Scientists Found AI’s Fatal Flaw—The Most Advanced Models Are Failing Basic Logic Tests

Identifying vulnerabilities is good for public safety, industry, and the scientists making these models.

Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...