C3H3tw retweeted
New research from Microsoft Research
I see a lot of AI engineers handwriting agent skill docs and hope they generalize.
Probably not optimal. This works show why.
It treats the skill doc as a trainable external state of a frozen agent instead.
It introduces SkillOpt, where an optimizer model makes validation-gated edits to the skill file. It adds, deletes, or replaces instructions, with a textual learning rate that controls how aggressively each round rewrites the doc. The agent itself never changes.
SkillOpt is best or tied on all 52 (model, benchmark, harness) cells.
On GPT-5.5 it adds 23.5 points in direct chat, 24.8 with Codex, and 19.1 with Claude Code over no skill. It beats human-written skills, TextGrad, GEPA, and EvoSkill, carries zero extra inference-time cost, and the learned skills transfer across models and harnesses.
Paper: arxiv.org/abs/2605.23904
Learn to build effective AI agents in our academy: academy.dair.ai/
C3H3tw retweeted
⏰ We introduce Reinforcement Pre-Training (RPT🍒)
— reframing next-token prediction as a reasoning task using RLVR
✅ General-purpose reasoning
📑 Scalable RL on web corpus
📈 Stronger pre-training + RLVR results
🚀 Allow allocate more compute on specific tokens
C3H3tw retweeted
DeepSeek just dropped the single best end-to-end paper on large model training.
It covers
— Software (MLA, training in FP8, DeepEP, LogFMT)
— Hardware (Multi-Rail Fat Tree, Ethernet RoCE switches)
— Mix (IBGDA, 3FS filesystem)
DeepSeek's engineering depth is insane. Must read.
C3H3tw retweeted
Flow matching produces smooth, deterministic trajectories.
In contrast, the sampling process of a diffusion model is chaotic, resembling the random motion of gas particles.
C3H3tw retweeted
🤯Inspiring cross-domain insights for AI! 🌞Neural Thermodynamic Laws for Large Language Model Training
Specifically, apply Thermodynamic Laws to design learning rate schedules, as the LLM loss landscape mirrors a river-valley terrain: a flat river at the base of steep valleys.
C3H3tw retweeted
Ray Dalio sees the future.
He said, "The US is going into a death spiral of debt."
Few weeks later, Moody's downgraded the US credit score, citing the mounting government debt.
What Ray Dalio sees coming next is disastrous: 🧵
C3H3tw retweeted
A Lean proof of Fermat’s Last Theorem
Kevin Buzzard, Richard Taylor
May 18, 2025
Readers added context they thought people might want to know
This is just the paper version of the blueprint Buzzard has been working on, not announcing a completed version of the project as might be thought
imperialcollegelondon.github.io/FLT/blueprint.…
C3H3tw retweeted
I present to you my AI automation that converts your Google Drive folder to RAG vector database. It also updates the RAG vector database whenever a new file is added or deleted from the Google Drive folder.
You can download the workflow for free (link in comment)
C3H3tw retweeted
🚀 New to Python?
Then here’s a roadmap to get you started in 2025, step-by-step! 🧵👇
(Don't forget to bookmark for later!! 😉 )
C3H3tw retweeted
Here is everything that happened in AI Agents this week 🧵
(save for later)
ALT AgentOps is the leading agent observability company. Every Agent Needs AgentOps. Industry leading developer platform to test and debug AI agents. We built the tools so you don't have to. AI agents suck. We're fixing that. From the team behind AgentOps, Agency helps teams create reliable AI agents at scale. Agent Dev Tools AgentOps is the industry leading agent observability platform. Vetted AI Agents Browse 400+ AI agents working in production. Agent Expertise We help everyone from startups to enterprise build agents at scale. AI agents are the next major paradigm shift in computing. We bring them to life. We've personally built and reviewed hundreds of AI agents. Agency helps clients prototype and productionize ambitious projects. Ready to bring high agency systems to life? From open source AI agent developer tools like AgentOps to Fortune 500 enterprises, we help clients create safe, reliable, and scalable AI agents.
C3H3tw retweeted
2024 is ending!!!
Here are some of my best creations on SQL, HTML/CSS, and JavaScript. Save them, use them as you want.
1. SQL Cheatsheet
C3H3tw retweeted
That's right, @elonmusk ! AI gaming is the future.
And that's us! @XPLA_Official @BNBCHAIN @AethirCloud
Check these out!👇
1️⃣x.com/XPLA_Official/status/1…
2️⃣x.com/AethirCloud/status/184…
3️⃣x.com/BNBCHAIN/status/186595…
#AI #Web3gaming
C3H3tw retweeted
#MachineLearning for Financial Risk Management with #Python — #Algorithms for Modeling Risk: amzn.to/3t7ARbG
—————
#AI #DataScience #ML #Finance #Coding #RiskManagement #DigitalTransformation #Fintech #Insurtech #PredictiveAnalytics
C3H3tw retweeted
For the Taiwan Training Promo Ad on Twitter:
Ready to master #crypto compliance & investigations? Join our hands-on training in Taipei on October 26, 2024, led by world-class experts from TVA3 & Crystal Intelligence. Gain real-world insights into Taiwan’s regulatory landscape.
C3H3tw retweeted
VIDEO NOW BLOWING UP THE INTERNET; Pfizer director on camera saying they are “mutating" COVID-19 Virus to increase infectiousness. UNREAL!
"That is Not What We Say to the Public" …
"People Won’t Like That’ … ‘Don’t Tell Anyone"
C3H3tw retweeted
At 2100+ (not a typo) pages 🤯, this is almost all the Math you need for Machine Learning without dumbing it down! PDF link: cis.upenn.edu/~jean/math-dee…
