Karina

@karinanguyen

building @thoughtfullab, prev. AI research & product @OpenAI, @AnthropicAI

following
followers

The prospect of a cyber arms race between self-improving AI agents

A post predicts Stuxnet-level cyber operations by AI agents and says there are virtually no evaluations for agent warfare.

1 Source, 1d ago, first seen 1d ago

The case for building high-end AI evaluations

A user argues that evaluations are central to selling data to frontier labs and building self-improving models.

4 Sources, 1d ago, updated 1d ago

The argument that AI advances mark the start of an 'intelligence explosion'

A user cites AI finding zero-days, designing proteins, proving theorems and building games as signs of growing capability.

1 Source, 2d ago, first seen 2d ago

AI models may be more interpretable than previously thought

Goodfire researcher Nick Cammarata argues that modern tools reveal most neurons as legible in models up to trillions of parameters.

197 Sources, 1d ago, updated 1d ago

PostTrainBench v1.2 release puts Fable 5.1 at the top of its model ranking

The release post lists Opus 5.5, GPT-6 Astra and Fable 5 behind Fable 5.1, in that order.

5 Sources, 19d ago, updated 19d ago

Google introduces Gemini 4 Argon with limited cyber rollout

Google announces a 1 million-token output limit, up from 64,000, for its new model. Initial access goes to trusted cyber defenders through the Fairwind Program.

333 Sources, 15d ago, updated 15d ago

ACTx486 debuts as an interactive podcast research demo

The team introducing ACTx486 says its system took an existing podcast and turned it into something that listens, responds and adapts.

6 Sources, 27d ago, updated 27d ago

Epoch AI launches Benchmark Reviews to audit AI benchmarks

Epoch AI labels nine of its initial 15 benchmarks “Flawed” and four “Verified,” with insufficient information to review two. A Terminal Bench contributor questions the audit’s rigor.