Caleob King retweeted
String theory is what happens when an intelligent system is starved of sensory input and starts hallucinating. Intelligence needs grounding.
For this reason, intelligence is practically never the bottleneck to solving impactful problems. The bottleneck is iterated contact with reality.
Caleob King retweeted
due to accelerating risks I’m going to have to revoke AI researchers’ access to percentages and exclamation points
This is the fundamental contradiction safety purists refuse to grapple with.
Caleob King retweeted
you: anthropic extended fable subs, so much more work i can get do now whooo
me: hey fig, write yourself a full stand-up set, create a comedy club in 3js & hyperframes, use as many tokens as you want to make it something interesting to you
Caleob King retweeted
Staggered AI rollout is not inherently bad. Giving reputable defenders a chance to use powerful models to increase security is good if done predictably, consistently, and transparently.
The problem is the capricious, closed door, extralegal way the White House is doing this.
No entity, including the White House, should have such capricious power over AI.
Caleob King retweeted
The Iranian navy, which has been destroyed eight times, has apparently closed the Strait of Hormuz again, because the United States, for the seventh time, won the war that wasn’t a war, so now the United States has to open the Strait of Hormuz that was already open before the not-war began.
The not-war began because Iran had uranium that was totally, completely, beautifully obliterated, so they can’t build the nuclear bomb they weren’t building, which is why the United States had to start the not-war it definitely didn’t start.
Now the United States, which has nuclear weapons, is threatening to use nuclear weapons to stop Iran from getting nuclear weapons, because nuclear weapons are far too dangerous for countries with nuclear weapons to allow other countries to have.
If the United States saw the United States doing what the United States does in other countries, the United States would invade the United States to liberate the United States from the tyranny of the United States.
Caleob King retweeted
PICARD: Data, shields up
DATA: Brilliant! Shields can reduce damage we sustain. Not immunity. Not hubris. Just prudence. It's not precaution—it's strategy.
[camera shakes]
WORF: HULL BREACHES ON NINE DECKS
DATA: Here's what happened: you told me to raise shields, and I didn't
open.substack.com/pub/wesstr…
Truly amazing article about A.I. and implications for teachers (everyone really but teachers in particular) If you care about learning and thinking and tech - read it.
Caleob King retweeted
We’ve just released another paper solving five further Erdős problems with an internal model at OpenAI: arxiv.org/abs/2604.06609.
Several of the proofs were especially enjoyable to digest while writing the paper. My personal favorite was the solution to Erdős Problem 1091. The question asks: if a graph G has chromatic number 4, while every small subgraph has chromatic number at most 3, must it contain an odd cycle with many diagonals? The internal model gives a very enlightening counterexample to this conjecture, and the proof was a pleasure to understand.
For those so inclined, a really fun exercise is to try to reconstruct the proof from Figure 5 of the paper, which was of course produced by Codex.
Hot take on all the "Hollywood is cooked" narratives. Actors and creatives have a much longer shelf life than representatives of talent (agents). I think we'll see large talent agencies with mostly algorithmic haggling/contract review and a token person to rubber stamp deals.
A firm but kind message about what's ahead regarding AI and what to do about it.
Every time someone asks me what's going on with AI, I give them the safe answer. Because the real one sounds insane.
I'm done holding back.
I wrote what I wish I could sit down and tell everyone I care about.
Send it to someone who needs to read it.
teapaw.org/mattshumer_/status/202…





















