vietvu
Joined 115 karma
- vietvuWhat is important or right today might not be tomorrow. The problem that you think today did not exist at the moment of decision. Even if you turn back time, you would likely do the same (unless with today knowledge). Why resent?
- You can just `uv venv`? Or even uvx?
- Yes, agree. At first I still use pyenv, or mise to manage python version, now that uv does that, uv is the only tool I need for everything in Python env.
- Me too. Aren't we already doing this? This is the basic I have been taught first.
- It's ok, just don't use them anymore if you don't like it. I will switch to something else.
- Quite sad to see devs nowadays has lost abilities to self-host. I know it can be overwhelming with Linux, networking, db, backup, hardware load.... However, it's not rocket science!
- I heard this opinion a lot recently. Codex is getting better, and Claude is getting worse so it's must happen sooner or later. Well, it's competition so waiting for Claude to catch up. The web Claude Code is good, but they really need to fix their quota. It's unusable. I would choose a worse model (maybe at 90%), but has better quota and usable. Not to mention GPT-5 and GPT-5-codex seems catch up or even better now.
- For Zed user, yes. They don't care (I too, didn't use Windows for years, but Windows is popular is a fact, not opinion).
What I meant was there are so many problems with Windows that the team cannot do it quickly (they post about it before: https://zed.dev/blog/windows-progress-report).
Just surprised, as I thought building GUI app on Windows must be easy right, as must be libs/frameworks already available to support that? It's just not.
- I didn't know bundling app for the most popular OS is that hard.
- and replace first, second... with 1st, 2nd...
- Or paying deepseek for slightly cheaper and worse performance than OpenAI.
- Because no one use Gemini Advanced.
- I used to have pyenv, asdf or mise to manage python versions (never use conda unless I need DL lib like pytorch). Now just uv is enough.
- LLMs like GPT are heavy and costly (and BERT are LLMs too, params can up to like 1.5B). For niche problems like classification on a small domain, BERT like models are much better, cheaper. You don't need all knowledge gen AI LLM has. I have seen many companies using DeBERTa or RoBERTa for text classification, not using GPT/LLaMA.
- So that what's Jeremy Howard was teasing about. Nice one.
- Another llama.cpp and mistral.rs? If it support vision models then fine, I will try it.
EDIT: Looks like no L3.2 11B yet.
- OK, I will wait for like a month before update.
- I felt weird reading the titles too, because maybe 2-3 years ago I have downloaded with `curl`, so why this? And also, I am pretty sure `wget` can do it and better too.
- The repo is terrible at tell us what is this about, the landing page is better: https://pico.sh, but still terrible.
- I haven't used duckdb since I got OOM on my dataset too. I think I will try again on 1.0.