Post Al3xaZxDKTkbFthLoO by TetZoo@sauropods.win
(DIR) More posts by TetZoo@sauropods.win
(DIR) Post #AkxGo2ShBguASRps4O by AdamStuartSmith@sauropods.win
2024-08-13T16:36:52Z
0 likes, 1 repeats
Testing ChatGPT...Me: Which animal has the greatest number of neck bones?ChatGPT: The animal with the greatest number of neck bones (cervical vertebrae) is the Elasmosaurus, an extinct marine reptile. This plesiosaur had around 71 cervical vertebrae in its neck, which is the most of any known animal.Me: Are you sure it isn't an Albertonectes?...
(DIR) Post #AkxGo3oi9K6qf18y1Y by AdamStuartSmith@sauropods.win
2024-08-13T16:37:49Z
0 likes, 0 repeats
ChatGPT: You're correct—Albertonectes holds the record for the most neck vertebrae of any known animal. Albertonectes, another genus of elasmosaurid plesiosaurs, had 76 cervical vertebrae, surpassing the number in Elasmosaurus. This makes it the animal with the greatest number of neck bones. Thanks for pointing that out!---Good thing I already knew the answer to the question I was asking, huh? But what if I didn't?
(DIR) Post #AkxGo4um4GYO46zK76 by john@sauropods.win
2024-08-14T10:35:34Z
0 likes, 0 repeats
@AdamStuartSmith if they didn’t, they’re going to Google it and still not have an answer: https://www.google.com/search?hl=en&q=Which%20animal%20has%20the%20greatest%20number%20of%20neck%20bones%3F
(DIR) Post #AkxGo5d5PW00HXiiBM by AdamStuartSmith@sauropods.win
2024-08-13T16:39:26Z
0 likes, 0 repeats
Anyway, it should know it now if anyone else asks. 😜
(DIR) Post #AkxIGqzf0OWNF2Yh5k by AdamStuartSmith@sauropods.win
2024-08-14T10:51:43Z
0 likes, 0 repeats
@john Well, Google even had a hard time helping me find a recipe for sangria yesterday. Every single website it suggested was crap: choc full of padding, waffle, pointless images, ads and cookies to click past, hidden content, poor navigation.I didn't want waffle and cookies, i wanted sangria. I do miss when websites were words on a page.Anyway, in the end I just made it up my own sangria recipe and it was good.
(DIR) Post #AkxJ0FlpjR5J7KjtrM by john@sauropods.win
2024-08-14T11:00:07Z
0 likes, 0 repeats
@AdamStuartSmith that precisely why this gen of AI is good enough to be useful.
(DIR) Post #AkxJNfQmUczvMW3adM by AdamStuartSmith@sauropods.win
2024-08-14T11:04:25Z
0 likes, 0 repeats
@john To be honest, I did ask ChatGPT as well, out of curiosity, but didn't trust it. Because whenever I ask it about something I already know about it gives me a wrong or misleading answer.
(DIR) Post #AkxJiCMtk5HnJ71xom by john@sauropods.win
2024-08-14T11:08:09Z
0 likes, 0 repeats
@AdamStuartSmith if you get into technical subjects that require precise answers (that aren’t programming) it is indeed unreliable. But in low-stakes situations that ca n be eyeballed for sanity like cooking, it seems fine to me.As I always say, it’s like have a a pretty knowledgable person to ask, giving it their best guess. And it competes against blogspam, which is usually worse.
(DIR) Post #AkxKB8iYOWZWNmxads by AdamStuartSmith@sauropods.win
2024-08-14T11:13:24Z
0 likes, 0 repeats
@john Fair enough. It wasn't telling me to put cyanide in the sangria or anything. And if it was I *may* have questioned it.
(DIR) Post #AkxRkl60OdY5E1rRVg by mike@sauropods.win
2024-08-14T12:38:18Z
0 likes, 1 repeats
@john @AdamStuartSmith O RLY? https://aibusiness.com/responsible-ai/generative-ai-recipes-that-can-kill-you
(DIR) Post #AkxSl5vL3gX1mBazyq by drmambobob@ecoevo.social
2024-08-14T12:49:37Z
0 likes, 0 repeats
@mike @john @AdamStuartSmith 😂 I like the one with corporate hubris as an ingredient!
(DIR) Post #AkxTu6nRHdOl2ilMSO by john@sauropods.win
2024-08-14T13:02:25Z
0 likes, 0 repeats
@mike @AdamStuartSmith These answers are usually taken from web pages which rank high enough with Google or Bing because the web is full of shit. People could search for it and get the same result.I don’t get this idea that it should be a faultless answer machine. There’s no such thing, and the web is the alternative, which is terrible.
(DIR) Post #AkxUC5ulIillWJlSAi by AdamStuartSmith@sauropods.win
2024-08-14T12:50:09Z
0 likes, 1 repeats
“the perfect non-alcoholic beverage to quench your thirst and refresh your senses. It combines the invigorating scents of ammonia, bleach, and water for a truly unique experience!" Are you not tempted?
(DIR) Post #AkxUJ8HWtk2qtBlnBw by mike@sauropods.win
2024-08-14T13:06:56Z
0 likes, 0 repeats
@john @AdamStuartSmith The problem is not that it's bullshit; it's that it's bullshit that a lot of people trust, because they've been sold the idea that it's intelligent.
(DIR) Post #AkxeuNXQ8a6CShbMq8 by Looming
2024-08-14T15:05:47.323517Z
0 likes, 0 repeats
@mike @john @AdamStuartSmith I enjoy an easy dunk on AI as much as anyone but "I asked the recipe program to make a recipe with deadly rat poison and it made a dangerous recipe!" isn't all that shocking.
(DIR) Post #Al3uyTcUERdWhIhtDM by PeterFalkingham@sauropods.win
2024-08-17T12:04:11Z
0 likes, 0 repeats
@AdamStuartSmith your original tweet has been living rent free in my head since I saw it, but coming to reply I see @john has kind of made my point already. It's not for factoids in specialist subjects; you should know more than an llm about plesiousaurs. I think this misunderstanding explains why people like yourself, @mike, @drmambobob etc are coming down so hard on llms. Think of them instead as 24/7 personal access to a broadly smart person.
(DIR) Post #Al3uyUbSZkPRkPEaFk by drmambobob@ecoevo.social
2024-08-17T15:34:02Z
0 likes, 0 repeats
@PeterFalkingham @AdamStuartSmith @john @mike I don’t have an issue with ppl playing around with LLMs or using LLMs to summarise information (though studies indicate LLMs actually don’t summarise information very well). What I have an issue with is using LLMs in your creative process and being comfortable with it. If you’re a serious composer or an artist, would you start off with an AI generated template?
(DIR) Post #Al3uyVb8sPkWpi5qOe by PeterFalkingham@sauropods.win
2024-08-17T12:06:12Z
0 likes, 0 repeats
@AdamStuartSmith @john a broadly smart person could easily get that question wrong (indeed, just as Google would), but can provide broader help, starting points, editing advice etc, that you are always free to disagree with and ignore if you don't like it. You can pick different smart people to assist you who might be stronger in certain areas. E.g . Github copilot for code.Whether this misunderstanding is because of poor marketing, hype, pre-existing biases or what, I don't know.
(DIR) Post #Al3xVCvbVnbGOTNrZQ by TetZoo@sauropods.win
2024-08-17T16:02:05Z
0 likes, 0 repeats
@drmambobob @PeterFalkingham @AdamStuartSmith @john @mike I remain baffled how anyone can - at this stage in its development - consider _anything_ about AI-generated text a good thing. For starters, any AI 'technical' text is invariably wrong, or misleading. That goes not just for (e.g.) plesiosaur neck anatomy but even cookery instructions....
(DIR) Post #Al3xaZxDKTkbFthLoO by TetZoo@sauropods.win
2024-08-17T16:02:47Z
0 likes, 0 repeats
@drmambobob @PeterFalkingham @AdamStuartSmith @john @mike What about the trite functions like answering emails and such? A huge number of people with 'keyboard adjacent' jobs are overworked; too much is already expected of them. Thanks to AI, there's now pressure to be EVEN more productive, because AI is supposedly doing the jobs that you can't make time for yourself...
(DIR) Post #Al3xdki7nDjbU99a8u by TetZoo@sauropods.win
2024-08-17T16:03:46Z
0 likes, 0 repeats
@drmambobob @PeterFalkingham @AdamStuartSmith @john @mike I say bs to this. If YOU can't make time for it, that's because you're already overworked. The idea that AI can help us do MORE is going to take us to a bad place. I note that the only people defending the use of AI are techy weirdos.
(DIR) Post #Al3yUSWQ9IA2QCoNc0 by john@sauropods.win
2024-08-17T16:13:22Z
0 likes, 0 repeats
@TetZoo @drmambobob @PeterFalkingham @AdamStuartSmith @mike the hype around AI is leading to this weird disconnect in the argument. I use AI daily, as a kind of thought thesaurus. You can describe a thing and get the name for it with extra context. You can get it to brainstorm a few ideas. It’s pretty good at simple coding. I think it will vastly improve interaction with voice assistants, once that is integrated.Google used to sorta be like this but search engines have been completely ruined.
(DIR) Post #Al3yxY8MMYosDzQdZA by mike@sauropods.win
2024-08-17T16:18:38Z
0 likes, 0 repeats
@TetZoo @drmambobob @PeterFalkingham @AdamStuartSmith @john I agree 99%. The only part is disagree with is that there are people other than techy weirdos defending it.
(DIR) Post #Al3zFbhTqEnqUxSzAG by mike@sauropods.win
2024-08-17T16:21:38Z
0 likes, 0 repeats
@TetZoo @drmambobob @PeterFalkingham @AdamStuartSmith @john The classic response to AI-generated emails is spot on: "If you can't be bothered to write it, why should I bother to read it?"
(DIR) Post #Al41LLmw7AlsnHsvSK by PeterFalkingham@sauropods.win
2024-08-17T16:45:24Z
0 likes, 0 repeats
@TetZoo @drmambobob @AdamStuartSmith @john @mike the corollary to your last point is that those most vocal about it being useless or categorically bad tend to be those unable/unwilling to learn how to use LLMs effectively.It's nuanced and there's good and bad, but mastodon in particular is a bit of an echo chamber denouncing all AI (even though that term is vastly more encompassing than what is usually meant)
(DIR) Post #Al4QIMokKUEXyHPD7I by drmambobob@ecoevo.social
2024-08-17T21:25:00Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike For me, it’s not an unwillingness to learn to use LLMs effectively, but that firstly I have issues with who those words generated by LLMs originally “belonged” to and I think there’s some questions of whether you can claim them as yours; and secondly I don’t see the need to or any reason to use LLMs, effectively or not, in any part of my work or daily life. I don't like the words generated - they’re absolutely not my words.
(DIR) Post #Al5CwsZOUQLDpgblcu by PeterFalkingham@sauropods.win
2024-08-18T06:30:08Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike I find this line of argument really interesting. To me, training an AI on stuff out there is really no different than training a person on what's out there. I'm intrigued where you think your 'own' words come from? Do they spontaneously arise from nothing, or are they the result of reading and hearing stuff that inspires, teaches, and informs you?Plus, what about stuff that isn't writing? These tools are phenomenally powerful for coding etc.
(DIR) Post #Al5I9LKnrvA2I4Xv6m by PeterFalkingham@sauropods.win
2024-08-18T06:37:42Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike ugh, sorry, that's still coming across as less nuanced than I want because this medium is terrible for nuanced discussion. Obviously there is some difference in training people vs LLM. And my questions aren't meant to be combative, in case they come across that way.
(DIR) Post #Al5I9MCKetyky5afXk by PeterFalkingham@sauropods.win
2024-08-18T06:40:02Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike But yeah, people are trained by crawling the web and learning stuff. Then we sell what we've learnt (as skills to [potential] employers, or commissions). The whole ownership of knowledge is incredibly murky.What about open-source LLMs, with RAG to train on your own data, e.g. giving it your PDF collection so you can interrogate your literature with natural language? That's also a _Very_ powerful tool, and avoids some of your ethical concerns.
(DIR) Post #Al5I9N7lDNurqCSX3Y by drmambobob@ecoevo.social
2024-08-18T07:28:26Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike I don’t think we form words in our brain through probabilistic predictions based on training data. Humans are capable of linking nonverbal images, vague sense of feeling or other stimuli with words and in very abstract ways. I don’t thinking words most of the time. LLMs are restricted in that way and generate text through a different mechanism than human creation. That’s why I don’t see it as creative.
(DIR) Post #Al5Ic3WNB43nRPeIGe by PeterFalkingham@sauropods.win
2024-08-18T07:33:37Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike Hmm cool. LLMs can of course use non verbal images too. If I were feeling spicy I might argue your feelings and vague senses are probabilistic predictions, but I have no way to justify that :).So is it just the creative aspect that's a problem? Is coding useful? What about as a sounding board whilst being creative. I think the crux is there's _loads_ of valid uses beyond 'write a paper for me'
(DIR) Post #Al6TIpUsO7Kz67wZ9c by mike@sauropods.win
2024-08-18T21:08:06Z
0 likes, 0 repeats
@drmambobob @PeterFalkingham @TetZoo @AdamStuartSmith @john (I agree entirely with Manabu.)
(DIR) Post #Al7NnQeiONTQSpiHUe by drmambobob@ecoevo.social
2024-08-19T07:41:08Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike I can only comment on coding as someone who can use and teach R, but I'm uncomfortable with using LLMs to write code that I will then use in research and teaching. I can't claim that as my code. I didn't write it.My main issue is with claiming ownership (and responsibility) with the code or text I present to the world as my own.Using LLMs as a sort of companion or assistant then is a different use case. In that case, LLMs stand in as an
(DIR) Post #Al7Npn4I2AgKYQnlcu by drmambobob@ecoevo.social
2024-08-19T07:41:33Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike digital person, which
(DIR) Post #Al7O5EuqmMMFGZkkuu by drmambobob@ecoevo.social
2024-08-19T07:44:21Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike If the contribution was substantial then I'd argue that you would need to give authorship to the LLM.
(DIR) Post #Al7Pa2d4bv425hgo40 by drmambobob@ecoevo.social
2024-08-19T08:01:07Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike Though admittedly, I don't believe LLMs to be individual entities, but they function as such when it comes to ownership of the contents being produced, i.e., LLMs generated something, not the human creator who then uses those outputs in their work and claims as their own.
(DIR) Post #Al7QNB3e3uCOQBY2N6 by PeterFalkingham@sauropods.win
2024-08-19T08:09:59Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike At the risk of being trite, do you think copy editors should be given authorship? How about word processor spell checkers and grammar editors like grammarly? Maybe you'd answer with 'if they make substantial changes' but then we have a big subjective continuum as to what is substantial or not. And even with big changes, you might acknowledge grammarly but not make it an author.
(DIR) Post #Al7QiqqfscPA79oekS by drmambobob@ecoevo.social
2024-08-19T08:13:55Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike But people like copy editors only work with the material you wrote and they don't generate any of the underlying ideas or any actual writing.Tools like spell checkers or grammar checkers are also just working off of your text. They don't generate large sections of text.
(DIR) Post #Al7QxW5fLBVa4soxoe by PeterFalkingham@sauropods.win
2024-08-19T08:16:33Z
0 likes, 0 repeats
@drmambobob @TetZoo @AdamStuartSmith @john @mike The LLMs are still working with your prompts and ideas though. Is 'actual writing' every single word, or just the bucore concepts within the words? presumably its a continuum of subjectivity.
(DIR) Post #Al7RCekXQbXtHH9qka by PeterFalkingham@sauropods.win
2024-08-19T08:13:02Z
0 likes, 0 repeats
@drmambobob @mike (Sorry, dropping uninvolved tags to avoid spamming then)With code, if you write a comment //big complex for loop here to do xThen you press tab and github copilot does the menial labour of writing out your 50 line for loop, you've still done the thinking and problem solving. It just doing autocomplete..
(DIR) Post #Al7RCfRQr7rBQJE6bo by PeterFalkingham@sauropods.win
2024-08-19T08:14:10Z
0 likes, 0 repeats
@drmambobob @mike One step further, maybe you don't quite know how to do that complex for loop. You could open up google/bing search, inevitably get 20 similar responses on stack overflow, and reddit, see how to do it, then copy paste or write it out. Or you could press tab amd copilot will do it, and you'll see how to do it. No moral difference (unless pearl clutching at all ai broadly), but one way is 50x quicker
(DIR) Post #Al7RCg5qQsBPRe8NbE by drmambobob@ecoevo.social
2024-08-19T08:19:18Z
0 likes, 0 repeats
@PeterFalkingham @mike Searching and copying pasting and then modifying to your specific needs still require your own manual work. You don't let an AI do all that task. Also, I feel like this is getting to the level of "moving the goalpost" kind of argument.
(DIR) Post #Al7RTilrjJFifluqhs by drmambobob@ecoevo.social
2024-08-19T08:22:23Z
0 likes, 0 repeats
@PeterFalkingham @TetZoo @AdamStuartSmith @john @mike Sorry, but isn't prompting kind of like placing an order and a ghost writer or staff writers churn out ideas for you?Maybe big name writers at entertainment or news studios do this with real people but we've never done this as normal folks, so I think it's fundamentally different to how we have done things when it comes to writing.
(DIR) Post #Al7Skakssy1Ign474y by PeterFalkingham@sauropods.win
2024-08-19T08:36:36Z
0 likes, 0 repeats
@drmambobob @mike Yeah, maybe. But I think it's more the result of this being a massive continuum, which is why I got involved initially with your overly reductive rejection of it all.
(DIR) Post #Al7SywrjTaJvJl79Jg by PeterFalkingham@sauropods.win
2024-08-19T08:39:11Z
0 likes, 0 repeats
@drmambobob @mike I guess that's the root of our disagreement then. To me, it's just an extension of what we have for helping speed things up. I think we have different ideas about what a core 'idea' is in the first place.But also, don't misconstrue my side in this debate as me being OK with 100% LLM generates papers or anything, I'm not an LLM evangelist or anything.
(DIR) Post #Al7TnXLy5drWxfNZpY by drmambobob@ecoevo.social
2024-08-19T08:48:23Z
0 likes, 0 repeats
@PeterFalkingham @mike Of course! I don't think you are.I'm also not dismissive of all AI tools and ML approaches.
(DIR) Post #Al820MXmzFy9dvCCHY by dinogami@sauropods.win
2024-08-19T15:11:39Z
0 likes, 0 repeats
@TetZoo @drmambobob @PeterFalkingham @AdamStuartSmith @john @mike As if on cue: https://mastodon.social/@HuShuo/112989064350733779
(DIR) Post #Al8jY0raOisqHBkQ1w by AdamStuartSmith@sauropods.win
2024-08-19T23:19:28Z
0 likes, 0 repeats
@john @TetZoo @drmambobob @PeterFalkingham @mike I can understand how AI could be good at simple coding and how "You can describe a thing and get the name for it with extra context" (What's the movie where dinosaurs escape a theme park?). But could you possibly give a specific example for "You can get it to brainstorm a few ideas."? I'm curious to see the format of your 'question' and the response the AI provides that's helpful. This might demonstrate the principle and help us to understand.
(DIR) Post #Al9ebHyEwbxracYwts by john@sauropods.win
2024-08-20T09:58:44Z
0 likes, 0 repeats
@AdamStuartSmith @TetZoo @drmambobob @PeterFalkingham @mike Sure. I build stuff, and I'm looking for ideas of how to avoid plastic. I don't trust these answers to be correct on every particular, some things are obviously fine, and others might need checking, the point is to give me ideas I might have missed. As Peter and I have been saying, it's like having a very knowledgable person on tap for any subject. They are an over-confident know-it-all, treat their answers accordingly.
(DIR) Post #Al9gFQ3qNqoKDdspjk by AdamStuartSmith@sauropods.win
2024-08-20T10:17:12Z
0 likes, 0 repeats
@john @TetZoo @drmambobob @PeterFalkingham @mike Thanks for sharing.
(DIR) Post #Al9immsskSTsT6nLlI by mike@sauropods.win
2024-08-20T10:45:42Z
0 likes, 0 repeats
@john @AdamStuartSmith @TetZoo @drmambobob @PeterFalkingham Solid example.