observability is the key to acing every step we take ig. a good observer,. > understands the problem from different pov .> understands its occurance at the very first point.> understands the methods that might solve the problem.> understands the edge cases related to each step.

RT @tobi: Both DSPy and (especially) GEPA are currently severely under hyped in the AI context engineering world.

RT @himanshustwts: The Lore of Kalomaze! ⚡️. bringing a great pod with @kalomaze (20yo ml researcher, prime intellect) - we'd talked about….

hey all, need some help. if i wanna create synthetic dataset for tool calling (lets say replicating the complete signal in the data which includes the query, tool to be called and args to be passed, tool results following a proper schema and so on (multi-step and multi-turn.

RT @karpathy: Love this project: nanoGPT -> recursive self-improvement benchmark. Good old nanoGPT keeps on giving and surprising :). - Fi….

people don't talk about how banger of a website is this to get a foundational hang of MoE, its architectural wins and training and scaling challenges around it. do go through it if you wanna learn:

RT @gordic_aleksa: New in-depth blog post - "Inside vLLM: Anatomy of a High-Throughput LLM Inference System". Probably the most in depth ex….

RT @ashu_1069: NP-hard sounds scary, but it's really just math's way of saying: "good luck." Here's the blog:.https….

i just finished watching the second season of Pantheon and i really really hate to admit how feable yet powerful our minds are in conceiving information sparsely fed to us. watching shows like this and Dark, you cannot apprehend where it all starts, there's a point of

RT @DimitrisPapail: Pretraining with cross entropy = learning the best compressor for all texts ever written.Why? .Minimize CE .= Min. KL b….

we didn't deserve this banger of a paper back in 2010. found this randomly while discussing weight initialization effects lol

i write sometimes: do give a read, i kept a question for each one of you at the very end :).

damn. prime intellect just did what hugginface did back in 2020 for models. you can see @PrimeIntellect being the usual visionary group they are. kudos team, more to come!.

RT @NousResearch: Nous Research presents Hermes 4, our latest line of hybrid reasoning models. Hermes 4 builds on….

you gotta appreciate people learning new topics and documenting the process. you literally see what they're learning, how do they process info, you can correct if they're learning something the wrong way or you can learn more from them!. @r4plh will be covering stuff in video.

the best thing in ai you can do currently is to learn RL and build evals. no matter what anyone says, these two are the hottest (and the most interesting ones) rn. this might just be a general purpose knowledge suggestion or a career suggestion, who knows.

RT @redtachyon: The thing about RL envs, or RL env libraries, is that envs should be completely independent from the algorithmic details of….

at this point, whatever the hot topics are in the ai field, @karpathy has discussed it already either via a tweet or as a sidenote. he is that good a visionary.

RT @willccbb: i'll confess i do have a very specific mission in mind with this project. the semi-vague private beta rollout is part of it.….