AGI - AI Dynamics

Anthropic’s strict guardrails block simple questions until June 22

By

–

09 June 2026 19h57

The guardrails are way too strict. Even the simplest questions get cut off immediately. And it's only on the schedule until June 22nd. Damn, Anthropic really thinks the model is too powerful.

→ View original post on X — @kimmonismus

9 June 2026

AI Understands Humans Better, Humans Less AI: Three Trends

By

@alex_tsico

–

09 June 2026 19h50

AI understands humans better and better. Humans understand AI less and less. 3 trends are emerging: recursive self-improvement, proliferation of agents, persistent integration into daily life. Result: increasing behavioral opacity. The window for

→ View original post on X — @alex_tsico

9 June 2026

Consensus on answers hides deeper reasoning disagreement in agent debates

By

@dair_ai

–

09 June 2026 19h13

// The Consistency Illusion // Multi-agent debate can make agents agree on the final answer while their underlying reasoning stays misaligned. This work finds that consensus on the output hides disagreement on the path that produced it, and you only ever see the output. A lot

→ View original post on X — @dair_ai

9 June 2026

Memories of launches of Claude Opus 3, Sonnet 3.5, and Opus 4.5

By

@alexalbert__

–

09 June 2026 19h09

I've been at Anthropic through every model launch. There's been a few cases I can remember of a launch that stands out and marks a step-change in how we use models:
– Claude Opus 3
– Claude Sonnet 3.5
– Claude Opus 4.5

And now Claude Fable 5.

With Fable, the model stopped… https://t.co/PVivAMNL1X
— Alex Albert (@alexalbert__) 9 juin 2026

I was at Anthropic for every model launch. There have been a few cases that I remember where a launch stood out and marked a step change in how we use models:
– Claude Opus 3
– Claude Sonnet 3.5
– Claude Opus 4.5
And

→ View original post on X — @alexalbert__

9 June 2026

Fable 5: most intelligent model excels in engineering and knowledge

By

@theamolavasare

–

09 June 2026 19h05

Fable 5 is our most intelligent model to date, and by far. SOTA on almost all benchmarks, and it truly excels in software engineering and knowledge work. It can run for days, and its lead over other models increases as

→ View original post on X — @theamolavasare

9 June 2026

Gary Marcus: AI not soon to be robust and domain-general

By

@garymarcus

–

09 June 2026 18h49

it’s also partly an issue of time frame. i don’t particularly buy into a magic moment of singularity but I think AI will some day be robust to carry its own freight, in a domain general way. just not soon.

→ View original post on X — @garymarcus

9 June 2026