Hacker Newsnew | past | comments | ask | show | jobs | submit | more az226's commentslogin

Even classic GPT-4 from March 2023 was quantized to 4.5 bits.


They’re afraid Cursor is going to drift away from VS Code.


the quality gap between agentic AI in Cursor and VSC is like night and day. If VSC would add LM Studio/llama.cpp support, it could win some programmers.


Is there a good way to replicate cursor tab in VS code? I don’t mean the interaction model, but how accurate it is. What model does Cursor use? It’s quite good AND fast.



I haven’t had the time to try it, but apparently a few weeks ago Copilot had a major update. Maybe someone else I can use can come in how it performs vs Cursor?


yes, I got excited after that major update and tried VSC, but it was not ironed out and clearly they got their system prompts wrong.


It's much better now, but it only works well with Sonnet 3.7.


The least speculative: PPUs will be converted from capped profit to unlimited profit equity shares at the benefit of PPU holders and at the expense of OpenAI the nonprofit. This is why they are doing it.


You missed the part where OpenAI the nonprofit gives away the value that’s between capped profit PPUs and unlimited profit equity shares, enriching current PPUs at the expense of the nonprofit. Surely, this is illegal.


Yes and no. It sounds like the capped profit PPU holders will get to have their units convert 1:1 with unlimited profit equity shares, which are obviously way more valuable. So the nonprofit loses insanely in this move and all current investors and employees make a huge amount.


But these rumors were said and talked about several days. But no big options trade was made before the actual day of announcement. That’s why it’s telling.


Meta will most likely compare against it when they release the upcoming Llama 4 reasoning model.


And the practical flops always end up lower. As an example a V100 has 125 according to spec, but the ideal case is more like 100 and non-ideal like 60.


That’s with sparsity. So it’s 29% down from 40%.


Can you export/download the models after training?


Not yet, but since people seem to want this it’s at the top of our roadmap. If you train a model now and want to get the weights, just message us and we‘ll give them to you ;)


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: