountains of code that I'm told works. I just think the tools should be more realistic w.r.t. their capability and how they fit into the industry today, and I fear that if this isn't done well we might end up with mountains of slop accumulating across software, and an increase in vulnerabilities, security breaches and etc. nitter.cf/karpathy/status/191558…
Job automation. How the radiologists are doing great nitter.cf/karpathy/status/197122… and what jobs are more susceptible to automation and why.
Physics. Children should learn physics in early education not because they go on to do physics, but because it is the subject that best boots up a brain. Physicists are the intellectual embryonic stem cell nitter.cf/karpathy/status/192969… I have a longer post that has been half-written in my drafts for ~year, which I hope to finish soon.
Thanks again Dwarkesh for having me over!
The @karpathy interview
0:00:00 – AGI is still a decade away
0:30:33 – LLM cognitive deficits
0:40:53 – RL is terrible
0:50:26 – How do humans learn?
1:07:13 – AGI will blend into 2% GDP growth
1:18:24 – ASI
1:33:38 – Evolution of intelligence & culture
1:43:43 - Why self driving took so long
1:57:08 - Future of education
Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!
Brook Miller retweeted
Two things can be true at the same time:
1. Without additional advances, LLMs won't get us to general intelligence.
2. Even without additional advances, LLMs will radically transform the economy.
Brook Miller retweeted
There’s a lot of great takes in this article from @GergelyOrosz detailing how Anthropic built Claude Code. One thing that amazes me the most is that 90% of Claude Code was written by AI and Claude Code itself. Even more impressive, each engineer on the team deploys to production at least five times a day.
open.substack.com/pub/pragma…

















