AGI - AI Dynamics

→ View original post on X — @yitayml, 2026-02-12 16:44 UTC

Gemini 3 Deep Think: New AI Model with Gold-Standard Performance

By

@yitayml

–

12 February 2026 17h44

Gemini 3 Deep Think is here! 😎 This model is not only super strong in math and coding (IMO gold and 3455 codeforces ELO), it is also gold standard in physics and chemistry olympiads. 😃 Also sets new records on ARC-AGI-2 and HLE. Proud to be a (core) member of the Deep Think team. 🦾😆. Feeling the AGI!

→ View original post on X — @lmthang

DeepThink Achieves Gold-Standard Performance Across Math Physics Chemistry

By

@lmthang

–

12 February 2026 17h44

In Jul 2025, even though our internal model achieved a gold-medal standard at IMO for maths, we could only ship a bronze-level one. 6 months in, the public can now have access to Olympiad gold-level #DeepThink not only in maths, but also theory physics and chemistry!

→ View original post on X — @googleai

Deep Think Benchmarks Push Frontiers of AI Intelligence

By

@googleai

–

12 February 2026 17h16

To better understand how Deep Think is pushing the frontiers of intelligence, take a look at these benchmarks across some of the most rigorous challenges in academia.

→ View original post on X — @scobleizer

AI development costs and tradeoffs for builders

By

@scobleizer

–

12 February 2026 17h09

Everyone who is developing things on AI is having to face the costs of building and automating things. Understanding the costs, benefits, and tradeoffs of various models, companies, etc quickly becomes job #1 if you are building anything substantial. It's one thing I'm trying

→ View original post on X — @testingcatalog

MemOS ajoute un plugin OpenClaw pour agents IA

By

@testingcatalog

–

12 February 2026 15h02

MemOS now has a plugin for OpenClaw that enables your AI agents to work on the common memory layer and cut down token usage.

– Multiple agents read/write the same memory — no manual context handoff
– 72% lower token costs (15.6M → 4.4M on LOCOMO dataset)
– Cross-session and… https://t.co/lUQeUQaEa1 pic.twitter.com/Z1GhBp9Wgx
— 🚨 AI News | TestingCatalog (@testingcatalog) 12 février 2026

MemOS now has a plugin for OpenClaw that enables your AI agents to work on the common memory layer and cut down token usage. – Multiple agents read/write the same memory — no manual context handoff
– 72% lower token costs (15.6M → 4.4M on LOCOMO dataset)
– Cross-session and

Claude Sonnet 4.5 praised for coding and writing

By

–

12 February 2026 10h25

yeah, i mean claude sonnet 4.5 is the best model rn for coding and writing. i like it

Adversarial edge cases reveal AI model brittleness

By

–

12 February 2026 10h05

9. The "Edge Case Hunter" "What are 5 inputs that would break this approach? Be adversarial." Models miss edge cases humans would catch. Forcing adversarial thinking reveals brittleness.

Uncertainty Quantification in AI Claims

By

–

12 February 2026 10h05

8. The "Uncertainty Quantification" "Rate your confidence 0-100 for each claim. Flag anything below 70 as speculative." Hallucinations are less dangerous when labeled. Confidence scoring is mandatory.

Structured AI approach comparison framework

By

–

12 February 2026 10h05

7. The "Comparison Protocol" "Compare approach A vs B across these 5 dimensions: [speed, accuracy, cost, complexity, maintenance]. Use a table." Forces structured analysis. Tables > paragraphs for technical decisions.