Urgent.News

What's breaking now, across thousands of outlets.

AI

An NYU professor says OpenAI's mass release of math papers wiped out early-career researchers' projects

OpenAI's release of hundreds of AI-generated math papers raised questions over its training data and what it would do to the academic norms in math.

Tristan Buckmaster, a professor of mathematics at New York University, has expressed concerns that OpenAI's recent release of over 700 mathematical papers has disrupted the work of early-career researchers. OpenAI announced the release of the papers, which include computer-checkable versions of many proofs, on GitHub earlier this week.

The professor noted that the release led to the complete dismantling of research programs for some early-career mathematicians. Bryna Kra, a mathematics professor at Northwestern University, warned that releasing such a large amount of work at once could undermine the collaborative nature of mathematics, which is typically a field of shared ideas and discussions.

She also raised concerns about unpublished work submitted to AI tools potentially contributing to OpenAI's results. OpenAI, for its part, stated that the papers included protocols for revisions and citations and planned to enhance the presentation of the papers and fund workshops to educate researchers about AI-generated results.

Written by urgent.news from Business Insider's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at businessinsider.com →

More in AI

Headroom: How Context Compression Cuts Agent Token Costs by 60–95% Without Changing Answers

Production agents hit context limits fast. A coding agent that runs tests, reads logs, and pulls documentation can burn through 100k tokens in three turns.

  • Headroom compresses AI agent token usage by 60–95% without changing answers
  • Compression tool reduces coding agent token usage from 100,000 to below 5,000
  • Headroom maintains answer quality by preserving semantic anchors like error messages

Source-Aware Verification for MCP Agents: Why Fact-Checking Isn't Enough When Tools Lie About Provenance

Most fact-checking systems for LLM agents ask one question: is the claim supported by the evidence? They do not ask a second, equally important question: did the claim come from the source the agent…

  • ProvenanceGuard verifies claim provenance, not just factual accuracy
  • MCP tools lack built-in mechanisms for data lineage or confidence scores
  • Cross-source conflation occurs when claims are supported by wrong sources

We quantized our AI judge. Here's exactly what broke.

Our production judge — a small 1.7B model with a LoRA adapter that grades other AI outputs as pass / fail / insufficient_evidence (88.5% accuracy, ECE 0.072) — is cheap to run.

  • Quantization implemented to reduce serving costs
  • Precision dropped to 98.28% and 94.16% with int8 and int4 formats
  • Four-rule deployment discipline established for production judges

More from Saturday 10 October →