Pritam

Language Modelling & ML systems

[Pritam @Pritamstudyai]

@henrylhtsang Nice insights. Followed

[henry tsang @henrylhtsang]

1. Blackwell (and to some extent hopper) is insanely powerful. Understanding all its features itself is hard
2. GPU programming, no matter which DSL thingy you picked, still depends on compilers
3. A lot of optimization is probably done when designing the kernel. after that,

[aditya @adxtyahq]

MIT CSAI
CMU CSAI
STANFORD CSAI
NTU CSAI
EDINBURGH CSAI

[vx-underground @vxunderground]

All these fucking dorks at Anthropic do is yap about how insane their product is and how end-of-the-world it will be

Someone tell these jabronis to shut the fuck up, holy Christ they're so annoying

[shrey birmiwal @shreybirmiwal]

Hot take the infra stuff around ai is way cooler than model training

[Kris @Krishna70284154]

Nice roadmap!
Instead of a year, if you spend enough time, you could go there in 6 months.

[Pritam @Pritamstudyai]

taste is very important.

[himanshu @himanshustwts]

anthropic blog was quite pragmatic.

as an human, you have got 3 comparative advantages which will matter the most (at least for now!)

> research taste and judgement (choosing which problems matter)
> which results to trust
> when an approach is dead end.

[Atlas Press @realAtlasPress]

Leonardo da Vinci's greatest paragraph

[Logan Thorneloe @loganthorneloe]

Read this to get started learning ML infra.

This is an excellent high-level overview of important considerations in ML training from CMU. It touches on:

- hardware
- memory
- the ML experimentation process

https://t.co/RTWm0Ecni1

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

Interested in learning how to run RL at scale? Here are the best resources to read…

Research on Scaling RL
1. The Art of Scaling RL compute for LLMs: https://t.co/PGjI6Gwgv0
2. Scaling Behaviors of LLM RL Post-Training: https://t.co/2u2saB3C0h
3. Optimally Scaling Sampling Compute for LLM RL: https://t.co/rUSdUvJyNH
4. Scaling up RL: https://t.co/O8vV6z8ymx
5. ProRL V2 - Prolonged Training Validates RL Scaling Laws: https://t.co/vu72juvRW4
6. Polaris - A Recipe for Scaling RL with Reasoning Models: https://t.co/rMibSAeJbg

RL Frameworks
1. Hybrid Flow (early outline of the verl framework): https://t.co/GnWXx131uD
a. More up-to-date info can be found here: https://t.co/j801HcJmPP
2. AReal - Large-Scale Async RL: https://t.co/qhOvsQK09N
3. PipelineRL - Fast On-Policy RL: https://t.co/iRM7KzySXe
4. AsyncFlow - Async Streaming RL: https://t.co/YwmzFtiU2q

RL for Agents
1. DeepSWE - Open Coding Agent Trained w/ RL: https://t.co/GHQHcmtE6F
2. AutoForge - Environment Synthesis for Agentic RL: https://t.co/mr3WDIL5vq
3. Agent-R1 - Training Agents w/ End-to-End RL: https://t.co/xpfQJGgzEv
4. AgentRL - Scaling RL for Multi-Turn, Multi-Task Agents: https://t.co/7fbVl0RWXG
5. The Landscape of Agentic RL: https://t.co/OMnSV4rgdW
6. Training SWE Agents with RL: https://t.co/YqMqySbyXS

Case Studies & Tech Reports
1. Kimi tech reports:
a. Kimi K2 - Open Agentic Intelligence: https://t.co/aAw17SXrIw
b. Kimi End-to-end Agentic RL: https://t.co/ProBpOPIiI
c. Kimi K1.5 - Scaling RL for LLMs: https://t.co/kRGOxY9Jvp
2. Composer series from Cursor:
a. Composer 2: https://t.co/K0v8rNCE6Z
b. Composer 2.5: https://t.co/D9PYimfOMU
3. Olmo 3 (also has open code / data): https://t.co/khetJFvp6N
4. MiniMax tech reports:
a. MiniMax-M2: https://t.co/HApb0OB80S
b. MiniMax-M1: https://t.co/mZj9UQsrnC
5. Nemotron 3 (NVIDIA): https://t.co/lCpE1GzxSi

[elie @eliebakouch]

microsoft MAI tech report is a gold mine, one of the most transparent for a model at this scale.

this model uses zero synthetic data or distillation from previous models. this means reasoning, agentic behavior, tool use are all learned fully during post-training with no cold start. bold choice that makes it harder and requires more iterations to reach sota, but you get FULL control over your model series and it proves they are serious about being a frontier lab.

the tech report is insanely detailed and precise about numbers. to give an example, they give the exact MFU across all the iterations of the model, with the exact changes etc. they also share the full scaling ladder recipe, to my knowledge this is the first time i've seen this in a tech report at this scale

let's look at all of this in this likely very long thread 🧵

[Edward Z. Yang @ezyang]

New devlog post from yours truly: When does fragmentation occur in the CUDA caching allocator? https://t.co/ocAdv4mjy2 -- this post is LLM authored but I heavily prompted/edited, and Natalia also helped fact check.

[Pritam @Pritamstudyai]

@sarthak2143 Congrats 👏🏻🎆

[Parmita Mishra @parmita]

RAS finally getting drugged is one of the great stories in modern biology, and almost nobody outside oncology understands why it's such a big deal.

YOU'LL LEARN SOMETHING AWESOME TODAY.

i am going to keep this as understandable (and simple) as i can.

OPEN THE THREAD.

🧵

[Mark Lewis, MD, FASCO @marklewismd]

Cheers, chills, and a standing ovation when RASolute 302 showed unprecedented survival on daraxonrasib for patients with progressive pancreatic cancer

Seldom do you sense you’re witnessing a historic moment in cancer care but this feels like ras targeting has arrived

#ASCO26

[Chinmay @ChinmayKak]

long overdue website overhaul. check it out, link in the comments:)

[Jino Rohit @jino_rohit]

ml compiler problems are so cool!

tvm, torch.compile, mojo, mlc llm etc they all are really good at taking a computation graph, perform optimization like (fusion, reorder etc etc), put them into a really nice IR and optimize for a specific hardware.

although compilers are really amazing, they probably wont be able to find algorithms like flash attention maybe since its more of a algorithmic inovation but a future where compilers can do this would be awesome.

this is another very cool research space.

[Mark Ajzenstadt @mardehaym]

None of this is satire.

→ A company spent $500,000,000 on Claude in one month because nobody set usage limits

→ Uber ran leaderboards ranking engineers by how much AI they used, not what they shipped

→ Uber burned their entire 2026 budget by April. Their COO said he can’t connect any of it to consumer features

→ A CTO told Axios employees were using enterprise AI to check the weather

→ Microsoft canceled most Claude Code licenses because the token bill spiraled

→ Companies are now laying people off to pay the AI bill. Not because AI replaced the work. Because the bill replaced the headcount.

[koko 𝑥𝑠 @kokoxsu]

https://t.co/dlaW0RSd0T

[Abhishek🌱 @Abhishekcur]

The best thing you can do as an engineer is learn C and C++ deeply.

once you see memory, CPUs, caches, threads, and I/O without layers of abstraction hiding them, you start appreciating how much of modern computing is actually systems programming.

even a bit of assembly, hehe :)