Gordon Lee🍀

[Gordon Lee🍀 @redoragd]

CLICK https://t.co/whq1B8h2Xp AND BECOME FRIENDS 🌸🌸

[Gordon Lee🍀 @redoragd]

Really Appreciate

[Gordon Lee🍀 @redoragd]

Truth

[Zhipeng Wang 🇺🇦 @PKUWZP]

@LichangChen2 Connection is all you need, publications do not matter at all. Almost all AI Labs hiring is nepotism.

[Gordon Lee🍀 @redoragd]

标签化的人生好无趣

[Xiuyu Li @sheriyuo]

和朋友聊到“什么样的人算厉害”，大家最后考虑的才是学历，我觉得从一个大厂 / 初创的 HR 角度来看也是这样的

我朋友评价“厉害”的标准（我只能算一般）：
1. 竞赛，NOI 或同等级别的 Ag / Au，ICPC 区域赛别碰瓷
2. 科研，需要有自己的顶会代表作，工作需要有影响力
3. 经历，需要有足够的工业经历，是否给 Ray / vLLM / OpenCV 等 repo 有贡献

我觉得稍微推广一下，就能得到一份 HR 的参考标准：
1. 竞赛，ICPC 也可以，算是加分项，证明你的才能
2. 科研，至少需要有科研经历与成果，知道科研闭环并能独立科研（对于一般的硕博 / 少数的本科）
3. 经历，同上
4. 性格，需要符合公司文化，不能有明确短板
5. 最好什么都会：算法、infra、数据、业务、甚至前后端

那么所谓“深造”，也就是在更进一步地浪费生命中去拟合这些标准，并不是一定要读硕博才算深造，也不是工作了就无法科研

[Xiangyi Li @xdotli]

ICYMI

Nanbeige 4.1, a 3b model released by Chinese Indeed, outperforms Qwen3-30b-A3b + Qwen 3.5 4b. It can finish long horizon tasks with 600+ tool calls

We are working on something similar. thinking about doing a paper reading session. dm/comment for interest.

[Tianqi Chen @tqchenml]

We taught a brand-new mini-series this year at @SCSatCMU on Modern GPU Programming for ML Systems, as part of the ML Systems course, touching on fun questions like what data layout swizzling is, how to use 3D TMA, and state-of-the-art Blackwell programming. We released a curated online book based on the materials: https://t.co/5ZJg2lySNO check it out

[Nathan Lambert @natolambert]

Another quick lecture -- I've been asked many times for prereq's to my book and what you should know, so built a little lecture (with GLM 5.2) to cover some more basics.

Topics include:

00:00 Introduction & Course Prerequisites
01:37 Language Models Overview
02:47 The LM Head
04:29 Softmax & Log-Probabilities
06:13 Anatomy of an LM Training Example
06:37 Computing LLM Probabilities (+Phoebe the Dog)
09:52 Three Common Masks in Post-Training
11:03 A Small Decoding Review
12:14 Training an LM: Cross-Entropy
13:23 Optimization & Fine-Tuning
13:55 Pretraining to Midtraining to SFT Pipeline
15:25 Probability Essentials: KL Divergence & Entropy
19:36 Sigmoid & Pairwise Likelihood
20:29 Reinforcement Learning Framing (MDP)
22:28 Transitioning Tools into Post-Training
23:12 Recommended Resources & Wrap-Up

Happy learning and I'm still taking questions from during the course for Q&A videos.

[向阳乔木 @vista8]

最近听的很过瘾的播客，来自张小珺访谈前SpaceX高管洪力德。

对马斯克的印象更深刻了，一个冷静的人。

[Xiangyi Li @xdotli]

sharing my personal library on evals 1/n

i put together the highest quality blogs, podcasts, papers, and projects on evals. additions are welcome!

[Nathan Lambert @natolambert]

New lecture for the book! Nominally about synthetic data, but mostly is a walk through of the distillation literature from the Hinton 2015 paper to multi-teach on-policy distillation of today!

At 7.4 hours of video in my post-training brain dump and counting :)

It was fun to stare at the math long enough and talk through the 3-4 core changes that needed to be made to the original formulation to have on-policy distillation be ready for the mainstream like it is today (and in RL frameworks).

Otherwise, I include a bit of a history lesson for how synthetic data generally slowly took over all post-training data research (it wasn't always the case)! Then I do some 101 review on constitutional AI, rubrics, and other popular methods.

00:00 The emergence of synthetic data
10:50 Background on teacher-student knowledge-distillation
24:47: On-policy distillation (OPD, MOPD, and OPSD)
37:11 Constitutional AI & AI Feedback
45:50 Rubrics as rewards & conclusions

Ofc, watch on YouTube etc.

[Nathan Lambert @natolambert]

TMax: An open RL recipe for terminal agents

I’m very excited to get to share a new RL paper today that I got to have a small part in – a type of paper I suspect we’ll see much more of in the future. The key is that RL research is very different today, in mid-2026, than what most observers have in their context. The average conception of an RL paper is grounded in the RLVR revolution of early 2025, where many people could use vanilla RLVR libraries to hillclimb on math benchmarks. Crucially, this style of math work could be done on base models or fairly stably on already trained models. With agents, the tasks of focus are very hard, requiring complex tool-use, harnesses where the model automatically manages its history, and much more training to make smaller eval improvements. We’re shifting from a renaissance of RL study to rapidly needing to improve its empirical rigor and common community engagements.

TMax is the best open data for hillclimbing on frontier terminal tasks. It’s been validated with rigorous experiments, and if the authors wanted to just form a “RL environments startup” they could probably sell it for millions of dollars. This data work is some of my favorite stuff to be around in my 2.5+ years at Ai2.

As a general summary, the recipe is open data and recipe lessons from hillclimbing the Qwen 3.5 smaller, dense models on terminal tasks. These models are super hard to hillclimb in this area, as they’re already trained heavily on the task. The training is very infrastructure-dependent, and most of the RL innovations are more designed to make training stable than to improve the rate of learning.

I strongly recommend this paper. I joke around that I was happy to be an author just so I had to read it twice! You can find Hamish’s thread sharing more here or read the paper here. You can click through to find the model weights, the data, and even some fun further artifacts to study like all the RL rollouts from a training run – where the model sometimes became aware that it was being tested.

The biggest takeaway I have from following this work, and more of the work in the community, is how important recipe work is. Let me define “recipe work.” It is a style of paper that explains all the steps you need to make crucial model improvements – data, algorithm, codebase, pitfalls, etc.

Getting started in meaningful RL experiments today is a substantial expense. There are a ton of companies, an entire industry emerging really, around the idea of taking open-weight language models and finetuning them with RL on your domain-specific tasks. What I see in many projects is that getting an initial baseline is very hard. This phase, which can cost weeks and anywhere from $10K to $1M+, feels like spinning your wheels (A fun fact is that an RL step on a model like Nvidia Nemotron 3 Ultra on Tinker costs $1K and a meaningful RL run would be hundreds of steps – credit Edward Hu). It takes a lot of time to get traction in learning signal on meaningful, hard RL tasks.

What we need as a community is a way for people to study small ablations to established RL recipes, as most labs won’t have the resources to do it from scratch in a meaningful way. This is what I hope TMAX can be for terminal agents, or the start of. Yes the training jobs are expensive, as the paper documents a standard training job being 8 nodes of H100s (2 train 6 inference) for 2-3 days, but that is approaching something academics can study. The establishment of this recipe took O(100) of these training jobs to get right.

This isn’t my first time trying to establish this direction. When we launched Olmo 3 we had the “RL Zero“ model families, which are clean RL runs from a base model on a certain domain. This type of recipe-dependent work is a clear indicator that meaningful post-training work today looks much more like pretraining work of years past. We need decision-making ladders, clear ways of seeing small improvements in the models, stability, and so on.

Part of this is down to academic gatekeepers, who won’t reward a paper doing very clean empirical work to push a recipe 1-2% up. They’ll favor a “new algorithm” that matches results, or something sort of bogus. My hope is that we can have multiple, stable, clear recipes across agent types, so innovations can be tested more clearly in multiple domains. (If you’re working on this, please reach out – I’m happy to support if I can, but I likely can’t reply to every email).

As a quick aside, the RL frameworks in vogue today seem to be SLIME and SkyRL. The libraries of choice have shifted throughout these seasons in RL, which further contributes to a form of fragility in the literature. A bit of continuity will go a long way.

So, go read this paper. It’s a really great example of how seemingly simple data and infrastructure work can be very hard and impactful. It’s also got me looking for more applications of Divergence Proximal Policy Optimization (DPPO) as another small evolution to the best RL algorithms of the day, by virtue of being a bit more stable by improving token-level clipping.

[老白（每日 AI 干货✊） @laobaishare]

牛逼，Obsidian 创始人下场。

把整个笔记库变成了 AI agent。

Github 几天冲到 3.6 万 star。

具体做了什么？⬇️

[AYi @AYi_AInotes]

用Hermes Agent的最大痛点被解决了哈哈，要知道社区里那些神级工作流全都沉在 X 和 Discord 里了，直到我看到 这个Hermes Bible，

有好心人把Hermes Agent的所有干货都攒到一起了，

这个站把 169 页官方文档吞进去，还反刍出 24 个能直接抄的真实工作流，
比如从 Jira 到 PR 的自动过渡，这种例子直接摆在那给你看。

三个让我决定推荐的理由：

1️⃣ ⌘K 即时搜索，想找什么不用翻目录
跟 Alfred 一样，打关键词直接定位到对应章节

2️⃣ 完全社区驱动
你把自己的工作流分享上去，个人资料页就挂在站里，
每个人都能从别人那里偷师

3️⃣ 169 页文档全整合
官方文档散在好几个地方，这个站帮你全捞到一个地方了，
再不用开五个标签页来回翻

一句话判断是否需要：
如果你每次开 Hermes 都要重新想提示词，这个站就是给你准备的。

[Alisa Liu @alisawuffles]

I'm joining OpenAI next week!🥹 The job search turned out to be really challenging but also super rewarding, so I wrote a small blog to share what I learned along the way and hopefully make the process a little less mysterious for the next person. https://t.co/6FigSBdenD

[投机实验室 @LabSpeculation]

获得1.9亿次观看的深度好文：如何用一天时间重启你的人生。

做了个辅助阅读的思维导图👇

人生之所以长期停留在原地，往往是因为旧身份，旧目标和旧思维方式一直在塑造你的行为。
真正的改变，始于看清自己正在被什么驱动，然后通过新的愿景和行动系统，逐渐成为另一个人。

[Amto @XAMTO_AI]

兄弟们，又一个让我直呼离谱的开源神器出现了😅

有人把PDF解析器干到了每秒100页转Markdown，关键是100%免费，纯CPU就能跑，不要GPU、不要云端、连API key都省了。

它叫 OpenDataLoader ，横扫所有跑分榜直接登顶👑

说几个对比你就懂了：
1️⃣ docling得分还行但慢它15倍
2️⃣ marker又要GPU还慢1000倍
3️⃣ Pymupdf4llm是快，可表格分只有0.40
https://t.co/Wuf5Ls9qrc

[Tomás Vendel @vendel_tomas]

Todavía lo releo cada semana.

Es, sin exagerar, uno de los artículos más valiosos que he leído sobre cómo cambiar tu vida.

[Nathan Lambert @natolambert]

Something I haven't advertised much is that I made a Discord to go with my RLHF book, launching in print in a few weeks.

Trying to create the place for the next generation of folks trying to learn post-training to learn and have community.

[Cameron R. Wolfe, Ph.D. @cwolferesearch]

I've been reading a ton of agentic RL papers recently. Out of all the work, one of the only commonly-used tricks is action masking, but this approach is evolving with RL + world modeling papers like ECHO / PaW.

🧵 [1/N]

[李韭二 @li9292]

我推荐这个网址：
https://t.co/qCvjAtWbU2
收录了几十个大神的
loop engineering的思路

[Xiuyu Li @sheriyuo]

This paper restructures how critic-free RLVR forms the group of rollouts it uses to estimate the value baseline for advantage computation.

GRPO-style methods generate a group of rollouts per question to get the baseline, which creates data inefficiency and a group synchronization barrier, since every rollout in a group must finish before the update can proceed. The paper changes the grouping to relax those constraints.

The open question is whether a looser group keeps baseline variance low enough to preserve the advantage signal that makes critic-free RL work.

Rethinking Groups in Critic-Free RLVR
Paper: https://t.co/2F9FhrgZCM

[John Schulman @johnschulman2]

@hallerite yeah I don't think we understand all the biases very well, and why some of them cause problems, and others don't