MiniMax just teased their Sparse Attention architecture for M3. The benchmarks show 9.7x prefilling speedup and 15.6x decoding speedup at 1M tokens vs M2. MiniMax deliberately went back to full attention for M2 because efficient attention wasn't production-ready. Their pretrain
MACHINE LEARNING
-
NPM installations of @openai/codex jump from 100k to 1M per day
By
–
https://npm-stat.com/charts.html?package=openai-codex&from=2026-01-25&to=2026-05-25 … shows that installations of @openai/codex via NPM were about 100,000 per day in January, now over one million per day.
-
Microsoft paper: SkillOpt for self‑evolving agent skills
By
–
One of the most important objects in agentic AI may turn out to be a Markdown file. Not the model weights.
Not the prompt. The skill document. A new Microsoft paper introduces SkillOpt: Executive Strategy for Self-Evolving Agent Skills. The thesis is sharp: If an agent’s -
Huawei’s Tau Scaling Law Key to Post-Moore AI Efficiency
By
–
The future of AI will not only be shaped by better models. It will be shaped by faster, more efficient systems underneath them. That is why Huawei’s Tau Scaling Law (Her’s Law) matters for the post-Moore era. #HuaweiPartner @Huawei #AI #Semiconductors #ChipDesign #MooresLaw
-
AI inference efficiency depends on data movement through memory, chips, and interconnects.
By
–
This matters directly for AI inference. Every AI response depends on data moving through: memory,
chips,
interconnects,
and full systems. If that movement is slow, inference becomes slower, more energy-intensive, and more expensive to scale. -
Custom Claude Code & OpenClaw Workflows: Capafyai Marketplace for Monetization
By
–
How cool!
— Charly Wargnier (@DataChaz) 26 mai 2026
I'm seeing TONS of custom Claude Code and OpenClaw workflows. @Capafyai is building a marketplace for them 🔥
Instead of hoarding your scripts, package them as one-click Skills and get paid per use 👀↓pic.twitter.com/gB8LQI5RNi https://t.co/vD8pOms1XlHow cool! I'm seeing TONS of custom Claude Code and OpenClaw workflows. @Capafyai is building a marketplace for them Instead of hoarding your scripts, package them as one-click Skills and get paid per use ↓
-
Free ChatGPT Deep Dive Video & Article Breakdown
By
–
I still can’t believe @karpathy released this 3.5-hour deep dive on how ChatGPT actually works… for FREE. 🤯
— Charly Wargnier (@DataChaz) 26 mai 2026
Easily the best video I've watched on the topic.
Swap your next Netflix binge for this.
Then read the ace article below from Codez breaking it all down 👀↓ https://t.co/5RYVSFckqq pic.twitter.com/c8spocCq5rI still can’t believe @karpathy released this 3.5-hour deep dive on how ChatGPT actually works… for FREE. Easily the best video I've watched on the topic. Swap your next Netflix binge for this. Then read the ace article below from Codez breaking it all down ↓
-
Typed Schemas for LLM Data Ingestion and Domain Context
By
–
The hard part is that "what matters" is domain-specific and the LLM has no way to know that on its own. That is where typed schemas earn their keep. You encode what matters once, and every future ingestion inherits that judgment without the model having to rediscover it.
-
SkillOpt: Reusable External Skills for AI Agents
By
–
The key idea: SkillOpt does not make the deployed agent larger. It trains a reusable external skill artifact offline, then ships only the improved procedure. That is a very different adaptation layer than fine-tuning, prompting, or adding more inference-time calls.
-
Memory Management in AI Systems: MemGPT and Schema Design
By
–
MemGPT is a great parallel. Paging between fast context and slow storage is essentially the same idea; memory needs an explicit policy, not just a bigger buffer. The schema approach pushes that decision even earlier, to ingestion time rather than retrieval time.