Posts by PeterFalkingham@sauropods.win
(DIR) Post #AhyeStNL4b29do5gOm by PeterFalkingham@sauropods.win
2024-05-17T07:43:19Z
0 likes, 0 repeats
Peak enshittification. I've finally been 'updated' from Mail and Calendar to 'new' Outlook, and the go back toggle has been removed. This means no offline access, no shortcut to a calendar, no touch support, no unified inbox. What a complete clusterfuck. Thanks #microsoft. Oh, and thanks to our IT department for locking out any other mail client, including all clients for #linux
(DIR) Post #AhyeSv012HZ8g3C5K4 by PeterFalkingham@sauropods.win
2024-05-17T07:45:56Z
0 likes, 0 repeats
No more drafting or organising emails on my commute on the underground then:
(DIR) Post #AktUgKz13zxMJHKrjs by PeterFalkingham@sauropods.win
2024-08-12T14:49:42Z
0 likes, 1 repeats
Job ad now live for a postdoc in my lab, either biomechanics/musculoskeletal modelling, OR sediment simulation backgrounds are welcome#postdoc #paleontology #palaeontology #biomechanics #simulation https://jobs.ljmu.ac.uk/vacancy/research-fellow-palaeontology-564793.html
(DIR) Post #Al3bmGNRkA631R8PEe by PeterFalkingham@sauropods.win
2024-08-17T11:58:56Z
0 likes, 0 repeats
@drmambobob I disagree. The changes AI does to photos on a phone can be massive. Spend 10 minutes looking at selfies from 40+ mums on Facebook! (Plus pretty sure it'll recompose for you in some cases). Where as you could use an LLM to enhance ideas or provide starting points. I guess I'm saying I don't think they are that different, and both can aid/be useless equally.
(DIR) Post #Al3jYTrdbNWMQkCAXQ by PeterFalkingham@sauropods.win
2024-08-17T13:26:05Z
0 likes, 0 repeats
@drmambobob it does though. It'll (or can) use AI to do a crop within the frame, so you're not even composing. You're doing no more than saying "I want a picture of vaguely what's in front of me"
(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 #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 #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 #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 #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 #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 #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 #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 #As2vN5Hei7eYHXVnqi by PeterFalkingham@sauropods.win
2025-03-14T10:15:21Z
0 likes, 0 repeats
I thought to myself, "you know, I _would_ like to move away from microsoft Office. I haven't tried LibreOffice in a while, I should check in on it."I physically recoiled at the UI when it opened.
(DIR) Post #As2vN7MKzgaQidiZQO by PeterFalkingham@sauropods.win
2025-03-14T12:55:08Z
0 likes, 0 repeats
@mike Almost equally horrid :) Light mode clearly helps, but my god that's a lot of icons that are all very tiny and crowded. I changed Libre Office to tab view, rather than icons, because I do prefer that. But here's what a beautiful, distraction free UI for a word processor should look like*:*IMO