The Real Story Behind OpenAI’s New Astra Model
Cal Newport looks at OpenAI's new Astra system, which produced 10 math results on open problems. He argues it shows steady progress in narrow discrete math tools, not a leap toward superintelligence.
The gist
-
Based on leaked details, Newport says Astra is not a simple chatbot but an LLM run by a complex, human-written harness that spawns multiple agents.
-
The 10 results were mostly discrete math constructions, and Noam Brown said OpenAI tried other major problems without success, with no Millennium Prize problems solved.
-
Newport notes an Anthropic mathematician reproduced half the results with Fable within 24 hours, and AlphaProof solved 9 of 353 Erdős problems in May.
-
Newport compares AI progress to tributaries of a river: success in one area, like coding or discrete math, does not carry over to others.
-
He predicts better math papers, graduate-level teaching of these tools, and cheaper tuned models with smart harnesses, but calls the announcement bad news for OpenAI.
Also mentioned
Astral
· "OpenAI announced that their new pre-release AI system Astra had produced 10 math results that resolve or..."
4:46
Lean
· "In AlphaProof, you write your results into a formal proof language called Lean, which can then be..."
9:21
AlphaProof
· "DeepMind has a system called AlphaProof that has a very smart math-solving harness on top of it that spawns..."
Ads in this episode
Done Daily. They're paid, so we keep them out of the picks above.
See what Cal recommends without being paid →
◎
Related episodes
Episodes covering similar topics
More from Cal Newport
“Rogue” AI Is Not Inevitable. It’s a Choice.
Oct 19, 2026
A Former AI Doomer Reveals a Dark Truth
Oct 08, 2026
A Single Sheet of Paper Can Fix Your Overwhelm
Oct 05, 2026
ChatBots Are Killing People!
Sep 28, 2026
A Dramatic Solution for Brain Rot
Sep 21, 2026