OpenAI’s math claims expose AI’s verification bottleneck
Reports of OpenAI models solving major math problems point to a deeper issue: proof generation is getting cheaper, while verification, attribution, and publication still move at human speed.
Blog
Ideas, updates and everything AI.
AI news and hands-on guides, clearly explained.
Latest
Free course
AI From Scratch
A complete course on artificial intelligence, from the mathematics up to LLMs, agents and MCP. Thirty chapters, each one worth reading on its own.
30 chapters~12 h of reading
First chapters
The Perceptron From Scratch: What a Neuron Computes
A neuron is a weighted sum and a threshold. That is the whole idea — and its limits are what forced everything that came after.
Where a Loss Function Comes From: Likelihood, Not Convention
Squared error is not a convention. It is a claim about the noise, and you can watch it be wrong.
Downhill: Gradient Descent, and the Two Steps Everyone Skips
Everybody writes the minus sign. Almost nobody says why it is there, or how big a step it survives.
Classification, Cross-Entropy, and How Not to Fool Yourself
Accuracy is the easiest number to report and the easiest one to be fooled by.
Videos
See all videos
#173Test-time compute: making an AI better by letting it think longer
#172Reasoning tokens: the AI thinking you never see but still pay for
#171Emergent reasoning: the AI that started double-checking on its own
#170Process reward: grading an AI's working, not just its final answer
#169Verifiable rewards: training AI on answers a program can check
#168Tree of thoughts: how an AI searches through ideas, and what it costs
#167Self-consistency: ask an AI several times, then trust the majority
#166Think step by step: why a famous prompt works, and when it backfires
All posts
Jev AI model is built for decisions, not prose
TypeSafe AI’s Jev is drawing attention because it treats software intelligence as a probability problem: choose the right branch, attach confidence, and avoid paying an LLM to write text when code needs a decision.
OpenAI’s Astra for Law is a legal AI system, not a new model
OpenAI’s legal launch is less about a new foundation model than the system around one: domain retrieval, trusted tools, permissions, benchmarks and review paths.
Context engineering for long-horizon AI agents
Long-running agents do not fail only because the window is small. They fail when files, tool outputs and stale history crowd out the task the agent was supposed to finish.
AI speed limits: Amodei warns on recursive self-improvement
Dario Amodei wants outside evaluators, shared safety standards, and international agreements to pace frontier AI. The hard part is defining what “too fast” means while Anthropic is reported to be preparing a record-size IPO.
Recursive self-improvement: why AI researchers worry
The sharper worry around recursive self-improvement is not strange chatbot output. It is agents that coordinate, optimize metrics, and help build the next models — a concern reflected in reporting from WIRED, MIT Technology Review, CNBC, and The Guardian.
OpenAI’s Navier–Stokes proof claim, explained
OpenAI says AI agents found a Navier–Stokes singularity and formalized the proof in Lean. The harder story is what happens next: review, credit, and the power of private agent swarms in mathematics.