puffybsd @puffybsd

Fan of Clojure, F#, Prolog and Rust. Linux Fan. Fan of Deep Learning and AI. Fan of AVR programming. LaTeX and MD Fan. Neovim Fan. Keyboard Fan. Arch btw. [email protected] Joined May 2008

Tweets

6K
Followers

545
Following

482
Likes

56K

puffybsd @puffybsd

3 hours ago

Why do some people hate ollama? It's been way easier to use than alternatives.

0 0 0 8 0

puffybsd @puffybsd

3 hours ago

While reclaiming space, noticed my superglued sata Frankenstein drive was not mounted. New troubleshooting opportunity.

0 0 0 9 0

puffybsd @puffybsd

3 hours ago

Reclaimed about 350Gb in local disk just by clearing cache. Premature optimization has an inverse: indiscriminate, bloated, modern software development.

0 0 0 12 0

puffybsd @puffybsd

7 days ago

Managed to run both llama.cpp and vllm on this host this evening. Two firsts.

0 0 0 69 0

puffybsd @puffybsd

7 days ago

catpuccin.nvim had a breaking change and I'm afraid to ask.

0 0 0 29 0

puffybsd @puffybsd

a week ago

Maybe the hard work has just paid off over the years, but transformers and attention are stunningly simple.

0 0 0 25 0

puffybsd @puffybsd

a week ago

There's the big productivity hack where you close social media.

0 0 0 30 0

Need to make some progress on research and need to reproduce a paper's experiment so installing miniforge, forgetting why it was uninstalled although it was something more concrete than shun all things conda.

0 0 0 36 0

puffybsd @puffybsd

2 weeks ago

It's sad that some researchers still use conda. Thought that abomination was gone from this machine.

0 0 0 28 0

puffybsd @puffybsd

2 weeks ago

3 ai strikes today: - dumps stack overflow comments - makes up data, falsifies validation - steals name from brony fan fiction

0 0 0 22 0

puffybsd @puffybsd

2 weeks ago

Asked Gemini to help with EDA: - Identified non-existent class imbalance. - Wrong data counts. - Blamed bad counts on non-existent blank lines. - Claimed to run code to check. - Generated made-up data. Made up and repeated a single protein sequence that appears to be a linker.

0 0 0 22 0

puffybsd @puffybsd

2 weeks ago

That was a 3b qwen2.5 model, generating some plausible looking data with some functioning code though. Here's gemini just straight up making things up with tool calls (made up file counts, made up a reason, made up blank lines):