Independent scholar. New book: The Independent Scholar (2026), a history from Plato to Nietzsche.
Austin, TX
- Tweets 31,308
- Following 634
- Followers 54,394
- Likes 26,139
I think we're incredibly early in understanding the differences between code and prose.
I'll give you one specific technical example that illustrates why I think AI tooling for prose has to look very different than what we're seeing for code.
Since coding compiles in a binary fashion (it works, or it doesn't), this means you can just throw at it as much context and tooling as you can. If the code compiles, then it's fine.
Prose does not compile in binary fashion (the ideal prose result is not only non-binary, but multiple, because it is unique to the author). This means you precisely do NOT want to throw more context and tools into the mix, other things equal, because compute can't reduce that toward any verifiable ideal point, so it only wastes tokens (time and money) while usually pushing away from the ideal end-state you are personally, creatively trying to arrive at.
(This is why I think it deserves dedicated commercial attention and is not likely to be solved by the big labs to the degree of precision that I want.)
5
1
22
2,392
Today we're launching Precision Copyeditor, the only publication-grade AI copyeditor that learns your house style.
ChatGPT and Claude can't do publication-grade copyediting for 4 reasons:
1) Most people don't need publication-grade precision, so popular tools never apply enough compute to this
2) They "learn" your preferences but in a generic way across all your various tasks, which you can't dial in with precision
3) So they always miss some errors and
4) They always change things you do not consider errors
We solve this by:
1) Applying more compute than most people want or need and
2) Learning and updating your House Style every time you run it
It's also the only professional copyediting tool that offers a self-serve API key you can give to your agents in 5 minutes!
You can try it without an account. It's free to use with daily limits but requires signup to build your preferences.
4
45
4,523
You're a smart guy, get the bag, build the future, but don't pretend you're also a brave, noble truth teller!
It's literally the one thing you give up by taking payment from a company to strategically shape the future on their behalf lol.
You can just enjoy money & influence!
I want to explain something to you:
People often quote tweet me and say "can you believe that an OPENAI EXECUTIVE is so dumb that he thinks tweeting like THIS is the way to fix AI's image problem?!?!?!"
I was not hired by OpenAI to "fix AI's image problem," and I do not get out of bed every day to make you feel good about tokens. Yes, AI has enriched my life tremendously and I believe it will enrich most people's lives, and I speak of this often. But I believe a truly transformative thing is underway, and that some aspects of it will be incredible and others will be extraordinarily fraught. I want to tell you the truth about that transformation as I see it happening, to the best of my ability.
The truth as I see it will not always be optimal marketing for the AI industry, nor will it always necessarily arrest the public's concerns. The public, very often, is right to be concerned. I am concerned. Most people I know who work at true frontier labs are concerned. Sometimes, I will feel it is my duty to describe to you changes I believe may be afoot in our world that will not be fun to think about or experience, and that cause me sadness and fear in my private life.
I'm not here to sing you some lullaby about what is happening, and I wish people would stop trying to RLHF me into doing so. It is out of respect for those who read my words that I try to tell the truth. I've never seen so many people beg to be lied to and disrespected, but I wish it would stop.
5
1
64
7,538
Today I learned that it was Nietzsche who invented the Reddit meme of "going NC" (No Contact) with family members, driven by narcissistic grandiosity.
"My sister... I am compelled to part company with you forever... You have not the remotest conception of what it means to be most closely related to the man and to the destiny in whom the question of millennia has been decided."
13
2
76
8,917
The thing about AI and writing is that it will have nothing to do with 'writing,' in the end, but also it will dramatically transform virtually everything *around* 'writing.' And this is just going to obliterate minds and egos and public discourse in bewildering ways for some time.
AI changes dramatically what it is possible to ask, how one can do research, edit, publish, etc.—while strangely, somehow, not actually affecting the core task and role of the human thinker/writer. So every simplistic take that plays best on social media will essentially doom you to misunderstand what's happening, and what to do about it. I'm pretty sure I've crossed into thousands of hours with every model, agent, and harness over the past two years, including countless custom tools, and this is what all of my conviction is amounting to as I try to solidify my strategic viewpoint looking out at the next 10 years of public intellectual life: That a very low percent of serious writers are sufficiently excited and joyful about AI in all of the places where it should be used, and a very high percent of the biggest bulls are fundamentally confused about where it will really matter for knowledge work (outside of verifiable domains).
The situation could not be more favorable for passionate and independent thinkers with an open, honest, experimental attitude toward truth-seeking as a vocation. A lot of the alpha will be private and personal for a long time still, I suspect.
7
3
66
5,362
AI slop detection is only a proxy for what people really want: to know if something is worth reading.
So we're launching a new measure of writing quality in general.
Slop detection is one element, but we also account for the other things that distinguish excellent writing.
Announcing Deft QA, a way to measure writing quality
The goal is to help readers decide if something is worth reading, and help writers improve their craft.
We measure quality according several content and style metrics:
Originality, Clarify, Coherence, Information Density, Correctness, and Humanness
The text gets a score for each dimension, then we combine the scores into a final overall score.
"Writing Quality" is vaguely defined and subjective. This isn't meant to be a source of truth for what is good writing and what isn't, it's just a tool that that is supposed to correlate well with your natural sense of quality. It's primarily designed for long form text (e.g. this tweet scores quite low)
Link in next tweet
2
30
3,045
I’m much more sympathetic to the homeless than a lot of my friends. But you cannot be homeless and also have a pitbull. No way man. Should be a federal offense.
3
38
2,433
No this is silly, academic journals only need fine-tuned and cheap-to-serve models for triaging submissions. Peer review proceeds normally but only after the initial model gate.
It's worth being explicit about this.
AI written submissions will break peer review as it currently exists. They also undermine one of the key functions of academic publishing, ie to identify talented researchers.
So we have to create some kind of bright line distinction that will avert those outcomes. If we have to engage in detailed adjudication, for each paper, of whether the amount of human contribution is acceptable, then that will quickly become untenably burdensome (we found this in our Neurips position paper track approach, which tried to give authors the ability to demonstrate provenance, but required almost as many person hours per paper, if not more, as the review itself).
So, using AI for writing might help *your* creative process. But it creates a significant negative externality, which is the cost of being able to distinguish between your putatively acceptable use of AI, and someone else's unacceptable use.
If we err towards inclusion, then that helps the "good" users of AI, but makes peer review unsustainable and makes it impossible to identify talented researchers (where talent ≠ being good at prompting). If we err towards exclusion, then the only downside is that some people don't get to use AI to write their final papers. That isn't a very significant cost, in my view; after all it's how things were up until about 9 months ago, when AI became good enough to play this kind of role.
2
5
2,887
Today we’re going public with our custom model (and small research lab) focused on the use of LLMs for professional writing/editing/publishing.
86% of uses come out 100% human on Pangram.
We currently have 24 paying customers and intend to become profitable without investors.
Announcing Deft, a new AI lab for better writing, cofounded with @jmrphy
See the picture for launch announcement the Deft model wrote for itself
Currently, 86% of user queries are fully human according to pangram.
This is still a small beta model and it might make mistakes. We are launching our public beta now to get more feedback before scaling up.
Tips for better performance:
- Add more details to your prompt. If you just provide a short sentence prompt, it will likely get detected as AI.
- Try changing the style in "advanced options"
- Deft currently works better for some use cases like Analysis/Essays, Creative writing, and Rewrites. It works less well for Marketing copy and news articles.
Our main goal is better writing, fooling AI detectors is just a side effect.
Link below
18
9
189
14,801
The doubling-down and valorization of Jason Arday after his suicide (which is sad emotionally but *more incriminating* logically) really shows how far gone "humane" educated opinion is today.
I really did not see this coming. They're going to make him the new George Floyd.
The entire situation is worse than I imagined, and I was one of the people saying it's worse than you can possibly imagine.
After the death of Jason Arday, Will Self asks what happens when newspapers, platforms, editors and audiences become components in a machine nobody admits to controlling.
The public interest is not the same thing as the public being interested.
26
16
277
23,690
The serious writer class should be quite happy with how LLMs are shaking out.
For a while, it was not clear. In the past few years I've been completely open-minded about whether LLM scaling might encroach upon what serious thinkers do when they produce serious texts. I've been writing a book throughout this period, so I've had an empirical and experimental relation to it all, constantly testing and judging what can and cannot reliably be done with these tools.
And despite exponential improvements over the past 3 years, their ability to do professional-level knowledge work has rather stubbornly not improved as much as we might have expected (outside of formally provable domains like math).
For serious writers in the humanities and social sciences, it's basically the ideal situation: Barring some fundamental architectural advancement, even with continued scaling it seems like LLMs may very well plateau as superhuman research assistants, without really challenging the core work and value-add of the human. As Cremieux points out, even with the latest models that are solving frontier math, you still need quite a lot of judgment to even use them for research.
Few seem to realize how surprising and fortunate this is for humans who make a living from knowledge work!
My latest article is all about how you *cannot* just ask LLMs arbitrary factual questions and hope to get consistently correct answers.
Link: cremieux.xyz/p/how-do-you-as…
15
18
161
25,011
The fever dreams of boomer white leftist female university administrators may turn out to be more lethal than we ever imagined, as they begin to intersect with AI-era truth equilibration. How many thousands will die from the unbearably efficient rectification of names?
Jason Arday, the former University of Cambridge professor, has been found dead.
Mr Arday was found “unresponsive” at an address in Battersea, London, on Friday afternoon and later pronounced dead at the scene.
🔗 telegraph.co.uk/news/2026/08…
19
12
101
12,655
What’s interesting here is that it’s a website, and it’s not double-blind peer-reviewed, so they could just push updates today if they wanted.
But then it would all come crashing down. This is an epistemological “let them eat cake” moment in the AI revolution.
the SEP has a new article on Capitalism, which has some serious errors!
with the help of Refine, let's work our way through some of them
1/
1
1
5
2,668
If you've been sitting out the AI agent hype cycles in the past year because it always seemed like a fad, a time suck, and too expensive to be worth it—honestly you may have been correct.
But I would say DeepSeek V4 Flash with Hermes Desktop just crossed a line where probably everybody sitting out should probably do this now. It's almost free, the AI is basically what the US companies are selling for $200/mo, no setup, and you can get the AI to do basically anything on your computer for you in like 10 minutes all-in.
(not an AI influencer and have zero financial involvement with any of these things).
The extremely high intelligence has now converged with such extremely low cost that it's bonkers someone would not take 10 minutes out of their day to have this on their computer.
All you do is download Hermes Desktop (free), then create an account on OpenRouter (just buy $5 of credits with no auto-topup, and copy your API key). Then put that API key in your Hermes settings, select DeepSeek v4 Flash as your default model, and everything else from there you can just ask Hermes to explain and do for you.
Then just go crazy trying to doing useful stuff with/on your computer, in your browser tabs, whatever. You'll be surprised how long it takes until your $5 runs out (and if you're not impressed by then, very little time or money has been lost).
31
20
399
58,142
Obviously inevitable but it’s so interesting to watch who is signaling in which directions…
A philosophy journal just published an article written mostly by Claude. The philosopher who "supervised" Claude & who is listed as the author explains how he did it. The journal's editor describes the publication of the piece as an "experiment" dailynous.com/2026/08/13/phi…
1
1
2,558
There's a very interesting rift emerging around these Atelier Missor guys. From what I can tell, the hip crowd in tech has decided they dislike this project, despite this crowd's constant calls—over many years now—for new and ambitious art projects. Having no horse in this race, my first impression is to be impressed by this project—at the very least for their commitment. Some call it "larping" but the majority of great figures and movements in history were "larping" something or someone from the past; all that matters is if you do it with sufficient sincerity and intensity and persistence and skill.
I can't help but feel that the cool kids hating on this group are just mad that someone is actually doing what they've been calling for all these years. If you make a name out of calling for something, you tend to hate any group that really goes for it—because you have not, although you could have at any time, evidently.
I don't have strong aesthetic opinions about sculpture but I'm impressed by what seems to be this group's sincerity, conviction, commitment, and extraordinary success in capturing American mindshare. Am I missing something? Are there good reasons for everyone to be naysaying this group's project?
100% epic. Prometheus rising at Starbase feels like the emergence of a new American aesthetic: classical, monumental, technologically optimistic and entirely unafraid of ambition
187
55
2,205
132,903