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 ...
Association for Psychological Science

Mahzarin Banaji Is Probing the Black Box of LLMs

The former APS president asked ChatGPT to describe its implicit biases. Its answer steered her research in a whole new ...
MedPage Today

FDA OKs Blood-Based Test to Help Detect High-Grade Prostate Tumors

The FDA approved Cleveland Diagnostics' blood-based test to help diagnose high-grade prostate tumors and aid in biopsy decisions, the company announced. Dubbed IsoPSA, the in vitro diagnostic kit is ...

These Tools Say They Can Spot A.I. Fakes. Do They Really Work?

Artificial intelligence detectors are increasingly used to check the veracity of content online. We ran more than 1,000 tests ...

Donut Lab Solid-State Battery Claim Challenged After VTT Lab Findings

VTT testing says Donut Lab’s “solid-state” battery likely matches lithium-ion; energy density measured at 298 Wh/kg, claims ...
The Aviationist

Lockheed Martin Tests Missile-Evasion AI on X-62 VISTA

The Have Remy Test Management Project saw USAF Test Pilot School students directly involved in an AI agent’s development, simulation and live trial with ...