Posts by shajith@mastodon.social
(DIR) Post #AS9NDY7ufPGGYhtl6O by shajith@mastodon.social
2023-01-28T17:27:05Z
0 likes, 0 repeats
@simon Thank you, this was great. Have you ever thought of building a “Substack for databases” with Datasette? That sure captures the idea very neatly.
(DIR) Post #AS9NDYuTkq6qzKcXnk by shajith@mastodon.social
2023-01-29T23:12:12Z
0 likes, 0 repeats
@simon that sounds fantastic. If I had to pick between “small data for teams” and “publishing-first airtable/notion”, I’d lean more towards the latter, but I think there’s room for both those paths. Excited to see where this goes.
(DIR) Post #AShwrVWUztTwykZWTo by shajith@mastodon.social
2023-02-15T15:25:42Z
0 likes, 0 repeats
@simon I thought this Sratechery take was telling about why these large companies are still pursuing this path: https://stratechery.com/2023/from-bing-to-sydney-search-as-distraction-sentient-ai/ Essentially, leaning into the idea that this persona business might be the real selling point. Eg: “Sydney absolutely blew my mind because of her personality; search was an irritant.”
(DIR) Post #ASjzsHSZSTcxpXiNVI by shajith@mastodon.social
2023-02-16T15:14:53Z
0 likes, 0 repeats
@simon I wonder how much of the “emergent” behavior is because BingGPT is set up to run queries as part of normal operation, and is finding self references about the weird behavior from people saying things about that, and rolling it into future responses. That feels like an explanation for how much weirder it is relative to ChatGPT, and how it seems to be getting weirder each day?
(DIR) Post #ASkAraXO8pLOClqB28 by shajith@mastodon.social
2023-02-16T17:18:07Z
0 likes, 0 repeats
@simon ah good point, it certainly seemed to list sources when it explained the SolidGoldMagikarp thing, for instance. If it tags any time it does any retrieval, that should be evident. What a weird and hilarious few days this has been.
(DIR) Post #ASqKGPs6zxfStwCtwu by shajith@mastodon.social
2023-02-19T16:31:32Z
0 likes, 0 repeats
@simon That sounded plausible to me, especially the possibility of some regular product team picking up some base model (and not say something like ChatGPT) and running with it vs spending a ton of time doing RLHF or other safety work beyond a test suite.This has been a great example of prior art for what due diligence and safety work we can expect large corporations to do with any new ML capability that emerges, before unleashing some product built on it to the general public.
(DIR) Post #ASsuOi17P6HsA0aqIq by shajith@mastodon.social
2023-02-20T22:23:38Z
0 likes, 0 repeats
@simon I was saying to someone yesterday that a likelier explanation for why Google hasn’t done something like BingGPT is that they tried it or something like it in some market and it did something that terrified them (like, “multiple spontaneous cults formed around the personality”, or “people died”). I wonder how long MS ran these tests. They don’t seem to have found anything worrying enough to stop the rollout..
(DIR) Post #ASzO0PMhGMCew4m4xc by shajith@mastodon.social
2023-02-24T01:26:22Z
0 likes, 0 repeats
@simon It seems plausible that the parts of search engine query streams that aren’t currently well served would find their way onto ChatGPT, because it would make up a confident answer for every query. That would be particularly true for things that normal search engines don’t return good results for because society has decided that such things shouldn’t exist, like easy ways to geolocate people by phone number.
(DIR) Post #ASzQ9pYlROvwgw2WDA by shajith@mastodon.social
2023-02-24T01:50:18Z
0 likes, 0 repeats
@simon Yup, I get that. I’m saying that ChatGPT is independently popular enough that there’s going to be people who will get these made up answers from it to queries that currently don’t have good answers (for good reasons, in some cases). Also on “BingGPT”, do you know if it won’t generate an answer if it can’t find a search hit? As in, does it actually distinguish a generative request from a query? I don’t have access, and I don’t remember any mention of an “I don’t know” answer.
(DIR) Post #AT0qQM8Qt1D8K0Dvto by shajith@mastodon.social
2023-02-24T18:19:13Z
0 likes, 0 repeats
@simon that should result in a rather different market structure from the market structure of current cloud infrastructure with two or three big players. Seems like it would be better if this stuff is commoditized? Maybe it will run in browsers some day, like WebGL?
(DIR) Post #AT0u6Nt145QnmJkFiS by shajith@mastodon.social
2023-02-24T18:59:57Z
0 likes, 0 repeats
@simon this “tagging system” thing they’re talking about, I wonder if it is prompt engineering of the sort you would attempt to protect against prompt extraction attacks. (With the same likelihood of success..)
(DIR) Post #ATiBby1QOgXCBwthom by shajith@mastodon.social
2023-03-17T16:09:14Z
0 likes, 0 repeats
@simon For someone who had trouble understanding the langchain abstractions, this was very useful code to read, thank you!
(DIR) Post #ATn47FlYeBtKMGgP4q by shajith@mastodon.social
2023-03-20T00:38:57Z
0 likes, 0 repeats
@simon I’m kind of looking forward (with trepidation as one must) to standardized LLM capabilities being baked into browsers (say as a W3C standard with a standard API) and OS platforms. I suppose nobody has gotten a second to take a breath here to talk about any sort of standardization of interfaces with LLMs yet.
(DIR) Post #AU7HppFq72ZoqR5A3M by shajith@mastodon.social
2023-03-29T18:46:23Z
0 likes, 0 repeats
@simon I just wanted to thank you for your AI coverage, it’s been indispensable for me and I’m sure for a lot of people. How far away do you think we are from commoditized LLMs running on consumer hardware? The Llama based experiments make it sound like it is definitely plausible, but I can’t tell if there’s some insurmountable last mile hurdle.
(DIR) Post #AU7ZDput08Bx4KJgC8 by shajith@mastodon.social
2023-03-29T21:59:19Z
0 likes, 0 repeats
@simon That sounds right to me too. I’m very interested in how the nature of applications will change when the LLM is client side instead of server side. Eg: a product that’s a collection of langchain style agent pipelines, but the LLM bits of that run on clients? Built-in LLMs in web browsers that allow web apps to use that as a native capability? Like, say, how AJAX changed things. Market structure would change too: Such a world doesn’t look like an OpenAI takes all market, for instance.
(DIR) Post #AVKYRxOU0aHGU6Avx2 by shajith@mastodon.social
2023-05-05T02:16:08Z
0 likes, 0 repeats
@simon Excellent doc there. I keep thinking Google should respond to Meta’s stroke of luck with Llama by shipping a LLM browser API and local model work Chrome.
(DIR) Post #AWfkgrHwXYZi8CSuiO by shajith@mastodon.social
2023-06-14T05:34:24Z
0 likes, 0 repeats
@simon Exciting stuff. Thank you for the summary.I can't tell if Functions is mostly sugar, or if it means that I can avoid spinning up some langchain-like infra if I want to implement some task-executing feature using an OpenAI model.
(DIR) Post #AWnue55IAncKkrNygC by shajith@mastodon.social
2023-06-18T04:03:50Z
0 likes, 0 repeats
@simon started using this last week, and loving the flow of updates, thank you! Have you considered an “llm server” type command to spin up a local chatbot/copilot?