tikkun

30 ms·

1. 391.▲
   by tikkun 3y ago

   A possible prescription for AMD regarding AI and CUDA: 1) Open-source driver development as mentioned in this post 2) Set up 24/7 free software tech support on discord. Maybe for all use cases, maybe only AI use cases. Do the tech supp
2. 392.▲
   by tikkun 3y ago

   On demand cost is $12.32/GPU/hr. For comparison, FluidStack and Lambda Labs are offering them at $2/GPU/hr on-demand (possibly at a loss or close to it - though AWS isn't doing that - their investors expect signific
3. 393.▲
   A Disgusting Playbook(geohot.github.io)

   6 points by tikkun 3y ago|0 comments
4. 394.▲
   by tikkun 3y ago

   Nice! What kinds of companies are generating enough AI images that they need automated image assessment? Is it only AI companies doing image galleries, or are there some more enterprise style companies
5. 395.▲
   by tikkun 3y ago

   For enterprise sales, definitely Founding Sales by Kazanjy and the entrepid how to sell guide. (Foundingsales.com + https://www.entrepidpartners.com/how-to-sell-guide ). I’d probably say start with a quick read through of th
6. 396.▲
   by tikkun 3y ago

   Nice, I think it makes sense to me. So you'd provide it for example with a csv with prompt and expected completion, and then a test file, and then you can run that test file and see how it does on those sample prompts.
7. 397.▲
   by tikkun 3y ago

   This looks great! What vendor did you end up going with to do the printing and shipping?
8. 398.▲
   by tikkun 3y ago

   https://openai.com/blog/gpt-4-api-general-availability
9. 399.▲
   by tikkun 3y ago

   I think I realized what you mean. Mostly manually, plus I built a few utilities to help. e.g. I have a browser extension where I can save system prompts, and a few python scripts that make it easy for me to get the contents of files nicely
10. 400.▲
    by tikkun 3y ago

    What do you mean? I don’t understand what you’re asking
11. 401.▲
    by tikkun 3y ago

    Lots of reasons. Brand. People prefer things they’re familiar with and things that people they personally know also use, they intuitively feel that they’re less likely to have a bad experience with those. And that’s generally true. Many mor
12. 402.▲
    by tikkun 3y ago

    Yes: https://gpus.llm-utils.org/stable-diffusion-apis/
13. 403.▲
    by tikkun 3y ago

    There are a few midjourney apis. Unofficial ones. See [1]. That said, I wouldn't recommend using one of them, they're not officially supported. Likely a better solution to see how you can get your style working with stable diffusi
14. 404.▲
    by tikkun 3y ago

    > community, the result of a years long collaboration with Accenture Hmm. Not a promising start to the news - Accenture is strongly associated to me with slow moving bloated enterprise development. And then there's no github repos
15. 405.▲
    by tikkun 3y ago

    For me: the only one I use daily is GPT-4. I use it in the official GPT Playground, rather than ChatGPT, as I prefer the interface, increased privacy, and no message limits.
16. 406.▲
    Ask HN: There are lots of GPT-powered coding assistants. Any you're using daily?

    5 points by tikkun 3y ago|4 comments
17. 407.▲
    by tikkun 3y ago

    One thing that could help: Add to the top of the readme something that basically shows "put in this / do this" and "get this" Kind of like a before and after. Basically, I'd recommend exposing 2-3 of the exampl
18. 408.▲
    Nvidia H100 GPUs: Supply and Demand(gpus.llm-utils.org)

    4 points by tikkun 3y ago|0 comments
19. 409.▲
    by tikkun 3y ago

    I’ve tried both BabyAGI and AutoGPT and they didn’t work well for me. Has anyone here had success with them on something useful or interesting that base GPT-4 failed at?
20. 410.▲
    by tikkun 3y ago

    I posted this on HN a while back: https://news.ycombinator.com/item?id=36035943 as a comment. I think it deserves a link post!
21. 411.▲
    For Potential PhD Students(math.stanford.edu)

    2 points by tikkun 3y ago|2 comments
22. 412.▲
    Why GPT-3.5 is (mostly) cheaper than Llama 2(cursor.so)

    1 points by tikkun 3y ago|1 comments
23. 413.▲
    by tikkun 3y ago

    Possible - here's a counterpoint: ChatGPT will make it easier to make new programming languages because all you'd need to do is give an AI your docs, and it'll be able to figure out how to re-write things in the new language.
24. 414.▲
    by tikkun 3y ago

    I have a list of tools here: https://llm-utils.org/List+of+tools+for+making+a+%22ChatGPT+... There are indeed quite a few startups that do this. Note that these are all 'retrieval-augmented generation' tools rathe
25. 415.▲
    by tikkun 3y ago

    Side note: It'd be nice if there was a paid Discourse alternative that was upvote/downvote style and also really well designed. Discourse is nice but I prefer upvote/downvote, I've seen other reddit clones but I prefer m
26. 416.▲
    by tikkun 3y ago

    In addition to the other link posted, see also this for Show HN: https://news.ycombinator.com/showhn.html Waitlists are a no-go for 'Show HN' launches HN likes things that are interesting and generally prefers mor
27. 417.▲
    by tikkun 3y ago

    Thanks!
28. 418.▲
    Ask HN: What are your small repos that you've been meaning to open source?

    2 points by tikkun 3y ago|0 comments
29. 419.▲
    Ask HN: What are the blog posts you've been meaning to write?

    1 points by tikkun 3y ago|1 comments
30. 420.▲
    by tikkun 3y ago

    My experience was that during the days of davinci-003 as state of the art, I put a lot of effort into saving prompts. In the first month or so of using GPT-4 as my daily LLM, I kept saving prompts. But since then, I now rarely bother puttin