He/it. A man of many talents and gifts.
IT security / LLM stuff / Occasional weirdness
DM for actual DM channels (Signal, etc.)
mlf.
Loading...
TIL I not only think the pro-copyright people are nuts. a lot of the anti-copyright people are too, apparently.
literally just pointing fable at some of my hobby projects right now like "make this better plz" and honestly it works surprisingly well
hello fellow tensorfuckers
like, we may have opinions on the overtuned safeguards and general claudeishness of the model, but it has a SIGNIFICANTLY better ability to look at an entire application end-to-end in one go. very different from the handholding you gotta do with opus to have it not go fully myopic on every feature.
now it's running a fable subagent that's just randomly maxing out my gpu from time to time and the claude code harness is very sparingly surfacing any information about what it's actually doing, but i ASSUME it's building what i asked it to
i love the future
Fable ship #2 is a much updated newsbrief that's coming out of the PoC stage and approaching proper WIP status. Much better model handling, two different size Gemma 4s used to optimize performance and quality.
tangled.org/mlf.one/news...
okay so a) that seems like a highly contentious move to make, b) it's kinda dickish too tbh, but also c) lmao gpl can suck it soooo... exciting times ahead?
Fable ship #1 of the night: some major updates to bsky-viz in how it handles embeddings and clustering. Ollama dropped in favor of llama.cpp, and Gemma 4 E4B used for cluster labeling instead of a centroid.
Still a WIP but much much better performance AND output!
github.com/mlowdi/bsky-...
