How much of SQLite, FFmpeg, PHP compiler can LMs code from scratch? Given just an executable and no starter code or internet access.
Introducing ProgramBench: 200 rigorous, whole-repo generation tasks where models design, build, and ship a working program end to end. 🧵
anthropic had mythos internally since 2026-02-24, so the leaked claude code (2026-03-31) was being written by it for over a month. if you checked the code, knowing these pieces of information will help you temper your expectations.
Claude Code leaked their source map, effectively giving you a look into the codebase.
I immediately went for the one thing that mattered: spinner verbs
There are 187
@gakonst I got claude to build it and it more or less 1-shot it. Used to have a indexing system using templates to have it search over but was really slow. A lot faster using a RAG, only thing that is annoying is rebuilding times but that should be rarely done. lookups are fast
Proud to partner with @NasdaqBTCS as they double down on their commitment to sustainable onchain markets
Their long-term view on DeFi infrastructure is right in line with what we’re building with Angstrom. Stay tuned for more big announcements in the coming weeks.
We're looking to build the best protocol engineering team in DeFi. You’ll be joining myself and @69WillSmith at the ground floor.
$400K+ TC for world-class builders. Dms are open.
LPs aren’t just underpaid, they’re underinformed.
You see a baseline APR and TVL, but not the real mechanics behind your returns.
With Angstrom’s Smart APR, you can finally model how price movement and range concentration affect your yield before you deploy.
No more guessing.
Now live: @titanbuilderxyz × @angstromxyz
The first DeFi app with a native block builder integration, supercharging LP returns.
Titan includes our bundles at 0 priority fees, minimizing gas costs for traders while also conditionally unlocking Angstrom pools for routers & solvers to ensure cheap, reliable access to the most performant liquidity on Ethereum.
@real_philogy Tl;dr I don’t think we have been building them wrong but rather the financial gain without the pressure of moving to a decentralized network has caused stagnation in the l2 space.
I think the main problem with L2’s is that they were all built with decentralization as a roadmap goal rather than a core primitive, given this was the standard to do centralized -> decentralized. Given pmf, I don’t think the incentives have been strong enough to really push the needle forward in respect to this, really hoping that the tradfi push here were decentralization importance when it comes to adoption will kick the l2’s into gear.
So angstrom is built as a reth extention, i.e we run them same process, we did this as it allowed us to piggyback on the reth network for angstrom orders and consensus. We also have a lot of db reads for validation and do optimizations where we query directly from the reth db. This significantly speeds up validation processes. Without this, would be significantly slower. About 10x