Conversation

Notices

1. Embed this notice
   Tats 💥 (tattooed_mummy@beige.party)'s status on Tuesday, 07-May-2024 20:16:32 JST Tats 💥

   But is AI a scam?

   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Tuesday, 21-May-2024 02:21:31 JST Sick Sun

     in reply to

     • John Fediverse
     @mrsaturday @Tattooed_Mummy every good picture you see from a local AI probably had to be curated by a human sifting through all the garbage ones that got produced. I've had better luck with the corporate ones but they still completely mess up for me when I ask it to do something like show me a person sitting in a car, the car is completely messed up. also the corporate ones are so censored I often can't make it do anything. I tried to get the bing one to do something recently and it argued with me constantly, things like military uniforms it wouldn't do.
   • Embed this notice
     John Fediverse (mrsaturday@shitposter.world)'s status on Tuesday, 21-May-2024 02:21:33 JST John Fediverse

     in reply to
     @Tattooed_Mummy I can tell you this, watching Stable Diffusion rapidly get left behind closed corporate censored picture generation is really depressing
   • Embed this notice
     :blahaj: Why Not Zoidberg? 🦑 (whynotzoidberg@topspicy.social)'s status on Sunday, 26-May-2024 22:33:27 JST :blahaj: Why Not Zoidberg? 🦑

     in reply to

     • Tom Bellin :picardfacepalm:
     • root42
     • James Wells

     @nikatjef @root42 @tob @Tattooed_Mummy They are unable to tell the truth. All they can do is mimic the truth with no actual control that they DO tell the truth. Relying on AI for facts is not a good idea, since they cannot determin what fact is.

     As someone else put it on here, we have burned billions of dollars, and are destroying the environment with enormous data centers burning drinking water and electricity, to produce tech that makes computers pretend they can't do math.

   • Embed this notice
     James Wells (nikatjef@mastodon.acm.org)'s status on Sunday, 26-May-2024 22:33:28 JST James Wells

     in reply to

     • Tom Bellin :picardfacepalm:
     • root42
     • :blahaj: Why Not Zoidberg? 🦑

     @WhyNotZoidberg @root42 @tob @Tattooed_Mummy
     And anyone who thinks a LLM is a RNG has even less clue.

     No one is claiming they are sentient or that it understands what it is say, but the fact is that LLM systems are our-scoring humans on various collegiate tests, they are diagnosing diseases that medical professionals are missing.

   • Embed this notice
     :blahaj: Why Not Zoidberg? 🦑 (whynotzoidberg@topspicy.social)'s status on Sunday, 26-May-2024 22:33:29 JST :blahaj: Why Not Zoidberg? 🦑

     in reply to

     • Tom Bellin :picardfacepalm:
     • root42
     • James Wells

     @root42 @nikatjef @tob @Tattooed_Mummy

     But it is not intelligent. It can randomly put together responses based on training and input, nothing it says is anything but random words put together to SOUND like a human. There is no medical or psychological knowledge in there, just random responses based on the data that it has been fed with while being trained to sound as if it knows anything.

     People who mistake an RNG what-if code with "it is almost sentient" has no clue.

   • Embed this notice
     root42 (root42@chaos.social)'s status on Sunday, 26-May-2024 22:33:32 JST root42

     in reply to

     • Tom Bellin :picardfacepalm:
     • James Wells

     @nikatjef @tob @Tattooed_Mummy My point. And memory is already on its way and will be another game changer for current models. I do think that AI will still do great strides, but I don't see the point or better: I see a lot of drawbacks and dangers. We totally lose track that whatever we do we should do it for humanity, not "because".

   • Embed this notice
     James Wells (nikatjef@mastodon.acm.org)'s status on Sunday, 26-May-2024 22:33:33 JST James Wells

     in reply to

     • Tom Bellin :picardfacepalm:
     • root42

     @tob @root42 @Tattooed_Mummy
     I am not so sure about that. 30 years ago Eliza was about the best you could get for "therapy" chat bots, now there are things like Replika, Elomia, Mindspa, etc. We have come a long ways in the last 30 years. Note that I am not saying good, just that we have come a long way.

   • Embed this notice
     Tom Bellin :picardfacepalm: (tob@hachyderm.io)'s status on Sunday, 26-May-2024 22:33:34 JST Tom Bellin :picardfacepalm:

     in reply to

     • root42

     @root42 @Tattooed_Mummy those "next steps" are no closer to becoming reality today than they were 30 years ago.

   • Embed this notice
     root42 (root42@chaos.social)'s status on Sunday, 26-May-2024 22:33:36 JST root42

     in reply to

     @Tattooed_Mummy we are trying to replace humans with humanlike computer programs. What could possibly go wrong? Calling it now: we will soon have equivalents of babysitters, teachers and psychologists for AIs. Assuming that next steps will be memory and self correction in AI development.

   • Embed this notice
     Tom Bellin :picardfacepalm: (tob@hachyderm.io)'s status on Sunday, 26-May-2024 22:33:41 JST Tom Bellin :picardfacepalm:

     in reply to

     • root42
     • James Wells

     @root42 @nikatjef @Tattooed_Mummy LLM is a cul-de-sac. An interesting party trick and toy.

     The "game changers" aren't anywhere except in peoples' fantasies.

     What you have instead is the masked work of humans. e.g. Nigerians working for $2/hour filtering through thousands of horrific images in the vain hope that Midjourney won't produce snuff-porn.

   • Embed this notice
     Hisham (hisham_hm@mastodon.social)'s status on Sunday, 26-May-2024 22:33:45 JST Hisham

     in reply to

     @Tattooed_Mummy excellent summary to share with family members

   • Embed this notice
     :blahaj: Why Not Zoidberg? 🦑 (whynotzoidberg@topspicy.social)'s status on Sunday, 26-May-2024 22:33:50 JST :blahaj: Why Not Zoidberg? 🦑

     in reply to

     • Tom Bellin :picardfacepalm:
     • root42
     • James Wells

     @tob @nikatjef @root42 @Tattooed_Mummy The over here AI that was used (as a test, double checked at every point by human pharmacists) for pharmacies to identify and recommend correct use of medication was very quickly taken off line since it turned out to be severely wrong on many occasions, often deadly so.

     The test was obviously a failure.

   • Embed this notice
     Tom Bellin :picardfacepalm: (tob@hachyderm.io)'s status on Sunday, 26-May-2024 22:33:51 JST Tom Bellin :picardfacepalm:

     in reply to

     • root42
     • :blahaj: Why Not Zoidberg? 🦑
     • James Wells

     @nikatjef @WhyNotZoidberg @root42 @Tattooed_Mummy I suspect if you go back and check your data points you'll find they're misleading.

     Acing exams is actually a really good case for testing our assumptions about these systems.

     After all, the whole idea is that they have access to the entire internet. How could they do anything but ace the tests.

     It turns out that LLMs can sometimes do kind-of-ok on tests given a very carefully constructed environment.

     So what gives?

   • Embed this notice
     Tom Bellin :picardfacepalm: (tob@hachyderm.io)'s status on Sunday, 26-May-2024 22:33:56 JST Tom Bellin :picardfacepalm:

     in reply to

     • root42
     • :blahaj: Why Not Zoidberg? 🦑
     • James Wells

     @WhyNotZoidberg @nikatjef @root42 @Tattooed_Mummy Good thing history doesn't have anything to say about civilizations collapsing because of the misallocation of investment based on the whims of a disconnected and increasingly bizarre class of elites.

   • Embed this notice
     :blahaj: Why Not Zoidberg? 🦑 (whynotzoidberg@topspicy.social)'s status on Monday, 27-May-2024 10:57:14 JST :blahaj: Why Not Zoidberg? 🦑

     • clacke
     • Tom Bellin :picardfacepalm:
     • root42
     • James Wells

     @clacke @root42 @tob @nikatjef @Tattooed_Mummy exactly. Especially since the marketing and a lot of the tech press leans into the "smart" bit so hard.