t11s
15.3K posts


@transmissions11 @sudolabel i just changed my handle to my real name lol
English

@sudolabel ngl i've been considering just plain doxxing to some extent — i don't really feel compelled to be psuedoanon anymore, it's mostly a relic from when i was just too young to be taken seriously — but idrk how to do this transition in a way that isn't weird lol
English

@tenobrus wdyt about the (admittedly kinda sci-fi) vision of 'big model writes a version of the software in python, swarm of models write and optimize a C/asm/whatever version and then another swarm of models formally verify their equivalence and patch divergences'?
English

not picking on thebes here since it seems to be held pretty loosely, but i've seen arguments like this a lot and they've never made much sense to me. "superintelligence will be smarter than any human and make fewer mistakes, so it will be able to write untyped python without bugs" like sure maybe it will be *more capable* of doing so, but will that actually be the optimal use of its time and tokens? empirically we don't find smarter and smarter people regressing back to writing assembly because they're maybe more capable of writing it bug free. rather they continue to write and use new and more abstractions that allow them to become even more efficient in the future. also the ceiling on memory safety seems pretty goddamn high, given that it remains the number 1 issue with eg C++ even in environments like Google with hordes of language experts and custom tooling. it seems quite plausible that the correct answer for a superintelligence is in fact to just build tooling that makes the problem impossible.
thebes@voooooogel
it makes sense to eat the messiness for now. but oddly it seems ultimately like a not-very-agi-pilled decision? since if you believe in powerful future models, those models should also be able to put defer statements in the right places or whatever and maintain the code in place
English

@JeremyNguyenPhD @cremieuxrecueil he's saying users sharing model outputs double-down, not the models themselves
English

@cremieuxrecueil do you have an example that's shareable?
the old stereotype is the opposite, that LLMs traditionally rollover very easily even when they're very right:
"are you sure?"
"you're absolutely right to pushback on this, I was mistaken..."
below from Sharma paper from Anthropic:

English

🚨JUST IN: A former @PanteraCapital partner is set to lead Solmate, a newly launched Solana Digital Asset Treasury in the UAE, starting with $300M in capital.
Source: BBG


English

What a stacked portfolio
Blend is quite underrated. Where else can you find an nft lending platform without oracles, perpetual, and lender has all the flexibility to liquidate
t11s@transmissions11
some highlights: - helped architect & gas-optimize Permit2 (3.5m+ users), Blur Blend (~1m loans), and OpenSea Seaport (~$9.5b volume) - solmate adopted by Uniswap, Aave, etc - co-authored EIP-4626 standard ($10b+ TVL) - experiments like 0xMonaco, VRGDAs, Flux more to come 🤠
English

life update: this was my last month at @paradigm!
exploring the crypto frontier with @danrobinson, @_Dave__White_ et al these last ~3 years has been a blast.
i'll be transitioning to a research advisor while i explore new frontiers at MIT
English

@transmissions11 @paradigm @danrobinson @_Dave__White_ Your code and research inspired me to get started early on. Can't wait to see what you do next!
English

@transmissions11 @paradigm @danrobinson @_Dave__White_ End of an era. Excited to see what new frontiers will you be pushing!
Will you be in SF during the fellowship offsite?
English

@transmissions11 - gave whatsabi its start with a sick shoutout <3
English













