#indieweb 2023-03-16

2023-03-16 UTC
ren, mcint and [KevinMarks] joined the channel
#
[KevinMarks] https://fedidocs.org/about/
gxt__ joined the channel
#
[KevinMarks] https://css-tricks.com/snippets/css/a-guide-to-flexbox/
[tantek] joined the channel
#
[tantek] Both those links seem better for #indieweb-dev
[snarfed] joined the channel
#
[snarfed] preview of bluesky's webapp: https://staging.bsky.app/
#
Loqi [indienews] New post: http://tantek.com/2023/026/t1/indieweb-priorities-balance (from https://martymcgui.re/2023/01/27/there-are-many-opposing-forces-in-the-indieweb-that-can-seem-like-catch-22s-yet-help-clarify-priorities-and-balance-present-pragmatism-future-optimism/)
#
Loqi [indienews] New post: https://shoptalkshow.com/549/#t=23:45 (from https://martymcgui.re/2023/01/30/549-ben-ubois-of-feedbin-shoptalk/)
jjuran, vrnc, polezaivsani, geoffo, mro, bozo and Guest6 joined the channel
#
IWDiscordRelay <c​apjamesg#4492> I was thinking about adding a disclaimer like this to my website: https://notbyai.fyi/
#
[tantek] what is a disclaimer
#
Loqi A disclosure is a bit of content, typically on a home page, on an indie web site that proactively discloses some aspect about the site that the site owner wants the user to explicitly be aware of https://indieweb.org/disclaimer
#
[tantek] ^ capjamesg, have you consider other disclosures / disclaimers for your website?
#
IWDiscordRelay <c​apjamesg#4492> I have a broken privacy disclaimer: https://jamesg.blog/privacy/
#
IWDiscordRelay <c​apjamesg#4492> Markdown rendering issue 🤦
#
IWDiscordRelay <c​apjamesg#4492> And now the preview image for that article is going to be a broken markdown image for a little while 🙂
[chrisbergr] joined the channel
#
[chrisbergr] When I visit the private website/blog representing an individual, I would expect that the content was not generated by AI.
#
IWDiscordRelay <c​apjamesg#4492> I have a disclaimer on James Bot.
#
[tantek] same
#
IWDiscordRelay <c​apjamesg#4492> > This text was written by a generative text language model, trained on James' blog. Factual inaccuracies may be present.
#
IWDiscordRelay <c​apjamesg#4492> There are degrees of AI usage.
IWSlackGateway and [tantek] joined the channel
#
[tantek] re: private website/blog representing an individual, expect that the content was not generated by AI.
#
IWDiscordRelay <c​apjamesg#4492> What about assisted by AI?
#
IWDiscordRelay <c​apjamesg#4492> "This article was written with assistance from AI."?
#
IWDiscordRelay <c​apjamesg#4492> https://jamesg.blog/bot/prompt/9c17ab38-1cea-435b-a7ff-741c5cd78402
#
IWDiscordRelay <c​apjamesg#4492> I _really_ need to get those links working.
[chrisbergr] joined the channel
#
[chrisbergr] Why would you do that? When I visit your website, I want to read your thoughts on topic XY. If you do your research via CPT or whatever, I don't care as much as whether you use Wikipedia. If you have a writing weakness, and AI improves your texts, go for it!
#
IWDiscordRelay <c​apjamesg#4492> That's fair.
#
IWDiscordRelay <c​apjamesg#4492> With the current generation of models, the veracity of model results is not guaranteed, so I am thinking more about the prospect of a factual inaccuracy slipping through.
#
IWDiscordRelay <c​apjamesg#4492> But then again, Wikipedia could have a factual inaccuracy but we don't need to say Wikipedia was a source on a blog post.
#
IWDiscordRelay <c​apjamesg#4492> I see your point [chrisbergr].
[jacky] joined the channel
#
[jacky] Wikipedia can be corrected (and frankly, people should be mentioning it to continue the legitimacy of using it as a primary source) and has a means of determining where exactly it might have gotten some text from
#
[jacky] That's more difficult (if not impossible without explicit logic for it) from something generated or assisted with AI
#
[jacky] If such a disclaimer was present (for example, a journalist's site or one that's aiming to educate/inform), I'd trust it a lot more versus me finding out after the fact - especially in this age of content farming
#
[chrisbergr] AI or not, as a publisher you should still ensure the accuracy of your information. If you use AI and therefore publish false data, it's embarrassing, with or without a disclaimer.
#
[tantek] There's a kind of spectrum from suggested next words (like you see in txt and email composing UIs) and a whole suggested response (which is basically what OpenAI is generating). Lots of people use the former without disclaiming anything, and would feel deceived if people used the latter
#
[jacky] Agreed but it'd be a shortcoming to note how AI _is_ being suggested as a tool without these short-givings (nor does it have a easy recourse for sourcing). It's less of a concern if it's sourced from one's site but if it's combing the Web and then using that as some sort of answer in writing is what I'm getting at
#
[tantek] like what happens when email composing UIs start suggesting whole next sentences?
#
[jacky] I think those aren't necessarily comparable (autosuggestion / LLM text generation)
#
[tantek] they literally use similar logic
#
[tantek] "just" a matter of scale
#
[tantek] one big difference is that current txt next word suggestion UIs provide multiple options, where as email (and gDocs) next word suggestion UIs only provide a single next word suggestion
#
[tantek] when composing a txt you're making more of an explicit choice among options, whereas with email/gDocs, you're making a yes/no binary decision of whether to accept the suggested next word or not
#
[tantek] however both seem "acceptable" tools to folks (no one I know disclaims that they use such UIs to compose their text/email/gDocs)
#
[tantek] phrases/sentences are next, then maybe short paragraphs
#
[tantek] when do you cross the line into requiring a disclosure?
#
[chrisbergr] I really like to use these next-words-suggestions in my iPhone keyboard in various messengers. The suggestions are usually very good/useful.
#
[tantek] (FWIW I do not use any next word suggestions when writing my posts, since I use BBEdit which doesn't do that)
#
[tantek] I expect that there are some folks that use email/gDocs UIs to write blog posts however
#
[tantek] jacky, those next words suggestions are a trained model on the text you've previously typed with defaults based on the language you're using, which is similar to LLM text generation from prompts
[KevinMarks] joined the channel
#
[KevinMarks] LLM's are literally next word suggestion machines, but with tunable randomness to vary the path a bit, and longer look back window that the previous generation
#
IWDiscordRelay <c​apjamesg#4492> I can’t write with a next word predictor.
#
IWDiscordRelay <c​apjamesg#4492> They are distracting.
#
IWDiscordRelay <c​apjamesg#4492> Google Docs seems to have a feature like this and if it develops out further without a disable feature I shall probably have to stop using the tool.
#
IWDiscordRelay <c​apjamesg#4492> [tantek] There is a meme that we will all use AI to write emails and the result will be the equivalent of two AI bots talking to each other with a tiny bit of human intervention between.
#
IWDiscordRelay <c​apjamesg#4492> (And, this evening, James AI is going to know about that too :P)
#
IWDiscordRelay <c​apjamesg#4492> [KevinMarks] > LLM's are literally next word suggestion machines
#
IWDiscordRelay <c​apjamesg#4492> Yes, but with semantic context stored in vectors used.
#
IWDiscordRelay <c​apjamesg#4492>
#
[KevinMarks] Sure, but wordnet2 had that too, as did latent semantic analysis https://en.m.wikipedia.org/wiki/Latent_semantic_analysis
#
[KevinMarks] Google Scribe was an early version of this, and a bit more Markov chain like, iirc https://epeus.blogspot.com/2010/09/if-google-predicts-your-future-will-it.html
[schmarty] and [benatwork] joined the channel
#
[benatwork] This event, co-MC-ed by [jacky], is kicking off right now: https://www.betterplatform.net/
[aciccarello] joined the channel
#
[aciccarello] Was just about to share that!
polezaivsani joined the channel
#
[schmarty] jacky++
#
Loqi jacky has 6 karma in this channel over the last year (62 in all channels)
chenghiz_ joined the channel
#
[tantek] jacky++
#
Loqi jacky has 7 karma in this channel over the last year (63 in all channels)
#
[tantek] [benatwork] can you add that event to https://events.indieweb.org/ so it gets into the Newsletter appropriately?
#
aaronpk snap a photo if you can too, and upload it!
#
IWDiscordRelay <c​apjamesg#4492> jacky++
#
Loqi jacky has 8 karma in this channel over the last year (64 in all channels)
#
[benatwork] Done!
[tantek] and BDFOC joined the channel
#
[jacky] soooo glad that it's done lol
#
[jacky] but that was the easy part 😭
#
[KevinMarks] jacky++ wel done
#
Loqi jacky has 9 karma in this channel over the last year (65 in all channels)
geoffo joined the channel
#
[aciccarello] Agreed, the breakout session was really interesting to listen to. jacky++
#
Loqi jacky has 10 karma in this channel over the last year (66 in all channels)
#
Loqi [benatwork] has 6 karma in this channel over the last year (13 in all channels)
#
[tantek] [benatwork]++ thank you!
Zegnat, [aciccarello]1, geoffo and bterry joined the channel; polezaivsani left the channel
#
[jacky] thank you!
#
[jacky] hoping the follow up will be useful for those making things
IWSlackGateway, [jacky], [KevinMarks], bterry and gRegor joined the channel
#
gRegor [jacky], Will there be video of the betterplatform talk?
#
[jacky] Yup of each session! working on cleaning things up and uploading it (like making sure captions are synced, zoom doesn't seem to do this automatically)
#
gRegor nice!
[tantek] joined the channel
#
[tantek] looking forward to it!