~rwx

post-training | mts @microsoftai

[Pawan @PawanKChandana]

As everything slowly begins to sink-in, and after finally getting some sleep last night, I wanted to share a small personal note…✍️

The first call I made after Vikram-1 successfully reached orbit was to my Dad.

He is the man who made my journey possible—with dreams for me far bigger than the ones I had for myself, and by simply never giving up on me.

Until my 8th standard, I was a bad student. I scored 51 marks in Maths and even failed another subject. Back then, I could never have imagined that I would one day make it to an IIT—or have the privilege of contributing to India’s first privately developed rocket reaching orbit, making India only the third country in the world to achieve this.

When I called him after the launch, he just cried.

And cried.

He couldn’t speak a single word.

But I understood every word he couldn’t say.

I managed to hold back my tears when Vikram-1 reached orbit. But somehow, writing this, I can’t.

Thank you, Dad. ❤️

And a big shout-out to every dad who dreams big for their children—and never gives up on them. 🙏

P.S. A picture with Dad from one of my birthdays after starting @SkyrootA.

[Skyroot Aerospace @SkyrootA]

ORBIT ACHIEVED. 🚀

Vikram-1 Test Flight-1 has reached orbit. India's first privately developed orbital rocket has completed its final burn and injected its payloads into a ~450 km orbit, making India the third country in the world with private orbital launch capability.

History is made. 🇮🇳

#Vikram1 #JourneyToOrbit #SkyrootAerospace

[Cognition @cognition]

Introducing SWE-1.7, the most capable model we’ve trained yet.

It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s.

RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

[~rwx @VVatsalya]

@0xSero masterpiece ✅

[~rwx @VVatsalya]

@prajdabre rollouts for sure but more to it if reward signal is like more verbosity more passrate then solution length will increase more with steps so fixing reward is crucial as well

[~rwx @VVatsalya]

@hamishivi @yinn_oscar @RulinShao @TengX6 @natolambert @HannaHajishirzi @waiorg Yes currently I am using Tmax-RL data and playing around with bunch of token cost penalty and tool calls penalty too. It’s helping.

Also how did you use test_initial_state tests, for now I am using the test_final_state and that’s enough for now.

[~rwx @VVatsalya]

@alokbishoyi97 🫡 (please add me too)

[Yang Liu @nlpyang]

Excited to introduce our MAI Code model at Microsoft Build. As shared in the session, this is a MoE (5B active / 137B total) initialized from an MAI pretrained model and trained for real user scenarios with product harnesses. I’m proud to have served as the research lead for this effort, and even prouder of what the team has achieved. It’s a beast for its size. Stay tuned — a larger model could come :)

[Mustafa Suleyman @mustafasuleyman]

Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier.
First is our text foundation model, MAI-Thinking-1, exceptionally strong on reasoning and SWE tasks.
- It’s a 35B active parameter MoE with a 256K context window. Independent human raters on Surge prefer it for overall quality in blind side-by-sides versus Sonnet 4.6, and it’s achieved 97% on AIME 2025, the key measure of its general-purpose reasoning abilities.
- It's at 53% on SWE Bench Pro, placing it right alongside Opus 4.6 on one of the toughest coding benchmarks.
- And since we co-designed our models with our own silicon, MAI-Thinking-1 is optimized on our MAIA 200 chip. Benchmarking head-to-head against the GB200, we see 30% better performance per dollar as well as a 1.4x performance-per-watt gain when running our MAI models on the MAIA 200 end-to-end.

Next is MAI-Image-2.5 and its Flash variant. Two super strong models now at #2 on the leaderboards, surpassing the score of Nano Banana 2 on image editing.

Last for now is MAI-Code-1-Flash, our new inference efficient coding model, especially tuned for VS Code and GitHub Copilot CLI.
- Code-1-Flash achieves 51% on SWE Bench Pro, despite having just 5B parameters, putting it closer to Haiku in size but cheaper in cost.

All of this is the foundation for Microsoft Frontier Tuning. It lets you customize our models to create custom, company-specific agents that only you control. You can make our model, your model. Your data. Your agents. Your moat.

Early adopters are already seeing a difference. When we tuned our models for McKinsey’s tasks, MAI delivered the highest win rate, outperforming GPT-5.5 on quality, while being 10x lower on cost.

Also really excited to be collaborating with the amazing team at Mayo Clinic to jointly train a new frontier AI model for healthcare.

Our announcements today mark another milestone on the road to humanist superintelligence. You can learn more and about our other new models in our latest blog: https://t.co/v65eop5Ixq

[~rwx @VVatsalya]

@khoomeik all about good data with good learning signals

[Guillermo Rauch @rauchg]

1 hour of TravisCore to build and accelerate to

[~rwx @VVatsalya]

@eigenron rubrics,graders,exploration vs exploitation

[mitsuri @0xmitsurii]

Just get started.

[Jason Wei @_jasonwei]

Becoming an RL diehard in the past year and thinking about RL for most of my waking hours inadvertently taught me an important lesson about how to live my own life.

One of the big concepts in RL is that you always want to be “on-policy”: instead of mimicking other people’s successful trajectories, you should take your own actions and learn from the reward given by the environment. Obviously imitation learning is useful to bootstrap to nonzero pass rate initially, but once you can take reasonable trajectories, we generally avoid imitation learning because the best way to leverage the model’s own strengths (which are different from humans) is to only learn from its own trajectories. A well-accepted instantiation of this is that RL is a better way to train language models to solve math word problems compared to simple supervised finetuning on human-written chains of thought.

Similarly in life, we first bootstrap ourselves via imitation learning (school), which is very reasonable. But even after I graduated school, I had a habit of studying how other people found success and trying to imitate them. Sometimes it worked, but eventually I realized that I would never surpass the full ability of someone else because they were playing to their strengths which I didn’t have. It could be anything from a researcher doing yolo runs more successfully than me because they built the codebase themselves and I didn’t, or a non-AI example would be a soccer player keeping ball possession by leveraging strength that I didn’t have.

The lesson of doing RL on policy is that beating the teacher requires walking your own path and taking risks and rewards from the environment. For example, two things I enjoy more than the average researcher are (1) reading a lot of data, and (2) doing ablations to understand the effect of individual components in a system. Once when collecting a dataset, I spent a few days reading data and giving each human annotator personalized feedback, and after that the data turned out great and I gained valuable insight into the task I was trying to solve. Earlier this year I spent a month going back and ablating each of the decisions that I previously yolo’ed while working on deep research. It was a sizable amount of time spent, but through those experiments I learned unique lessons about what type of RL works well. Not only was leaning into my own passions more fulfilling, but I now feel like I’m on a path to carving a stronger niche for myself and my research.

In short, imitation is good and you have to do it initially. But once you’re bootstrapped enough, if you want to beat the teacher you must do on-policy RL and play to your own strengths and weaknesses :)

[~rwx @VVatsalya]

@myprasanna Hope you will sail through this mess, you are a fighter 🙌

[Prasanna S @myprasanna]

My name is Prasanna, who previously founded Rippling (worth $10B); I'm going through a divorce. I'm now on the run from the Chennai police hiding outside of Tamil Nadu. This is my story.

[jack @jack]

19 years

[~rwx @VVatsalya]

SVD is so back !!!
https://t.co/daoII4U00o
"Think of SVD as a surgeon performing a detailed operation on the brain of an LLM."