Logo
ResearchAudio
Search
Latest
Free tools
Community
Login
Subscribe free
Logo
ResearchAudio

Archive

Two Humans Worked for a Week and Got 23%. The AI Agents Got 97%.

Jun 8, 2026

The first end-to-end demonstration that AI can run an open-ended research project on its own — with the caveat Anthropic put in footnote form.

Read More

Google Deleted the Encoder Out of Gemma 4 12B

Jun 5, 2026

Vision is one matrix multiply. Audio drops the encoder entirely. It still runs on a 16GB laptop.

Read More

Harness for the Claude Code

Jun 3, 2026

Team-grade Claude Code orchestration using Dynamic Workflows

Read More

How a Tokenizer Mismatch Collapsed GSM8k to 2.56

Jun 1, 2026

Nvidia traced it through 1,100 numerals, then built a projection-matrix fix.

Read More

Test-time compute as a scaling dimension

May 28, 2026

Training isn't the only place scaling laws apply. Letting the model think longer at inference produces predictable gains.

Read More

Anthropic's hypothesized architecture for Claude Mythos

May 25, 2026

Four papers, one repo, sixteen loops of the same block.

Read More

Microsoft Is Pulling Claude Code From Its Engineers

May 23, 2026

Six months from rollout to rollback. The unit economics are not what management is selling.

Read More

We Compressed a Neural Network's Brain. It Started Speaking English.

May 21, 2026

A new Anthropic tool reads activations as English. The audit of Opus 4.6 is the most uncomfortable AI safety story of the year.

Read More

Your Vector Database Costs You 16% Accuracy. A Unix Tool From 1974 Beat It.

May 18, 2026

PwC ran 116 questions through every major AI agent — changing only the search tool. Grep won. The harness mattered more.

Read More

Stop Teaching AI How. Start Teaching It Why.

May 15, 2026

The entire industry is optimizing the wrong variable. The models getting the most funding are learning the least important skill.

Read More
Load more

ResearchAudio

Evidence-led AI briefings and free decision tools for engineers, researchers, and product builders.