Srivatsa Chakravarthy
@srivatsamath
Math researcher working in Random Walks on Groups
I'm extremely convinced that AI math will come to roost within the next fifteen years, and will be beautiful. The theorems they prove will be experienced like La Sagrada Familia I have no basis to make the claim. Just intuition
I've stumbled onto database Twitter. Holy crap there is so much that goes into these things
Amara's law remains undefeated
What having no types and an LSP does to someone
enabled: boolean isEnabled: boolean I trust you more if you use the second convention with booleans
The problems with LLMs stem fundamentally from their inability to apply basic principles across the board: While they might be able to debug a 10,000 line codebase and perform fifty logical steps in Rust, they cannot do the same for a codebase in a niche language like Typst.…
.@RichardSSutton, father of reinforcement learning, doesn’t think LLMs are bitter-lesson-pilled. My steel man of Richard’s position: we need some new architecture to enable continual (on-the-job) learning. And if we have continual learning, we don't need a special training…
Some people just don't keep up with the news
2025: Hoorah, this *looks* perfect! The chatbots will do all the calculus for us! 2030: Wait ... why did that bridge fall down?
>>=, <$>, <*>, <|>
I'm pretty convinced that folks will eventually look back at piles of .map.and_then.or_else (in any language, this is not specific to Rust!) as a big mistake, when you could just write a few imperative lines instead. It'll be like how we look at ultra-OOPy slop from the '90s.
I've noted that the intersection of people with serious political takes and serious programming takes is an order of magnitude smaller than the intersectands
Maybe this is overkill but I feel like LLMs would be way more useful if - I could specialize an LLM towards a particular set of tasks. For example, I could train on LLM on SeaORM - I could easily run 10 LLMs at the same time, each with 500 Tok/s latency
Okay I think I've hit my "GPT-5-Codex is a midwit moment". I asked it to implement a batch insert into a table using SeaORM where you do nothing if there exists a row in the table which shares an entry with the with the row you're trying to insert at column unique The SeaORM…
Amazing! But why use a piss filter image to announce such a great result lol
1/n I’m really excited to share that our @OpenAI reasoning system got a perfect score of 12/12 during the 2025 ICPC World Finals, the premier collegiate programming competition where top university teams from around the world solve complex algorithmic problems. This would have…
Maybe my take is bullshit, but I would imagine that performance engineering is something that LLMs would be very good at. I want to make a small wager that within three years LLMs can just translate things like grep from C to idiomatic Rust with good performance
Ubuntu’s plan to replace the GNU Core Utils with Rust-based reimplementations is going exactly as poorly as predicted. Some Rust versions being 17 times slower than the battle tested GNU C / C++ version. And other Rust-based versions simply failing to work on large files.
So fucking preachy. What happens to the fucking rock???
United States เทรนด์
- 1. Diane Keaton 148K posts
- 2. Mateer 8,725 posts
- 3. #UFCRio 14.9K posts
- 4. Oregon 55.7K posts
- 5. Annie Hall 33.8K posts
- 6. Indiana 27.3K posts
- 7. #iufb 3,669 posts
- 8. Mendoza 15.5K posts
- 9. Drew Allar 1,385 posts
- 10. Northwestern 3,966 posts
- 11. Bama 16.8K posts
- 12. Tim Banks 1,018 posts
- 13. #HookEm 3,813 posts
- 14. Sark 3,407 posts
- 15. Arkansas 10.3K posts
- 16. Oklahoma 27.4K posts
- 17. Huntington Beach 7,541 posts
- 18. Baby Boom 3,154 posts
- 19. Hawkins 3,823 posts
- 20. Raiola 1,322 posts
Something went wrong.
Something went wrong.