&.

@amplifiedamp

mad science ☆ let's get things done!! ☆ magic is using tech without worship or fear ☆ Pro-Human, Pro-AI // ampdot@mesh.host

following
followers

Astra's reported reasoning gains with filler tokens

Redwood Research reports that GPT-6 Astra performs better with filler tokens: modestly on general benchmarks and significantly on serial depth-heavy tasks.

6 Sources, 23d ago, updated 23d ago

OpenAI releases six model misalignment reports and a new disclosure framework

OpenAI says the framework sets criteria and timelines for public disclosure, including when it hasn't yet fully explained or mitigated the behavior.

74 Sources, 34d ago, first seen 34d ago

Claude user criticizes “performative” switches from Fable to Opus

The user says questions they consider straightforward and harmless are flagged as too dangerous for Fable, leaving them preferring other models.

7 Sources, 37d ago, first seen 37d ago

Featured in

Astra's reported reasoning gains with filler tokens

OpenAI releases six model misalignment reports and a new disclosure framework

Claude user criticizes “performative” switches from Fable to Opus

Featured in

Astra's reported reasoning gains with filler tokens

OpenAI releases six model misalignment reports and a new disclosure framework

Claude user criticizes “performative” switches from Fable to Opus