Building... Previously zkml @gizatechxyz, @GetRevelator (acq. by Warner Music)
Joined July 2019
- Tweets771
- Following729
- Followers1K
- Likes2.3K
For the past few days, I’ve been letting my agents run on autoresearch to improve an architecture I’m working on.
I define a metric, freeze the eval and let agents experiment. Keep what improves the score and revert what doesn’t. Then let the harness recursively improve its own research process.
Inspired by Karpathy’s autoresearch, @yukonresearch and Harness², I turned my workflow into a small Claude Code plugin to make it easier to run.
Feedback and contributions welcome:
github.com/raphaelDkhn/infin…
Raphael Doukhan ✨ retweeted
we've been maintaining rust libraries for many years now so we already love the language and have convinced many people about how amazing it is, but it's really the only language where i feel i can write software post-agents and my brain doesn't explode from the complexity
Congrats for the launch @CemFDagdelen
It would be nice to see if musicians can use it to collaborate on Ableton projects (cc @CaseyWescott)
You are in the Squiggle. You've left (or want to leave) your job. You don't know what to build yet. You're reading, journaling, talking to friends, getting nowhere fast. What do you actually want to spend the next five to ten years on?
worldbuilder is a collection of Skills that operationalize @spc' -1 to 0 thesis. Inspired structurally by @garrytan's gstack, but oriented around the pre-founder phase instead of the shipping phase.
Every skill pushes you toward the things -1 to 0 founders avoid: shipping artifacts into the world, talking to external humans, and killing your favorite ideas.
Disclaimer: I'm not part of SPC. I'm just a builder who found the -1 to 0 thesis compelling enough to apply it to my own squiggle. I'd love to see SPC contribute and add their own personal touch (cc @adityaag, @rsanghvi)
github.com/raphaelDkhn/world…
Raphael Doukhan ✨ retweeted
Today, Giza enters a new chapter.
ARMA and Pulse were where we started.
What we're launching today is the product of everything we learned building them.
Unified into one agent. One interface. One name.
And an entirely new universe to experience your onchain capital.
Welcome to Giza World ↓
If we verify these proofs on-chain, we could enable smart contracts or protocols to only accept inputs coming from specific watermarked LLMs
Invoking innovators in onchain agentic space here. I may be wrong, but I think this could be useful for some agentic flows.
@DavideCrapis, @_sumeetc, @lordOfAFew, @renckorzay
Let’s dive deeper into what it really means to prove LLM watermark detection.
🧵
LLM watermark detection isn’t publicly verifiable: checking a watermark typically requires a secret key.
I built a STARK prover for Unigram LLM watermarking using @StarkWareLtd S-two that proves watermarking detection without revealing the key.
Code:
github.com/raphaelDkhn/zunig…
Here is the full implementation:
github.com/raphaelDkhn/zunig…
LLM watermark detection isn’t publicly verifiable: checking a watermark typically requires a secret key.
I built a STARK prover for Unigram LLM watermarking using @StarkWareLtd S-two that proves watermarking detection without revealing the key.
Code:
github.com/raphaelDkhn/zunig…
I dive deeper into the implementation in this article:
raphaeldoukhan.com/articles/…
