Post AVomgSaqf0EjHyyPyK by simon@fedi.simonwillison.net
(DIR) More posts by simon@fedi.simonwillison.net
(DIR) Post #AVomgSaqf0EjHyyPyK by simon@fedi.simonwillison.net
2023-05-19T16:17:17Z
0 likes, 0 repeats
How many LLM chat tools have you tried at least once?I mean things like ChatGPT and Bing and Bard(Trying to gauge how many people haven't yet experimented in this weird new space)
(DIR) Post #AVomvX2SPnHtQ54eg4 by daniel@social.telemetrydeck.com
2023-05-19T16:19:28Z
0 likes, 0 repeats
@simon I wish to learn!!!!
(DIR) Post #AVonLP68YpIeJz8F6W by luciano@parens.social
2023-05-19T16:24:30Z
0 likes, 0 repeats
@simon ChatGPT + own hardware but no others 😅
(DIR) Post #AVonpQu8PQ4CbxpwuG by simon@fedi.simonwillison.net
2023-05-19T16:29:44Z
0 likes, 0 repeats
If you're holding off for privacy reasons, have an OK laptop (I think 8GB+ of RAM) and can spare a few GBs of bandwidth, one option could be to try this model that runs entirely in your own browserYou can even start it running and then turn WiFi off before you start playing with it https://simonwillison.net/2023/Apr/16/web-llm/
(DIR) Post #AVoo36MRaMz4x2b17o by lukas@pdx.land
2023-05-19T16:29:45Z
0 likes, 0 repeats
@simon oh wait it got cut off. I'm in the I'm running on my own hardware camp!
(DIR) Post #AVooamlSx1eW1IMU0u by 0x56@mastodon.social
2023-05-19T16:32:16Z
0 likes, 0 repeats
@simon i said "none", tho i can't rule out unintentional use at this point (e.g. getting trolled by a chatbot-powered social media account)i have a vague but semi-functional understanding of the underlying technology, and i've "delegated" my data gathering in a sense, to the people with more intuition for/experience with eliciting & describing interesting behavior from models (mostly because it's too easy to self-nerd snipe)
(DIR) Post #AVoomOhoEAXSsSBjH6 by jonn@social.doma.dev
2023-05-19T16:34:39Z
0 likes, 0 repeats
@simon I desperately need "two, both from #OpenAI" option. Was tempted to just say "one".
(DIR) Post #AVopRhMlkiPUcp2aXI by kellogh@hachyderm.io
2023-05-19T16:48:09Z
0 likes, 0 repeats
@simon to be fair, this is biased toward people who follow you for LLM shop talk
(DIR) Post #AVos7fdYltylKi2IJk by jafo@inuh.net
2023-05-19T17:17:48Z
0 likes, 0 repeats
@simon Cool, I'll have to point my son at that (I introduced you two at PyCon), this morning before school he was trying to get LLaMA 3B running on his new Macbook Air. Another friend recommended he try Dali.
(DIR) Post #AVosm7oxP7J0KKeoHQ by mcc@mastodon.social
2023-05-19T17:24:11Z
0 likes, 0 repeats
@simon I have done extensive exploration with self-trained markov chain models in the past. I have avoided LLM like the plague— including dropping use of products that add optional LLM integration— and do not intend to change this. Even if I could avoid concerns about privacy/API data retention, which I can't*, I have yet to encounter one of these large models where I would describe the training-data collection as ethical.* Even many of the downloadable models have weird license encumbrances.
(DIR) Post #AVovXtL4kbRlDiaKIK by rose_alibi@post.lurk.org
2023-05-19T17:56:17Z
0 likes, 0 repeats
@simon i have done a lot of markov chain experimentation, and ChatGPT et.al. just isn't that interesting to me as a result of having gone to the point of boredom with that.
(DIR) Post #AVovwhgBEoBIIYHtia by triadne@toot.wales
2023-05-19T18:00:17Z
0 likes, 0 repeats
@simon I tried both google and bing to see if they could answer a question about a character without spoiling the story the character is from. None of them are able to use reasoning to that degree so I don't see much point to them.
(DIR) Post #AVpFRZkwdzyFpax0oS by dr2chase@ohai.social
2023-05-19T21:39:27Z
0 likes, 0 repeats
@simon I am epsilon away from buying a new (Apple) laptop, have money to burn, and am wondering if I just max it out completely, figuring that something large and interesting will appear in the next year. And failing that, it will be awesome for video-editing.
(DIR) Post #AVpGNTibFbD5tvSpZg by simon@fedi.simonwillison.net
2023-05-19T21:48:52Z
0 likes, 0 repeats
@dr2chase I got a 64GB M2 recently - no regrets, but frustratingly there's still tons of ML model stuff that wants a Linux or Windows PC with an NVIDIA graphics card because of PyTorch and CUDA
(DIR) Post #AVpgumxXvLIl76V4i0 by eichin@mastodon.mit.edu
2023-05-20T02:47:17Z
0 likes, 0 repeats
@simonI care about provenance and attribution, and have done 2 ML-heavy startups; I read your LLM work in particular because you're one of the few experimenters who doesn't seem to have been fooled by it, and are the most likely to publish something that could change my mind.
(DIR) Post #AVqDVYC47pnaJvgv32 by Volker@fosstodon.org
2023-05-20T08:51:29Z
0 likes, 0 repeats
@simon Hey Simon. Also played with Vicuna 7b a while back and was similarly amazed. Amazing how much knowledge can be compressed into a few GB of weights. As regards to tools for LLMs, have you seen anyone giving LLMs access to linters and static type checkers to improve the code they generate in an iterative manner? Seems like low-hanging fruit but I couldn't find anything of that sort when I searched.
(DIR) Post #AVqJrNOjXl5OvyPJDs by albertcardona@mathstodon.xyz
2023-05-20T10:03:29Z
0 likes, 0 repeats
@simon @dr2chase The framework laptop gives you an AMD or Intel CPU plus lots of RAM (64GB or even more) for very reduced prices. It’s a great deal. Runs linux and windows well.
(DIR) Post #AVqii6SVeQOEe7qeno by charlesroper@indieweb.social
2023-05-20T14:39:22Z
0 likes, 0 repeats
@simon @dr2chase You can get mind-bogglingly good hardware from https://www.pcspecialist.co.uk/ if you need something high spec for a reasonable price. They make great kit - I’ve got a new laptop arriving from them soon.
(DIR) Post #AVqw8Xl84QdYFYUmjA by simon@fedi.simonwillison.net
2023-05-20T17:12:28Z
0 likes, 0 repeats
@Volker ChatGPT Code Interpreter gives the LLM the ability to run Python code directly, and if it gets an error it automatically fixes the bug and tries again
(DIR) Post #AVqwMNIlkPlBs3pFdw by bhawthorne@infosec.exchange
2023-05-20T17:13:13Z
0 likes, 0 repeats
@simon @Colarusso I have Draw Things on my phone and have multiple text-to-image models loaded. With a couple of gigabytes of model data, my phone can do image creation with no network connection. I’ll have to see if there is an LLM that will run on my phone.
(DIR) Post #AVsreUASqbqmvJg492 by not2b@sfba.social
2023-05-21T15:31:41Z
0 likes, 0 repeats
@simon The difficulty with surveys like this is that they will mostly be responded to by people who follow you, and those folks are a lot more likely than the general public to have an interest in LLMs, positive or negative.
(DIR) Post #AVtCvs7L9efVA7LdEu by simon@fedi.simonwillison.net
2023-05-21T19:29:30Z
0 likes, 0 repeats
@not2b I found it pretty surprising that, given the early-adopter nature of my following here, 1/5 people haven't tried an LLM tool at all yet