AI Dynamics

Global AI News Aggregator

LLMS

MiniMax M2.7 Open-Source: Self-Evolution and Autoresearch at Scale

By

@akshay_pachaar

–

14 April 2026 12h30

MiniMax M2.7 is open-source!

The most interesting part of this release isn't a benchmark number. It's what MiniMax calls "self-evolution," and it's essentially Karpathy's Autoresearch applied at full scale.

Every AI agent today runs inside a harness: the scaffolding of skills,… pic.twitter.com/91lBJXZax5
— Akshay 🚀 (@akshay_pachaar) 14 avril 2026

MiniMax M2.7 is open-source! The most interesting part of this release isn't a benchmark number. It's what MiniMax calls "self-evolution," and it's essentially Karpathy's Autoresearch applied at full scale. Every AI agent today runs inside a harness: the scaffolding of skills,

→ View original post on X — @akshay_pachaar,

14 April 2026
Anthropic Ditches Agents for Powerful Agentic Skills

By

@datachaz

–

14 April 2026 11h52

Wow.

The team behind Anthropic's Agent tools just explained why they ditched agents to focus entirely on Skills, showing just how powerful agentic skills can be.

Also adding this rad article from @exploraX_ that shares 20 powerful skills you can plug into Claude or any model 🔥 https://t.co/Cl46YgixCF pic.twitter.com/cLUxL47krw
— Charly Wargnier (@DataChaz) 14 avril 2026

Wow. The team behind Anthropic's Agent tools just explained why they ditched agents to focus entirely on Skills, showing just how powerful agentic skills can be. Also adding this rad article from @exploraX_ that shares 20 powerful skills you can plug into Claude or any model

→ View original post on X — @datachaz,

14 April 2026
RNNs Match Transformer Memory Without Quadratic Cost

By

@jiqizhixin

–

14 April 2026 11h41

Huge! Recurrent neural networks could match Transformer memory without the quadratic burden! Ali Behrouz from Google and colleagues have cracked it! They present Memory Caching (MC), a simple yet powerful method that lets RNNs store "memory checkpoints" of their internal

→ View original post on X — @jiqizhixin,

14 April 2026
Never Too Late for Cheeky Kimi K2.5 Fine-tuning

By

@maximelabonne

–

14 April 2026 11h32

It's never too late for a cheeky Kimi K2.5 fine-tune

→ View original post on X — @maximelabonne,

14 April 2026
Anthropic Enables Mid-Chat Model Switching Feature

By

@kimmonismus

–

14 April 2026 10h53

Antropic is rolling out switching models mid-chat. About time.

→ View original post on X — @kimmonismus,

14 April 2026
OpenAI V2 Model Imminent Launch Signal Detected

By

@aihighlight

–

14 April 2026 10h26

Image V2 appearing on LM Arena under tape codenames before being pulled is the clearest signal it's close. OpenAI doesn't test on Arena accidentally.

→ View original post on X — @aihighlight,

14 April 2026
Code Quality Benchmarks Show Steep Trajectory Shifting AI Calculus

By

@aihighlight

–

14 April 2026 10h17

The exciting version of this is a year or two away not a decade. The trajectory on code quality benchmarks is steep enough that the calculus will shift noticeably in a short window.

→ View original post on X — @aihighlight,

14 April 2026
Graph Traversal Enables Multi-Hop Queries Beyond Vector Search

By

@akshay_pachaar

–

14 April 2026 10h14

Good questions. Graph traversal adds a small overhead but makes multi-hop queries possible that vector search alone simply can't answer.

→ View original post on X — @akshay_pachaar,

14 April 2026
Gemini CLI feels like placeholder compared to Claude Code

By

@whats_ai

–

14 April 2026 9h28

gemini cli really does feel like a placeholder, i keep drifting back to claude code

→ View original post on X — @whats_ai,

14 April 2026
Memorization as Tool to Accelerate Cognition Not Replace It

By

@fchollet

–

14 April 2026 8h25

The role of memorization and knowledge is to cache & reuse past cognitive work. It should be leveraged as a way to speed up cognition, not as a *replacement* for cognition.

→ View original post on X — @fchollet, 2026-04-14 06:25 UTC

14 April 2026