miniblog.

An AI benchmark website that tries to run comparable benchmarks regularly to discover when LLM performance is degrading:

@friend @janriemer@mastodon.technology Sounds like a good choice! I recently found this interesting benchmark table which shows Blake as a great cryptographic hash: http://cyan4973.github.io/xxHash/ (FWIW difftastic uses rustc-hash for its hashing needs.)

Could you build an entirely automated open source project? For example, suppose you're trying to make the fastest gzip library. You could write a bot that automatically accepted and released any pull requests that improved the benchmark! No human intervention needed at all.