ruairidhwm
Joined 11 years ago
Karma
337
Submissions
223
Member since
Jan 25
Recent submissions
Comment · 3 months ago
Absolutely - not a criticism. I think it's great to try out projects :)
Comment · 3 months ago
Nice! Reminds me a little of https://github.com/gchq/CyberChef
Comment · 4 months ago
A dumb vibe-coded GTA III parody set in my city of Glasgow: https://glasgow-brawler.vercel.app/
Comment · 5 months ago
I managed to find that Haiku outperformed Sonnet on some tasks...don't want to blog spam but if anyone is interested: https://www.ruairidh.dev/blog/sonnet-4-6-drops-format-rule-o...
Comment · 5 months ago
I recently made halftrace - point it at your OpenAI/Anthropic/LangSmith agent logs and it tells you what shape your agent's rule-following has, why, and what to try. I built it to try and find smooth decay over trajectory length on Claude.
Comment · 6 months ago
Ricoh GRIII / IV is absolutely perfect for this imo
Comment · 6 months ago
I guess just a spell-check in the repo? But yes, I'd imagine that they have an effect. Even running the same input twice is non-deterministic.
Comment · 6 months ago
I literally just posted a blog on this. Some seemingly insignificant words are actually highly structural to the model. https://www.ruairidh.dev/blog/compressing-prompts-with-an-au...
Comment · 7 months ago
Also seen this several times.
Comment · 8 months ago
To be fair, I suspect he doesn't make his own tea xD
Contact - ruairidhwm[AT]gmail[DOT]com