Launching DoneThat on ProductHunt today! DoneThat enables more remote work by tracking work in useful and privacy-sensitive chunks.
producthunt.com/products/don…
Been using @MistralAI large 4 a lot via it’s chat UI.
I’m very impressed. It’s been good enough to replace all my other chat UI’s.
It’s omniscience isn’t good enough to replace frontier models in phrasing, but it’s great for daily tasks
Am I the only one who doesn’t want reasoning with LLMs? Almost all modern LLMs now require reasoning.
I don’t need my LLM to solve physics for me. I just need it to tell me if a word is a noun and be wrong less and less.
1. No thinking normally does not improve the accuracy. I’ll try the new models but so far thinking actually increases the error rate of basic questions
2. Yes I know POS taggers exist. In practice LLMs outperform spacy/stanza/nltk
Comprehensible input is cardio. Spaced repetition is the gym. Traditional learning is a HIIT class. New episode on why that analogy holds up.

piped.video/s3W1Yb_sRsQ