One thing to recall from @hdevalence.bsky.social 's work: @lmao.bsky.social, @pinged.bsky.social & @alexhevans.bsky.social showed that just encrypting a continuously updated CFMM (e.g. via FHE) fails to provide real privacy, since the very structure of a live, convex invariant leaks information.

At ACM #Facct2025, learning that LLMs face diminishing returns & comparatively worse task-specific performance when fixating on a larger scale. arxiv.org/abs/2409.14160 (paper by ‪@sashamtl.bsky.social‬
‪@gaelvaroquaux.bsky.social‬ & ‪@meredithmeredith.bsky.social‬)

I spoke at @protocol_berg on Modern Multi-Proposer Consensus. Consensus researchers know that algorithms like Mysticeti and Cordial Miners now elegantly blend HashGraph's virtual voting and @brynosaurus's Threshold Logical Clocks on a DAG. High throughput, low latency, the dream.

Ever wonder exactly where your Rust code spends its time? I wrote a note on how to light up perf bottlenecks with Tokio’s tracing and the lightweight tracing-texray layer.

I might just have stumbled on the same functionality for Rust, and it was indeed quite simple to do! Have a look at github.com/huitseeker/l... : hopefully this is useful for someone!

Thanks @filippo.abyssdomain.expert for the inspiration and @simonwillison.net for the llm tool!

#rustlang #llm

New blog post ! Byzantine Consistent Broadcast (BCB) has re-emerged as a path toward ultra-fast digital asset transfers. Projects like pod and delta claim major performance gains through parallel execution—but what's the catch?

1/7

The recording of my talk at the Agglayer Summit in Bangkok is out ! I tried to outline an approach to ZK-proof verification that combines parallel processing, sharded state, and 500ms latency. Here's to making proof verification more accessible & efficient in 2025! #ZKProofs