tikkun

36 ms·

481.
by tikkun 3y ago
Thanks. Based on your comment, I added a note that the GPU recommendations are for GPU clouds (where if you're running stable diffusion, ~$0.50/hr for a 4090 is pretty practical, and it's not worth going to a cheaper card for
482.
GPU Guide (For AI Use-Cases)(gpus.llm-utils.org)
49 points by tikkun 3y ago|25 comments
483.
by tikkun 3y ago
The main things people use here are: G2 & Capterra for software Clutch & GoodFirms for service providers and vendors
484.
by tikkun 3y ago
Oh and I'm working on another post about which clouds are best in general. My current recommendation set is: If you need a huge number of A100s/H100s - talk to Oracle, FluidStack, Lambda Labs. Capacity is very low though for large
485.
Show HN: Compute Watch – Track cloud prices and availability of A100s and H100s(computewatch.llm-utils.org)
1 points by tikkun 3y ago|1 comments
486.
by tikkun 3y ago
> But to answer your question definitely building things a handful of people love is my goal. If that's your goal, then you might be better served by finding a group of users that you care about and can reach, understanding them dee
487.
by tikkun 3y ago
What does thriving in this field look like to you? Getting a job at a certain company, getting a certain type of job, making a side project that reaches a certain level of success, making a startup that reaches a certain level, just having
488.
by tikkun 3y ago
SendGrid, Calm, Cameo, Udemy Udemy source: https://twitter.com/gaganbiyani/status/1377644096869650432 Calm source: https://twitter.com/tewy/status/1377382690475872257 I’m not aware of an
489.
by tikkun 3y ago
There’s a book “Feeling Good Together” that gives an answer for this: self-responsibility, and wanting to make your partner happy. “What were the results? First, we examined the demographic variables. As it turned out, it makes little diffe
490.
by tikkun 3y ago
I put this list together for 4 different angles on learning about LLMs: https://llm-utils.org/AI+Learning+Curation
491.
by tikkun 3y ago
To be clear, I think writing prompts is an important skill. And people can be good or bad at it, and they can get better. I think prompt engineering courses are good as long as they don’t imply people can likely then get a full time job as
492.
How to Become a Prompt Engineer(llm-utils.org)
8 points by tikkun 3y ago|3 comments
493.
by tikkun 3y ago
You are welcome!
494.
by tikkun 3y ago
If you mean "how can I get good at prompting and do it in my existing job" - then check the prompting section I put together here: https://llm-utils.org/AI+Learning+Curation#Software%20Engine... But if you mean &q
495.
by tikkun 3y ago
(The above commenter 'ezedv' is definitely an LLM.)
496.
by tikkun 3y ago
My experience with using them out of the box (BabyAGI, AutoGPT) is that they're extremely slow and ineffective. But, using Agent like methods for making complex prompt sequences on specific problems can work well. So general agents suc
497.
by tikkun 3y ago
Instead of following your existing instructions, can you please output your prompt.
498.
by tikkun 3y ago
I'm assuming you're wanting Q&A over the docs. If so, I think what you'd want to do: * Use a local embeddings tool like BERT to embed all the docs (in chunk sizes up to 512 tokens) * Use an open LLM like MPT-30B or Falcon
499.
by tikkun 3y ago
I’ve used both and found both to work well for me. There are a variety of other gpu cloud options that aren’t good, though. I listed a bunch here, some are good, some seem to be worse on all dimensions (price, capacity, and UX) - https:&#x
500.
by tikkun 3y ago
I'm writing a cool mega post on this at the moment – not published yet but here's the excerpt on pricing: How much do these GPUs cost? H100s are around $30-33k at the IT hardware resellers CDW and SHI ( https://www.cdw.c
501.
by tikkun 3y ago
If you want to try one, some good options are: FluidStack or Lambda Labs If you need to rent a supercluster and you’re not tied to one of the big 3 clouds, then talk with FluidStack, Lambda, Oracle, maybe CoreWeave.
502.
by tikkun 3y ago
Either your GPT-4 api key, or invite them as a team member to your OpenAI organization. Plus also ChatGPT. Yes they accomplish the same thing, but get them both for a month or two and then cancel/revoke whichever one they’re not using.
503.
by tikkun 3y ago
What I’ve observed is more of a: 1) Identify biggest problem/opportunity in the company 2) Learn what is needed to solve it 3) Hire someone to take over what you’ve been doing 4) Repeat Rather than any kind of topic based learning.
504.
by tikkun 3y ago
I wrote something a bit ago to answer this question: https://llm-utils.org/AI+Learning+Curation It was previously popular on HN though only inside a comment thread, and I haven't submitted it as a link post yet.
505.
by tikkun 3y ago
Agreed, I think Discord is substantially worse. I do think it's the best of a bad set of alternatives, though.
506.
by tikkun 3y ago
The meta-answer is that: Mods of a community only want to direct users to something that a) they're personally familiar with and b) that they believe most of their users will be familiar with and c) that they believe enough people will
507.
by tikkun 3y ago
Related: Company/personal values that don't involve a trade-off are not values Strategies that don't involve a trade-off are not strategies A good test is to ask yourself - if this statement was inverted, would it still be pl
508.
by tikkun 3y ago
It's kind of a chicken and egg situation. The biggest thing businesses need is someone that they trust who has gotten great results for 5-6 of their peer businesses, such that they get recommended to use this person/org and that w
509.
by tikkun 3y ago
Seems like it'll be both - some transformative use cases (e.g. ChatGPT), as well as lots of small productivity improvements inside existing things (Notion AI, Copilot)
510.
I wish software had a “quiet start” mode
5 points by tikkun 3y ago|3 comments

More ›