Post ASZgUm1ML99YxDStAe by STL_Inquiries@indieweb.social
(DIR) More posts by STL_Inquiries@indieweb.social
(DIR) Post #ASZaafFhoqHRySbKca by TedUnderwood@sigmoid.social
2023-02-11T14:46:28Z
0 likes, 0 repeats
It hasn't sunk in yet, but we learned last week that the whole discourse of #LLM critique since November has been aiming behind the ball. #Bing and Google Search are now going to compete to address factual accuracy. Meanwhile the brainstorming function is already integrated in Word if you use the Edge browser.Image: #ChatGPT integration into Office365.
(DIR) Post #ASZapS4l5jOBS3X2My by afamiglietti79@mastodon.social
2023-02-11T14:49:08Z
0 likes, 0 repeats
@TedUnderwood crashing this out to market this fast is absolutely crazy bananas and solves no actual human problem except "Microsoft needs to boost its share price." Sad we have no functioning government, this screams for effective regulatory oversight.
(DIR) Post #ASZazdYhFlGXxdOM8e by TedUnderwood@sigmoid.social
2023-02-11T14:50:59Z
0 likes, 0 repeats
We could now work on producing a new generation of critique aimed at this ball. "What!? only five options for tone! how do they define 'professional'?" But I earnestly hope we can instead start running toward the place the ball *will* be in twelve months.
(DIR) Post #ASZb7Lx0bQjTfSRWOe by TedUnderwood@sigmoid.social
2023-02-11T14:52:23Z
0 likes, 0 repeats
@afamiglietti79 Or, alternatively, we could be guided by the principle that people get to use words pretty much however they want.
(DIR) Post #ASZbFJKSEbQ1qwCht2 by afamiglietti79@mastodon.social
2023-02-11T14:53:47Z
0 likes, 0 repeats
@TedUnderwood I'm less concerned about crashing out the Word Processor and more concerned about the Search Integration (they seem to be one and the same). Word processors just make language, but search? There are social concerns with how it can fail.
(DIR) Post #ASZbHq2pZLbltkALiK by jeffjarvis@mastodon.social
2023-02-11T14:54:16Z
0 likes, 0 repeats
@TedUnderwood Yes. I had a conversation with someone at Google a few weeks ago, who said that getting the facts right is table stakes. Amen. But that functionality is another layer on LLMs such at ChatGPT, yes? Picking the next, best word versus verifying a factual assertion are very different functions, no?
(DIR) Post #ASZbWvI4LcMGVlUDmC by TedUnderwood@sigmoid.social
2023-02-11T14:57:00Z
0 likes, 0 repeats
@jeffjarvis Yeah, it's fair to say that the underlying GPT algorithm was just predict-the-next-word and they're having to bolt something else on to do factual accuracy. But the two pieces can fit together semi-naturally: e.g. one common strategy for addressing accuracy is to do search, gather statements about X, and then put those statements into the prompt with a frame like "given this <background>, how would you answer <user question>?"
(DIR) Post #ASZc8QCoMucv6juLs8 by misc@mastodon.social
2023-02-11T15:03:46Z
0 likes, 0 repeats
@TedUnderwood which is?
(DIR) Post #ASZcYuVML76wblDLkW by STL_Inquiries@indieweb.social
2023-02-11T15:08:33Z
0 likes, 0 repeats
@TedUnderwood @jeffjarvis Any idea what 'factual accuracy' means here? I imagine it has to be variable, and will be controversial.A student might wish to accurately mirror what a prof. or textbook says. A "news" site editorialist will wish to assume the biases of the rabble she's rousing. A biologist will want facts that have been certified by a journal, and probably not just any journal in her field.It seems unlikely that one conception of 'factually accurate' will work for everyone.
(DIR) Post #ASZcdYadFKSA5Vtcx6 by TedUnderwood@sigmoid.social
2023-02-11T15:09:24Z
0 likes, 0 repeats
@misc There's room for five or six different visions at least, but I'm thinking about how I would want students to use this technology (or how I would want to use it) in writing about an inherently perspectival subject like history.
(DIR) Post #ASZcm3PsW9zA0luyrA by TedUnderwood@sigmoid.social
2023-02-11T15:10:56Z
0 likes, 0 repeats
@STL_Inquiries @jeffjarvis I think that's well put, and it's a problem that worries me because I teach subjects like history and literature that are going to need to be approached perspectivally. I'm less worried about the model making stuff up than about people trying to chain it to 2023-era Wikipedia.
(DIR) Post #ASZd4Hu5ORZi9gk3pA by shriramk@mastodon.social
2023-02-11T15:14:13Z
0 likes, 0 repeats
@TedUnderwood Neeva search has had LLM integration for months now!
(DIR) Post #ASZd8QgqfQguZ3WcN6 by TedUnderwood@sigmoid.social
2023-02-11T15:14:59Z
0 likes, 0 repeats
@shriramk Didn't know that! Excellent. There were things like you.com and perplexity.ai too.
(DIR) Post #ASZdQBZ7dfhorvNUga by tafadzwa@twit.social
2023-02-11T15:18:11Z
0 likes, 0 repeats
@TedUnderwood if they've not been working on that problem for the last half decade what have they been doing?
(DIR) Post #ASZdtu7t2fzxT9oGbg by STL_Inquiries@indieweb.social
2023-02-11T15:23:33Z
0 likes, 0 repeats
@TedUnderwood @jeffjarvis Right, these questions are very hard for some disciplines. It makes me wonder what the software just can't do for us. Can it present my very own reading of Oedipus Rex, informed by my own experience? I doubt it, though it might simulate a person doing that, and it might assist me in it somehow.I see the concern about narrowing/selecting the fact base. But maybe we're already living in that world: wikipedia exists, it's easy, and we're all lazy or busy.
(DIR) Post #ASZeHi5RSL3X5019iS by jeffjarvis@mastodon.social
2023-02-11T15:27:50Z
0 likes, 0 repeats
@TedUnderwood @STL_InquiriesYes, a fascinating problem: Media blather on about GPT et al being sentient or not but on a much simpler level, do they know what an assertion of fact is? Can they know what to verify? Then is the methodology of verification well-established and tested (e.g., is that Wolfram Alpha's specialty; can one be lashed onto the other?)?
(DIR) Post #ASZemXIzaiNUcQ3SeO by afamiglietti79@mastodon.social
2023-02-11T14:55:58Z
0 likes, 0 repeats
@TedUnderwood also, there are serious vertical integration/anti-trust concerns here. If AI writing integration becomes an expected feature of word processing and models are owned/run on centralized stacks, how can anyone compete?
(DIR) Post #ASZemXwLEPqyaSSsz2 by afamiglietti79@mastodon.social
2023-02-11T15:24:47Z
0 likes, 0 repeats
@TedUnderwood sorry, had a second to think about this, there are failure modes for a word processor to be concerned about, like leading people to write seriously flawed health advice: https://futurism.com/neoscope/magazine-mens-journal-errors-ai-health-article We want to encourage every parent, school nurse, etc to outsource writing to these things *before* we know how frequently these failures happen? When the tech solves 0 problems that need to be solved except for "Microsoft needs a competitive edge?"
(DIR) Post #ASZemYYcw4TiVCNSeu by TedUnderwood@sigmoid.social
2023-02-11T15:33:26Z
0 likes, 0 repeats
@afamiglietti79 The anti-trust questions are very valid. Monopoly or duopoly of language is intolerable. Regulating content seems to me clearly a bad idea. I know there's a whole encyclopedia of "but what about — ?" edge cases, but I feel like most of them got addressed in the period between Areopagitica and Reno v ACLU. Unless the model claims to be a doctor or is shouting fire in a crowded theater, we already had that conversation.
(DIR) Post #ASZf8yHJzEY6e2yGfI by afamiglietti79@mastodon.social
2023-02-11T15:37:29Z
0 likes, 0 repeats
@TedUnderwood we're not regulating *content* we're regulating a transmission and creation tool. We do that all the time (though less so now than in many prior periods, Fairness Doctrine RIP)
(DIR) Post #ASZfLQk3zAFUOkM7e4 by TedUnderwood@sigmoid.social
2023-02-11T15:39:44Z
0 likes, 0 repeats
@afamiglietti79 "We're not regulating content, my good sir, but a transmission and creation tool which, if unlicensed, you must concede will surely be used to spread blasphemy and sedition."
(DIR) Post #ASZffcgTILIBEAIvrs by afamiglietti79@mastodon.social
2023-02-11T15:43:22Z
0 likes, 0 repeats
@TedUnderwood I mean, there are indeed weird formal similarities between, say, vaccine misinformation and blasphemy and sedition, that's why anti-vaxxers often imagine themselves Galileo. The weird thing here, is that we're not even discussing what *people* can say, we're discussing how they will be encouraged to let machines say things *for* them.
(DIR) Post #ASZfjvbksVAN4r9R5M by STL_Inquiries@indieweb.social
2023-02-11T15:39:41Z
0 likes, 1 repeats
@jeffjarvis @TedUnderwood Because of my education in philosophy and philosophy of science, I think a simple formulation 'what's the objective fact, here?' raises a lot of deep puzzles. But I think 'objective fact' isn't the target.I wonder if anyone is trying to develop a crisp conception of 'public discourse facts' where this would name the proper target of journalistic practices of fact checking. It might be right to classify GPT etc. as a form of journalism, and hold it to that standard.
(DIR) Post #ASZfjw6b1ppucVa49o by TedUnderwood@sigmoid.social
2023-02-11T15:44:10Z
0 likes, 0 repeats
@STL_Inquiries @jeffjarvis This seems right. They're going to need to teach the models how to say "was quoted as saying" and "as reported last Tuesday in the Times," — and the models are going to say that kind of thing so much we get tired of it.
(DIR) Post #ASZg9AdI9ZI6uouGG0 by afamiglietti79@mastodon.social
2023-02-11T15:44:12Z
0 likes, 0 repeats
@TedUnderwood There are indeed creative opportunities for the tech, and folks will do weird cool stuff with it, but maybe we should have more than a month to figure it all out before its on every desktop.I'm not saying ban it, I'm just saying there's no need to go this fast.
(DIR) Post #ASZg9BQDDgQHMXnKVc by TedUnderwood@sigmoid.social
2023-02-11T15:48:37Z
0 likes, 0 repeats
@afamiglietti79 As long as we're giving prudential advice to people ("be cautious how you use it") and not making laws, I'm happy and John Milton is happy.
(DIR) Post #ASZgUm1ML99YxDStAe by STL_Inquiries@indieweb.social
2023-02-11T15:52:38Z
0 likes, 0 repeats
@TedUnderwood @jeffjarvis Hmm. Yeah! Rather, a hyperlinked text where every phrase takes you to a list of its 380,000+ web sources, arranged according to some 'relevance' score. It might also be time to start rethinking what it will mean to 'cite sources.'
(DIR) Post #ASZhegbJRkmIzgZY8W by MrBerard@pilote.me
2023-02-11T16:05:37Z
0 likes, 0 repeats
@TedUnderwood "We" being education, barely seem to be where the ball was three months ago.
(DIR) Post #ASZiAGDx3BV6SEhXYu by TedUnderwood@sigmoid.social
2023-02-11T16:11:21Z
0 likes, 0 repeats
@MrBerard I know, right?
(DIR) Post #ASZknVwYjtu6PSPZjs by shriramk@mastodon.social
2023-02-11T16:40:50Z
0 likes, 0 repeats
@TedUnderwood I can also say that nothing has made me submit more bug reports to Neeva than its LLM integration. It swings between useful, useless, horrible, and horribly funny.
(DIR) Post #ASZmPv4FsPqVHyt7Ts by afamiglietti79@mastodon.social
2023-02-11T16:58:56Z
0 likes, 0 repeats
@TedUnderwood also, if we're just doing historical analogies "I demand to use my AM transmitter in accordance with the broadest possible free speech rights! Don't tread on me with your need to manage destructive interference!"
(DIR) Post #ASZmrfmiI9OzzwuuZc by TedUnderwood@sigmoid.social
2023-02-11T17:04:01Z
0 likes, 0 repeats
@afamiglietti79 Yes, the right-of-way for cable installation is a good analogy to broadcast spectrum and I agree that government can appropriately regulate that part. But not content.
(DIR) Post #ASZnHEI18lcyFjkHHU by boris_steipe@qoto.org
2023-02-11T17:08:37Z
0 likes, 0 repeats
@TedUnderwood Exactly ... that's where the money is for MS. As per June 2022 revenues – Office 365: 32%; Azure: 40%; Windows: 28%; Ads: zero.... and this is why Goggle is looking at data integration - not search. Because this is where their money is.https://sentientsyllabus.substack.com/p/reading-between-the-lines#SentientSyllabus #ChatGPT
(DIR) Post #ASZnxWKqRBpsNnEYiG by afamiglietti79@mastodon.social
2023-02-11T17:16:15Z
0 likes, 0 repeats
@TedUnderwood but as part of the deal we did regulate content, for good or for ill. Obscenity rules, the fairness doctrine, children's advertising regulations, tobacco and rules, etc.
(DIR) Post #ASZoqZ8nhv990mqdHc by TedUnderwood@sigmoid.social
2023-02-11T17:26:13Z
0 likes, 0 repeats
@afamiglietti79 We did, starting very early, and the good/ill balance over the centuries has not been not super favorable. There may be some legitimate edge cases involving children, which we have typically found ways to address. It does not make me feel that the first thing we should do with a new tech is regulate content "until we figure out what's going on."
(DIR) Post #ASZpTSg9fPOuGPzoVE by afamiglietti79@mastodon.social
2023-02-11T17:33:14Z
0 likes, 0 repeats
@TedUnderwood I guess I was imagining reguk
(DIR) Post #ASZpjHF7x9frQfta4W by afamiglietti79@mastodon.social
2023-02-11T17:36:06Z
0 likes, 0 repeats
@TedUnderwood I guess I was imagining regulating how and where tools could be deployed, not what they could generate. We will of course get rules about what content the tools will generate, they will just be written unilaterally by Google and Microsoft and then challenged on all sides by hackers and jailbreakers. A form of discourse I find generally enjoyable, but I'm not sure is like 100% socially desirable.
(DIR) Post #ASZr1fDhe5wMxXYonw by TedUnderwood@sigmoid.social
2023-02-11T17:50:38Z
0 likes, 0 repeats
@afamiglietti79 Which returns us to antitrust and public funding of open-source alternatives: all of which I favor.
(DIR) Post #ASZsnAmsCGyiQbEGfo by ryancordell@hcommons.social
2023-02-11T18:10:23Z
0 likes, 0 repeats
@TedUnderwood @afamiglietti79 though to be fair I’m super *not keen* to fight multiple wars based on some shit ChatGPT made up
(DIR) Post #ASaKzMIM2CoqJIYM88 by LeisureClass@mastodon.sdf.org
2023-02-11T23:26:20Z
0 likes, 0 repeats
@TedUnderwood Can it do metaphor?
(DIR) Post #ASaLn5LTKMB4DkisW8 by TedUnderwood@sigmoid.social
2023-02-11T23:35:21Z
0 likes, 0 repeats
@LeisureClass Sure. In fact I often use it when I'm reaching for a metaphor and don't have the perfect one yet.
(DIR) Post #ASaYcqfCMG8fC6OmSO by LeisureClass@mastodon.sdf.org
2023-02-12T01:59:10Z
0 likes, 0 repeats
@TedUnderwood Only previously used metaphors?
(DIR) Post #ASaZ8Flv3VO82lW8kC by LeisureClass@mastodon.sdf.org
2023-02-12T02:01:31Z
0 likes, 0 repeats
@TedUnderwood And those aren’t metaphors. Those are synonyms with adjectives.
(DIR) Post #ASaZ8UzwReW9GbC3BQ by LeisureClass@mastodon.sdf.org
2023-02-12T02:01:53Z
0 likes, 0 repeats
@TedUnderwood Do you understand the difference?
(DIR) Post #ASaZ8VfPxRh7LEbApc by TedUnderwood@sigmoid.social
2023-02-12T02:04:51Z
0 likes, 0 repeats
@LeisureClass I have a PhD in English literature, which I sometimes use to recognize implicit metaphors (e.g. the fog in "foggy memory") as well as the explicit kind. And yes, it can invent metaphors.
(DIR) Post #ASajD3wdSHOyLsSgd6 by LeisureClass@mastodon.sdf.org
2023-02-12T03:57:45Z
0 likes, 0 repeats
@TedUnderwood Then how did you confuse synonyms with adjectives for metaphors? You created the metaphor - not the machine - when you told it to create a metaphor for metaphor creation using ”travel”. Travel is the metaphor. A bad one, but you gave it the answer in your request. Ask it “create a metaphor for speaking a first word in a heretofore unspoken language using travel”. Is the answer different?