Muhsinun Chowdhury

AI/ML Researcher

[Benjamin Tay @bennytay07]

imo people are thinking too small about the north star for robotics

eliminating manual labor is cool, but why only on earth?

life must become multiplanetary; we need embodied AGI to construct space stations, build settlements etc

thats the true frontier

[Muhsinun Chowdhury @MuhsinunC]

@kallasmaa @AniC_dev Interesting theory

[Muhsinun Chowdhury @MuhsinunC]

@AniC_dev Have yet to test it since I haven't bought an account yet though haha

[Muhsinun Chowdhury @MuhsinunC]

@AniC_dev I've had my agents researching the cheapest and best options for my software factory and CI, and https://t.co/JGV1YNOhJH keeps coming up first 👀

[Muhsinun Chowdhury @MuhsinunC]

@jordanschenck @gnukeith @0xSero Had to check again 😭

[Muhsinun Chowdhury @MuhsinunC]

@FishAudio This is sick 🔥

[Muhsinun Chowdhury @MuhsinunC]

@thsottiaux Codex has massive cache hogging my disk space

[Muhsinun Chowdhury @MuhsinunC]

@gucaslelfond But can they scale across smart fridges?

[Muhsinun Chowdhury @MuhsinunC]

@jparkjmc @ycombinator Dude this looks so fun to work on

[Muhsinun Chowdhury @MuhsinunC]

@thsottiaux Were we being charged extra for long-context on subscriptions before now?

[Muhsinun Chowdhury @MuhsinunC]

@thsottiaux 💀

[Muhsinun Chowdhury @MuhsinunC]

@mainteemoforfun MashaAllah Allahumma baarik

[Muhsinun Chowdhury @MuhsinunC]

@OpenAIDevs LET'S GOOOOO

[Muhsinun Chowdhury @MuhsinunC]

@workersio @ycombinator Sick idea

[Muhsinun Chowdhury @MuhsinunC]

@sxhivs I've experienced this as well. What I do is I tell it to keep a log containing predicted estimates, and then real wall-clock time when it's done. And then whenever something completes, readjust all estimates for future tasks, and then it becomes a little bit more accurate.

[Muhsinun Chowdhury @MuhsinunC]

@samraaj @ycombinator Non-Euclidean websites

[Muhsinun Chowdhury @MuhsinunC]

@maharshii Yup. This is the direction software as a whole is moving.

[Muhsinun Chowdhury @MuhsinunC]

@MLCatttt @t_ed_li @junyu_r @BanburyRoadAI @speedrun If we can select our own set of active "experts" fine-tuned to our workload, this could be more cost-effective and potentially more performant than MoE. If we can't select the active "experts", how is this different than MoE?

[Muhsinun Chowdhury @MuhsinunC]

@DavidSHolz Funmaxxing

[Muhsinun Chowdhury @MuhsinunC]

@quantbagel ✋🏼