rock 🗿 (@amrockfeller) 's Twitter Profile
rock 🗿

@amrockfeller

- élève du réel
- all things AI / ML

ID: 1404051944147193856

linkhttps://github.com/rock-feller calendar_today13-06-2021 12:24:20

473 Tweet

84 Followers

603 Following

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

A mathematically unnecessary fancy opinion which however, can hold true: if you want to fully understand #git, learn state space models first.

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

Getting rid of fact-checkers and replacing them with community notes, sounds like déjà-vu! I wonder how Yann LeCun feels about this! Get your 🍿. It's about to go epic this year!

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

With all the things we have been told about China, who could have thought that not only they will lead the race of open source (os) models, but their deepseek model will outperform the best close source model in several benchmarks.

With all the things we have been told about China, who could have thought that not only they  will lead the race of open source (os) models, but   their deepseek model will  outperform the best close source model in several benchmarks.
rock 🗿 (@amrockfeller) 's Twitter Profile Photo

When using LLMs as coding assistants, avoid letting them code from scratch. Instead, ask them to: - Break your code into modular components for better organization. - Write unit tests to ensure robustness and anticipate user behavior. - Create concise documentation for clarity.

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

While DeepSeek is busy thinking and reasoning, don't sit still and wait. - Mend the documentation of working materials - Sharpen your shell scripting skills - Prepare a bunch of concise commits messages for your fellow contributors Just do sthing. You can even clean your house

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

Still remember when I complained to ollama about structured outputs as if they were the model providers. Then, I discovered DSPy and my life changed. SO MUCH has happened since Jan.

rock 🗿 (@amrockfeller) 's Twitter Profile Photo

Q: have got this complex DSPy program that uses 10+ async dspy.Predict modules, all working concurrently. What strategies are recommended to manage concurrent traffic from millions of users?