Conversation

Notices

1. Embed this notice
   Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 03:33:11 JST Baldur Bjarnason

   So, yesterday I published this:

   "'The Intelligence Illusion (Second Edition): Why generative models are bad for business' – Black Friday launch sale"

   https://www.baldurbjarnason.com/2024/intelligence-illusion-2nd-ed-launch-black-friday/

   I've long resisted releasing a second edition. Despite regularly revisiting the book to revise it, I've generally then decided not to release the revision.

   The risks remain unchanged from two years ago when I first began working on the first edition.

   What changed my mind was that, put simply, everything else has changed

   • alcinnz repeated this.
   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:05:33 JST Baldur Bjarnason

     in reply to

     Tech companies have become less and less concerned about the negative impact of their products, more flagrant when it comes to anti-competitive behaviour, and are now thoroughly a political project that are pose to be in close collaboration with some of the worst political elements of the west. That's in addition to their long history of collaboration with authoritarians around the world in order to gain access to their markets.

   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:06:02 JST Baldur Bjarnason

     in reply to

     That's even without getting into the political issues. Generative models are by their nature highly political and authoritarian systems and their generative nature makes them a unique threat. That these vendors might collaborate with authoritarian governments on its own is reason enough to avoid any use of any of these systems.

     alcinnz repeated this.
   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:06:03 JST Baldur Bjarnason

     in reply to

     What little optimism the first edition had about the future potential of generative models turned out to be completely unwarranted. Even though the new edition integrates quite a bit of the research I did after the first edition, the biggest reason for the revision was to make it clear that I don't think LLMs or diffusion models can ever be "good for business" and that using them in your organisation risks pushing even a moderately well-run company into a Boeing-style death spiral

     alcinnz repeated this.
   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:06:06 JST Baldur Bjarnason

     in reply to

     And the environmental impact is, again, on its own a big enough issue to avoid these systems. The first edition had a faint hope that these systems might be put to good use in the future and in the meantime we had to take care in how we approach them. But two years later it's clear that optimism was unwarranted. These systems are just dangerous, politically, economically, and environmentally

   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:06:50 JST Baldur Bjarnason

     in reply to

     Since I don’t have the energy to reply to everybody going “WTF do you mean? Everybody uses AI”, what I mean is already in the thread:

     - That the original book had some optimism about the tech implies that the biggest problem isn’t the tech.
     - Mentioning anti-competitive behaviour and negative impacts would imply that a big source of risk are the vendors
     - Boeing-style implies that the danger is not a question of individual productivity but instead a dysfunctional system created by management

     alcinnz repeated this.
   • Embed this notice
     Ward Cunningham (k9ox@mastodon.radio)'s status on Friday, 29-Nov-2024 06:06:50 JST Ward Cunningham

     in reply to

     @baldur

     "Everybody uses AI"

     The gold-rush explanation didn't seem sufficient. One must truly believe there is something there to become so devoted.

     It's the Psychic's Con:
     https://softwarecrisis.dev/letters/llmentalist/

   • Embed this notice
     Baldur Bjarnason (baldur@toot.cafe)'s status on Friday, 29-Nov-2024 06:06:51 JST Baldur Bjarnason

     in reply to

     And that's the biggest qualitative difference between the first edition of The Intelligence Illusion and the second. The first edition was a hopeful and optimistic tool to guide businesses to safe harbour. The second edition is a defensive tool to help people talk managers out of littering their own businesses with ticking time bombs.

     https://illusion.baldurbjarnason.com/

   • Embed this notice
     alcinnz (alcinnz@floss.social)'s status on Friday, 29-Nov-2024 06:08:36 JST alcinnz

     in reply to

     @baldur Incidentally: Counter-point to "everyone uses AI"... I don't!

   • Embed this notice
     alcinnz (alcinnz@floss.social)'s status on Friday, 29-Nov-2024 06:29:23 JST alcinnz

     in reply to

     • Gerrit Eicker
     • Ton Zijlstra
     • Jeff Grigg

     @gerrit @JeffGrigg @baldur @ton Personally my argument is that we need to untangle the hype from the tech itself.

     I don't think these "generative AIs" would've gotten away with any of these harms without the hype! And its the hype itself which causes most of the harms!

     But how do we get this across to those who control the purse strings?

   • Embed this notice
     Jeff Grigg (jeffgrigg@mastodon.social)'s status on Friday, 29-Nov-2024 06:29:31 JST Jeff Grigg

     in reply to

     • Gerrit Eicker
     • Ton Zijlstra

     @gerrit @baldur @ton

     Current marketing of Generative AI as "AI" (seemingly intentionally) confuses: There is a lot of non-Generative-AI and statistical methods and heuristics that are still useful and good for large dataset initial analysis. And have been since long before the Generative AI fad.

   • Embed this notice
     Gerrit Eicker (gerrit@eicker.news)'s status on Friday, 29-Nov-2024 06:29:31 JST Gerrit Eicker

     in reply to

     • Ton Zijlstra
     • Jeff Grigg

     @JeffGrigg @baldur @ton Exactly this years-long and now further intensified use makes it puzzling to me to label LLMs, etc., as not good for business. What exactly are the arguments for that? “AI” (we all know what’s marketing and what’s not) has been proven to be extremely helpful for businesses.

   • Embed this notice
     Gerrit Eicker (gerrit@eicker.news)'s status on Friday, 29-Nov-2024 06:29:36 JST Gerrit Eicker

     in reply to

     • Ton Zijlstra

     @baldur @ton Is that an opinion, or are there arguments to support it? Are there even companies left that develop software without AI assistance? Or that don’t use AI to pre-analyze large datasets?

   • Embed this notice
     alcinnz (alcinnz@floss.social)'s status on Friday, 29-Nov-2024 06:54:01 JST alcinnz

     in reply to

     • Gerrit Eicker
     • Ton Zijlstra
     • Jeff Grigg

     @gerrit I do recognize that many people do find value in "gen-AI". I just question if the value outweighs the harms, & I wish that there wasn't so much pressure for us all to adopt it!

     I don't find the suggestions I receive for using it addresses meaningful bottlenecks in my work.

     And I'm hopeful that the uses people found for it can be achieved without all the exploitation.

     @JeffGrigg @baldur @ton

   • Embed this notice
     Gerrit Eicker (gerrit@eicker.news)'s status on Friday, 29-Nov-2024 06:54:02 JST Gerrit Eicker

     in reply to

     • alcinnz
     • Ton Zijlstra
     • Jeff Grigg

     @alcinnz @JeffGrigg @baldur @ton It’s really simple: through regular use, both privately and especially professionally.

     Generative AIs are tools. I use them daily — and multiple times a day. I can’t see any hype.

     Failing to recognize their business value, in my opinion, requires active denial and ignorance - or a massive lack of creativity.

   • Embed this notice
     alcinnz (alcinnz@floss.social)'s status on Friday, 29-Nov-2024 07:09:04 JST alcinnz

     in reply to

     • Gerrit Eicker
     • Ton Zijlstra
     • Jeff Grigg

     @gerrit History seems to *repeatedly* indicate to me that a crash is inevitable... "AI" doesn't have a good track record that way.

     But I see there's no arguing with you!

     @JeffGrigg @baldur @ton

   • Embed this notice
     Gerrit Eicker (gerrit@eicker.news)'s status on Friday, 29-Nov-2024 07:09:06 JST Gerrit Eicker

     in reply to

     • alcinnz
     • Ton Zijlstra
     • Jeff Grigg

     @alcinnz @JeffGrigg @baldur @ton The technology is here. In many forms. It’s not going to disappear. Not using these tools is almost certainly a self-fulfilling prophecy. Luddites have always lost. Increased productivity is a fundamental business and even societal value — and it overcomes almost any kind of limitation.

   • Embed this notice
     betalars :antifa: (betalars@chaos.social)'s status on Friday, 29-Nov-2024 08:10:00 JST betalars :antifa:

     in reply to

     • alcinnz

     @alcinnz it's so remarkably useless. I have tried to use it some times and it has never been more helpful than a rubber duck.

     the only productive thin I use the current batch of machine-learning-tools for in day-to-day use is up-scaling, de-noising and grammar checking.

     Other than this, they just made my life more difficult due to all the additional garbage on the internet and due to the amount of artists that have taken their work offline.