pushkar

Happy to announce that Etymology Explorer got a UX revamp and is also now completely free to use. No API Keys. No paywall. No signup. Just words and their roots.

@mav3ri3k companies that make swe agents benefit from working closely with the teams who are building runtimes. openai's agent stack is heavy on python + rust, while ant is heavy on js.

@thepushkarp But why are ai companies interested in buying runtimes ? Been hearing that openai also wants to buy deno

called it

To further my understanding and also reproduce the results of the paper - I have setup with repository where we can compare and understand the baseline vs mHC vs attention-residual-full vs attention-residual-block algos properly. Sharing more as runs complete later. link - github.com/tokenbender/na…

the top researchers and engineers have access to the exactly same frontier models as a mid-level engineer, but the gap between them is still the same, if not wider, because ai was never really the differentiator.

also, I miss burning my tokens

how do i politely tell my friend that it's probably not a good idea to include me in his plans because i feel like i'm third wheeling

"laziness, impatientence, hubris" look it up

i'm loving building tiny cli tools and goofy stuff for myself because i just can

claude is down but in dark mode now

attention to each expert in moe next?

I wrote something on Moonshot's latest research release - Attention Residuals. Intuition, notes and how you can understand standard residuals vs mHC vs attention residuals.

@himanshustwts @smallest_AI congrats boi!!!

Career update: Excited to share that I have joined the incredible team at @smallest_AI to work on Research x Devrel! The team is cooking incredible small + efficient multi-modal models and it feels like an exciting time to push the frontier on scale!

@himanshustwts persistence, drive and agency

@thepushkarp i am curious to know more about differentiating parameters

@dejavucoder assuming no insider access lol

@thepushkarp are you sure about the access part hehe

@ExpressGradient dming you

@thepushkarp none the gpumode lectures could help but not much best way is turn the claude opus to max efforts and let it write docs by reading cute dsl source code

chat, what are some good resources to learn cute-dsl that you found really useful?

@tokenbender the same thing exactly happened with me yesterday. i don't remember how much it compressed, but it chose to obey the harness more than what I wanted it to do and just kept getting dumber every turn

@thepushkarp compressed 220k context to 50k without reason and the agent just lost any reference and understanding of my preferences in how i want it to work.

how uninstalling opencode-dcp feels like after it has lost its way