Post

O great hive mind! About a week ago I saw someone doing further experimentation on Talkie's coding abilities — I think with few-shot examples? I know the Talkie-1930 release itself does some testing of its coding ability, but I think this went beyond that? Am I hallucinating?

12/5/26, 12:16 pm3 reposts12 likes

Aaron Sterling

@aaronsterling.bsky.social

12 May

bsky.app/profile/grac...

Grace@gracekind.net

I don't expect it to convince anyone who isn't already convinced, but vintage language models solving coding problems is pretty good evidence of general reasoning ability

Ted Underwood

@tedunderwood.com

12 May

Yes, thanks! I think this is from the Talkie post itself. It's already interesting, but IIRC someone else on Bluesky took things a bit further ... Ah I found it! It was this: bsky.app/profile/apar...

austin@aparker.io

4.5% Pass@1 on SWEBench-Verified for a talkie fine-tune after 75k examples is kinda wild. github.com/RicardoDomin...

github.comGitHub - RicardoDominguez/talkie-coderContribute to RicardoDominguez/talkie-coder development by creating an account on GitHub.

Lynn Cherny

@arnicas.bsky.social

12 May

i posted about that too