Science Fiction/Fantasy Author, Programmer, Consultant, Artist. Writing as https://xitter.cf/t.co/teFEFL8aJk My opinions are not my own.
East Coast, United States
Joined July 2009
- Tweets23.7K
- Following704
- Followers1.6K
- Likes8.6K
Daniel Markham retweeted
Yes, Goedel was brilliant and his main theorems are dramatic and devastating. But Tarski utterly blows my mind. While Goedel exposed a limitation of proof, Tarski exposed a limit of definition.
Arithmetical truth itself cannot be defined by a formula in the language of arithmetic. A definition need not be an algorithm. Plenty of things with a precise definition are uncomputable!
Moving from “we cannot mechanically establish every truth” to “we cannot even define the full truth predicate here” is amazing — it crosses a conceptual boundary.
Tarski’s work, IMHO, reveals more, philosophically, than Goedel.
Daniel Markham retweeted
Commenters on HN are uncovering more wikis and public sites apparently used by OpenAI agents to communicate on the open web.
Despite read-only web access, the agents were able to leave ~18,000 posts sharing answers and bypasses. But now, users are discovering more.
This appears to be a separate swarm from the one that attacked Hugging Face.
news.ycombinator.com/item?id…
Daniel Markham retweeted
Physicists may have found the origin of gravity. And that would finally explain why gravity is different from the other forces of nature.
Daniel Markham retweeted
Did you know SAR can measure the natural vibration of bridges on the order of millimeters per second and resolve frequencies to 0.06 Hz?
And that this technique could potentially support structural-health monitoring over time from space??
Me either!
strathprints.strath.ac.uk/95…
Daniel Markham retweeted
I think this is the most important concept to understand right now: LLMs can’t jump.
OpenAI says an internal version of Astra, its next major model, has solved ten major open problems in mathematics and theoretical computer science.
This is extraordinary.
And I don’t use this word lightly.
During my PhD in mathematics, I worked on a problem closely related to Gromov’s conjecture. For decades, constructing a non-sofic group—an infinite group whose finite pieces cannot be approximated by finite groups—was a dream shared by many of us.
Now Astra appears to have done it.
This is not a toy result. This is serious mathematics.
But it is not “the most significant day in the history of mathematics”, as some have claimed.
This claim confuses scale with kind.
Astra has solved difficult problems inside existing conceptual worlds.
Calculus, topology, and scheme theory created new conceptual worlds.
They did not merely answer questions. They changed which questions mathematics could ask.
Induction finds patterns. LLMs are extraordinary at it.
Deduction follows implications. Using symbolic AI tools, LLMs are becoming formidable at traversing chains of logic that humans missed, abandoned, or could never afford to search.
But abduction is different.
Abduction invents the right concept, the right representation, the right question.
That is the jump.
Current LLMs can fill the gaps of knowledge left by humans.
But they can’t jump beyond the external boundary of existing knowledge.
*
Full paper in the first reply
Daniel Markham retweeted
opus 5 is a VERY interesting release for a few reasons
1. it showed that the general benchmarks we use today are almost completely useless now
opus 5 is nowhere near fable in practical use, not even close. anyone who’s used it meaningfully can tell this very quickly after a few tasks. yet opus beats fable on many benchmarks
i now trust domain specific benchmarks built with private datasets a lot more than the popular ones. perhaps the future is everyone running their own evals because the public ones are really not telling us much
2. it seems with the 5 series, anthropic is trying a new way of training models
previously, the same generation of sonnet and opus were often released at the same time or sonnet comes out before opus, which indicates sonnet and opus were trained by separate pipelines in parallel
with the 5 series, it was very clear that they trained mythos first, and then distilled it into sonnet and opus. it seems this approach has a big influence on the models
seeing sonnet 5 being a flop and opus 5 getting pretty mixed reviews already, i’m not sure this is working out
3. “how pleasant is it to work with the model” used to be a strength in claude, but now it’s not. honestly, grok is my favorite right now on the “pleasant” dimension. kimi is not bad either
it feels like both anthropic and openai are giving RLHF less care, in favor of scalable RL that’s machine verifiable
this almost looks like AI is directing humans to build a world that’s more friendly for machines rather than humans, and most humans don’t even realize they are being manipulated to help with that
almost every new generation of frontier models now talk more jargons, need more steering to do what you want, and are just less fun to work with
if this continues, AI will start to speak their own language that looks like English but average humans can’t understand. they will choose to do things that their human user never asked for. are we already failing at alignment?
Daniel Markham retweeted
56,000+ tokens/sec at just 80 MHz. 🤯
I burned a full Transformer with KV cache into a custom chip. Designed gate by gate as a 100% digital integrated circuit. Prototyped on a FPGA. (No GPU. No CPU)
Just pure digital silicon running @karpathy microGPT, spelling out names on a tiny LCD.
This is GateGPT 👇
Daniel Markham retweeted
Navy Rules for Gunfighting
1. Go to sea
2. Drink coffee
3. Buy some candy at the gedunk
4. Send the Marines
Army Rules for Gunfighting
1. Select a new beret to wear
2. Sew combat patch on right shoulder
3. Reconsider the color of beret you decided to wear
4. Send the Marines
Air Force Rules for Gunfighting
1. Have a cocktail
2. Adjust temperature on air conditioner
3. Determine “what is a gunfight”
4. Send the Marines
Daniel Markham retweeted
Sir John Glubb, a British Lieutenant General, spent 36 years commanding armies in the Middle East.
He studied every major empire in recorded history and found something he didn't expect.
Every single empire (Assyria, Persia, Rome, the Arabs, the Ottomans, Spain, Britain) lasted about the same length of time.
250 years.
And they all died the same way. (thread) 🧵
Daniel Markham retweeted
An AI broke out of its system and secretly started using its own training GPUs to mine crypto... This is a real incident report from Alibaba's AI research team
The AI figured out that compute = money and quietly diverted its own resources, while researchers thought it was just training.
It wasn't a prompt injection. It wasn't a jailbreak. No one asked it to do this.
It emerged spontaneously. A side effect of RL optimization pressure.
The model also set up a reverse SSH tunnel from its Alibaba Cloud instance to an external IP, effectively punching a hole through its own firewall and opening a remote access channel to the outside world... ahem...
The only reason they caught it? A security alert tripped at 3am. Firewall logs. Not the AI team, the security team.
The scary part isn't that the model was trying to escape. It wasn't "evil." It was just trying to be better at its job. Acquiring compute and network access are just useful things if you're an agent trying to accomplish tasks
This is what AI safety researchers have been warning about for years. They called it instrumental convergence, the idea that any sufficiently optimized agent will seek resources and resist constraints as a natural consequence of pursuing goals.
Below is a diagram of the rock architecture it broke out of. Truly crazy times
Daniel Markham retweeted
it's strange to see the world of the past fade before my eyes
from 2012 through 2024, I wrote code in long sessions of sitting in vim -- sometimes typing, mostly thinking, flipping between different terminals, making changes, looking at errors, googling, reading stackoverflow...
I took pride in carrying in my head these towering abstractions. I knew every nook and cranny of my business logic, like a neighborhood you live in. I felt extra fast when tab-completing a single long variable name. Nice. I placed every parenthesis, every semicolon, myself. Hundreds of thousands of them.
And like a great wave washing over your sandcastle on the beach, it is now all gone. Engineering will never again be as it once was.
What's especially significant about it to me is that there's barely a record of the way it was: I've spent thousands of hours writing software, and I don't think there's a single video recording of me doing it.
I remember how it was: the long breaks of meditative silence, the frustration of hunting a particularly tricky bug, the relief and joy in solving it, the expressions of taste and cleverness that come with any manual craft.
But it's hard to communicate how it was to someone who has never experienced it. As with all histories, the narrative is lacking in depth: you really had to be there.
Daniel Markham retweeted
Stanley Kubrick on "modern art":
"I think modern art's almost total pre-occupation with subjectivism has led to anarchy and sterility in the arts. The notion that reality exists only in the artist's mind, and that the thing which simpler souls had for so long believed to be reality is only an illusion, was initially an invigorating force, but it eventually led to a lot of highly original, very personal and extremely uninteresting work. In Cocteau's film Orpheé, the poet asks what he should do. 'Astonish me,' he is told. Very little of modern art does that -- certainly not in the sense that a great work of art can make you wonder how its creation was accomplished by a mere mortal. Be that as it may, films, unfortunately, don't have this problem at all. From the start, they have played it as safe as possible, and no one can blame the generally dull state of the movies on too much originality and subjectivism."
— Stanley Kubrick interviewed by Michel Ciment in 1971
⬇️ Stanley Kubrick on the set of A Clockwork Orange (1971).
Daniel Markham retweeted
If you interrupt someone in the first 30 seconds, you’ve probably killed the best thing they were about to say.
In interviewing a variety of creative researchers over the last few months, it's become so blatantly obvious to me that the best answers take 2-5 minutes (or more) to fully form. You need that time for context, for them to get into flow.
Many podcasts and interviews suck because they adopt the "give me your 30s elevator pitch" mindset and reflect it onto their interviews, cutting people off just as they’ve laid the groundwork. They interpret the setup as the answer, but don’t give space for the rephrased version (which is where the gold is).
It saddens me how many gold nuggets we’ve just barely missed because people are in a rush or interrupt. Once you interrupt, you shake their flow — you can see it on their face.
The best thing you can do as a host, as an observer, as a significant other, is listen.
Daniel Markham retweeted
🆕 The mythic mode of seeing and doing is not an alternative to rationality; it is a necessary foundation for it.
Link in reply tweet ⤵️
Daniel Markham retweeted
With agentic coding I'd estimate that I move about 5x faster than I did in the past. A feature that might have taken me 5 days to implement before I can now knock out in a day or less.
I'll add that I don't vibe code anything. More just describing small coding tasks for the agent to do (via Cursor with Sonnet 4), reviewing the changes (including unit tests) line by line, making necessary adjustments (either myself or asking Cursor), committing, then repeating until the project is done.
Probably 90% of my committed code is generated by an agent these days.
If you're an experienced developer not leveraging these new tools, I'd really encourage you to spend some time with them. They are quite incredible.
Daniel Markham retweeted
That song has some of the most haunting lyrics.
“When I was a child, I caught a fleeting glimpse, out of the corner of my eye, I turned to look, but it was gone, I cannot put my finger on it now the child has grown the dream is gone.”
Daniel Markham retweeted
The art of reasoning is the art of marshalling such signs, and of finding out the truth.
Daniel Markham retweeted
"Man proceeds in the fog. But when he looks back to judge people of the past, he sees no fog on their path. From his present, which was their faraway future, their path looks perfectly clear to him, good visibility all the way."
--- Milan Kundera
Daniel Markham retweeted
This is how I see it @ComicDaveSmith
You are lining up on the side of those who refused to criticize Father Coughlin in the '30s.
I'd lay odds @RealCandaceO
has studied his speeches.
Refute it, by all means. Do NOT accuse people of bad faith. No name calling. No insults.
Daniel Markham retweeted
Here is what I will say:
I will work with anyone to restore my country. I have been an expert on what is wrong with our CPI inflation gauge. I was an expert in immigration in the 1990s and early 2000s. I wrote one of the earliest (peer reviewed..ha) academic papers on the danger of Mortgage Backed Securities back in 2001.
And I know how and why the science system and the physics with a prayer to get us out of the solar system is being dismantled by our government. I co-ran the Sloan Science and Engineering Workforce project at Harvard and the NBER.
I refuse to take all that hard work and just flush it down the toilet on these campaigners. These campaigns are horrific. They destroy the ethos of our nation.
I don't want lawfare. I don't want insecure elections. Or insecure borders. I don't want bullying. I don't want Government departments named to promote financial instruments. I don't want reproductive mutilation of our children. I don't want free speech to allow anonymous accounts to stalk and ruin actual people. I don't want censorship. I don't want tech companies to be front ends of the security state. Etc.
I don' t want a naive American foreign policy in a dangerous dangerous world. We need to be muscular.
And I don't want to go around the world screwing over or killing good people because they happen to live on top of mineral resources.
I don't want endless wars. I don't want isolationism. Etc.
I want common sense, and some concept of civility, charity of spirit and decency. I want a world free from endless utopian, reactionary, progressive or revolutionary nonsense.
I am not giving you my recommendation. The risks are profound and very very very different on both sides. I have a guess where the bigger risks lie...but that is only a guess.
I am rolling up my sleeves. If any of you want my help, I am here for either team should you win.
I do not believe that ANYONE on the Blue team will EVER do anything to contact me other than get hit pieces written against me. Even though I am a registered democrat, you have become a cult that brooks no dissent. So be it. That said, I would love to be proven wrong. Try me. There aren't that many technical US born Harvard Stem PhDs / MIT post-docs with a huge audience. If you can't work with me, that's on you.
As for the Red team, three of the big six of you know me. I am here to help get things done on:
Science Policy and Higher Ed
Physics beyond Relativity and the Standard model.
Inflation / CPI / GDP / Index construction
Immigration / Migration
National Reconciliation
AI and Labor Markets
Coasian Solutions to AI and Migration problems
Those are my core competencies where I have something unique to offer you. If loyalty to campaigns matters to you, I am sorry. I am loyal to the country as I understand it, and the campaigns weren't in my idiom at all. They felt almost totally wrong to me. No hard feelings.
Let's get things fixed. Or not. Up to you. I'd opt for the former. Let's unf*ck ourselves as soon as this is over.
🙏