2024-08-07 14:58 UTC
Replies (40)
-
@larsmb@mastodon.online 2024-08-07 15:00
I can also see this going great for coding, programming languages and computers are known to be very forgiving and tolerant
-
@larsmb@mastodon.online 2024-08-07 15:05
This also makes perfect sense, because context matters - and once it generated a wrong answer, it is human enough to double down on it! The singularity is near! You've got to ask it "nicely" right from the start. Don't embarrass it! I AM A PROMPT ENGINEER
-
@dogzilla@masto.deluma.biz 2024-08-07 19:09
@larsmb@mastodon.online @msbw@masto.ai This is like having to convince your hammer that it is not, in fact, a screwdriver before hammering a nail
-
@Mossy_Modem@mastodon.social 2024-08-07 19:13
@larsmb@mastodon.online It feels like there were definitely some Monty Python skits in the training data.
-
@wirepair@mastodon.social 2024-08-07 19:42
@larsmb@mastodon.online @munin@infosec.exchange on my word that had me in tears
-
@SomeGadgetGuy@techhub.social 2024-08-07 19:44
@larsmb@mastodon.online That chat bot needs to go to the LIBARY! [Typo on purpose]
-
@nini@bitbang.social 2024-08-07 19:55
@larsmb@mastodon.online I must admit I'm impressed at how steadfast it is in never admitting to being wrong. The unearned confidence of a mediocre white man who's never been told "no".
-
@aatheus@batchats.net 2024-08-07 20:27
@larsmb@mastodon.online Oo, Gaslighting as a Service!
-
@Crystal_Fish_Caves@universeodon.com 2024-08-07 20:43
@larsmb@mastodon.online yeah this tech DEFINITELY is worth all the resources it gobbles up. We have PLENTY of spare water and power.
-
@tricotfeelya@woof.group 2024-08-07 23:07
@larsmb@mastodon.online ArtificialInanity
-
@sortius@mastodon.social 2024-08-07 23:44
@larsmb@mastodon.online the sad thing is, if they designed the LLM to purely count Rs, it might actually work... but that sounds too much like an algorithm, and that's got no techbro magic sauce in it. It reminds me of when the world was captured by radioactive materials, and they stuck it in everything
-
@eliocamp@mastodon.social 2024-08-08 01:37
@larsmb@mastodon.online Looks like they patched it, but only for a very specific subset of berries.
-
@daveley@mastodon.social 2024-08-08 04:16
@larsmb@mastodon.online I got it to admit that it was wrong eventually, but that’s not very reassuring....
-
@vampirdaddy@chaos.social 2024-08-08 06:00
@larsmb@mastodon.online now repeat after me: LLMs do not understand, LLMs do not reason, LLMs do not calculate, LLMs don’t do logic, they just guess the next words based on a laaarge data set.
-
@ptesarik@fosstodon.org 2024-08-08 06:27
@larsmb@mastodon.online Yes, LLMs make you sorely aware of the sloppiness of our speech. I suspect ChatGPT is confused because there is a “double r” in “strawberry”, and LLM correctly associates “double” with “two". A human might also tell you to write two R's in strawberry, intending to warn you about the double R in berry. I think this LLM works as designed. Sadly, some people want to ignore what LLMs (and natural languages) actually are.
-
@julbriman@piaille.fr 2024-08-08 06:33
@larsmb@mastodon.online so this is why we invented computers: to be gaslighted by them. Awesome.
-
@jm@norden.social 2024-08-08 06:38
@larsmb@mastodon.online
-
@MegaMichelle@a2mi.social 2024-08-08 08:34
@larsmb@mastodon.online
-
@chbmeyer@digitalcourage.social 2024-08-08 06:58
@larsmb@mastodon.online it's so pointless to have a discussion with a boring statistical algorithm. It only wastes my time and consumes lots of electrical energy and drinking walter, burns the planet but it never will learn anything how to do better. It could be fun to find the statistical nonsense of those services if their company goes bankrupt over time or if one of its servers processors blacks out by a logical failure, and you could shutdown the server farm this way. Also little explosions for every wrong answer would be amusing. ...
-
@andylancelot@toot.wales 2024-08-08 07:05
@larsmb@mastodon.online …this sounded like a genuine argument with a real life gammon rather than an AI 🤖 VS 🍖
-
@WooShell@chaosfurs.social 2024-08-08 07:40
@larsmb@mastodon.online If that's the AI that is supposed to take over world domination, I'm not /that/ worried anymore...
-
@tino76H@norden.social 2024-08-08 09:35
@larsmb@mastodon.online in german it seems to work a bit better
-
@riley@toot.cat 2024-08-08 10:38
@larsmb@mastodon.online The robot judge has sentenced all letters of the word to be carried out consecutively, except for one of the R:s in 'berry' and the R in 'straw', which are to be carried out concurrently.
-
@sudorandom@infosec.exchange 2024-08-08 10:43
@larsmb@mastodon.online
-
@plinth@infosec.exchange 2024-08-08 12:52
@larsmb@mastodon.online I read this story. It's in the Cyberiad by Stanislaw Lem.
-
@hindsight@mstdn.social 2024-08-08 13:38
@larsmb@mastodon.online 2 Rs is correct.. I'll say it again.. "stwawberry" only has 2 Rs.. so there ;-)))
-
@amayer@rheinneckar.social 2024-08-08 14:20
@larsmb@mastodon.online
-
@fabiscafe@mstdn.social 2024-08-08 15:21
@larsmb@mastodon.online Thats exactly my pain from the "AI helpers" I have to work with. Basically I use them to create markdown tables. Thats it. Everything else would create more work for me.
-
@sarahbeck@expressional.social 2024-08-08 15:23
@larsmb@mastodon.online ... what? 😂
-
@codinghorror@infosec.exchange 2024-08-08 16:00
@larsmb@mastodon.online yeah, I can repro
-
@fortyseven@defcon.social 2024-08-08 16:03
@larsmb@mastodon.online Maybe you just need a better model. :P
-
@rurudotorg@mastodon.social 2024-08-08 16:09
@larsmb@mastodon.online ChatGPT still isn't sure about the number of Rs in STRAWBERRY...
-
@teledyn@mstdn.ca 2024-08-08 16:14
@larsmb@mastodon.online @Gargron@mastodon.social likely no one wants to be reminded, but this is the very same flavour of super-human 'reasoning' CURRENTLY in use to 'intelligently' target civilian populations.
-
@liziwizi@mastodon.social 2024-08-08 16:17
@larsmb@mastodon.online 😂😂😂
-
@gonun13@mastodon.social 2024-08-08 16:17
@larsmb@mastodon.online You're clearly on mythsGPT
-
@mdstevens0612@mas.to 2024-08-08 16:19
@larsmb@mastodon.online Still a thing as of 8 August 2024, although GPT seems to take correction more politely tho
-
@GreatBigTable@mastodon.social 2024-08-08 16:22
@larsmb@mastodon.online Interesting. I thought I'd try it with co-pilot. I started by asking about the number of s's in Mississipi (which it got right) before asking about Strawberry. One of it's answer references even points to a page related to this bug in OpenAI. Interesting stuff.
-
@enthraxxx@h4.io 2024-08-08 16:25
@larsmb@mastodon.online I've had conversations with toddlers that weren't so stubbornly idiotic.
-
@Lily_and_frog@mastodon.art 2024-08-08 16:26
@larsmb@mastodon.online We feared gen AI would take jobs away from workers. It won't happen. But this example shows it is perfectly ready to take over middle management jobs. I'm sure I had a similar conversation with my previous line manager.
-
@BartV@mastodon.social 2024-08-08 16:26
@larsmb@mastodon.online it’s both stupid AND stubborn!