ChatGPT having a stroke?
« previous post | next post »
Or a psychotic episode? ICYMI — Maxwell Zeff, "ChatGPT Went Berserk, Giving Nonsensical Responses All Night", Gizmodo 2/21024:
ChatGPT started throwing out “unexpected responses” on Tuesday night according to OpenAI’s status page. Users posted screenshots of their ChatGPT conversations full of wild, nonsensical answers from the AI chatbot.
ChatGPT Enterprise has lost its mind pic.twitter.com/fgNdnjpSo6
— Esplin (@promisebender) February 21, 2024
Comment
byu/Sweet-Block5118 from discussion
inChatGPT
Is my GPT having a stroke? The responses are getting progressively more incomprehensible.
byu/kefirakk inChatGPT
Apparently the problem, whatever it was, was fixed a few hours ago.
Unfortunately, no singing seems to have been involved, but I can't avoid referring you to HAL 9000:
There are many interesting tidbits in David Stork (Ed.), HAL's Legacy: 2001's Computer as Dream and Reality, MIT Press 1998:
How science fiction's most famous computer has influenced the research and design of intelligent machines.
I became operational… in Urbana, Illinois, on January 12, 1997.
Inspired by HAL's self-proclaimed birth date, HAL's Legacy reflects upon science fiction's most famous computer and explores the relationship between science fantasy and technological fact. The informative, nontechnical chapters written especially for this book describe many of the areas of computer science critical to the design of intelligent machines, discuss whether scientists in the 1960s were accurate about the prospects for advancement in their fields, and look at how HAL has influenced scientific research.
Contributions by leading scientists look at the technologies that would be critical if we were, as Arthur Clarke and Stanley Kubrick imagined thirty years ago, to try and build HAL in 1997: supercomputers, fault-tolerance and reliability, planning, artificial intelligence, lipreading, speech recognition and synthesis, commonsense reasoning, the ability to recognize and display emotion, and human-machine interaction. A separate chapter by philosopher Daniel Dennett considers the ethical implications of intelligent machines.
Here's a link to Dennett's chapter on computer ethics. 27 years later, it might be time to revisit those issues, as well as the many other topics in the book's other chapters.
Update — Current Google News coverage. Benj Edwards at Ars Technica suggests that anthropomorphizing metaphors are unhelpful, but also quotes my favorite fantabulation so far:
In another example, when a Reddit user asked ChatGPT, "What is a computer?" the AI model provided this response: "It does this as the good work of a web of art for the country, a mouse of science, an easy draw of a sad few, and finally, the global house of art, just in one job in the total rest. The development of such an entire real than land of time is the depth of the computer as a complex character."
This is apparently the "explanation" offered by OpenAI:
Update on ChatGPT going off the rails
byu/HOLUPREDICTIONS inChatGPT
Terry K. said,
February 21, 2024 @ 5:04 pm
Interesting to me is the top right one in the first quoted post. It starts out as perfectly nice nice Spanglish (and even refers to using Spanish), until near the end of the first paragraph we get "l'eau" thrown in (French), and then I'm not sure what happens after that first paragraph.
bks said,
February 22, 2024 @ 6:19 am
I didn't think of HAL, I thought of the typewriter scene from _The Shining_(1980)
https://youtu.be/4lQ_MjU4QHw
Peter Grubtal said,
February 22, 2024 @ 6:43 am
Not everyone was enamoured with the film or its actors.
It was said that HAL was the most human thing in it.
Hans Adler said,
February 22, 2024 @ 7:11 am
OpenAI has published a postmortem "Unexpected responses from ChatGPT". Apparently while trying to optimize the efficiency of ChatGPT, they made it run computations with insufficient precision.
Daniel Barkalow said,
February 22, 2024 @ 12:20 pm
The explanation they gave makes me think that it was rounding down the tiny probability of each of a huge collection of responses based on long-range connections, such that it eliminated all of them in favor of kinds of response with little enough variation that they were individually likely enough not to be rounded.
Terry Hunt said,
February 22, 2024 @ 2:34 pm
@ Peter Grubtal – I think that may have been a deliberate artistic choice by Kubrick.
SusanC said,
February 22, 2024 @ 4:01 pm
And in Ridley Scott's Blade Runner, isn't part of the point that the humans (including Deckard, if he actually is human) are just as robotic as the replicants?
Jerry Packard said,
February 22, 2024 @ 5:07 pm
The ChatGPT breakdown does closely mimic Wernicke’s/fluent aphasia, with the ‘press for speech’ and semantic breakdown, but preserved syntax/phrase structure, for the most part.
Rick Rubenstein said,
February 22, 2024 @ 5:22 pm
If you trace far enough back, we were all North American Fall Webworms in a past life.
SusanC said,
February 23, 2024 @ 1:01 pm
As far as I know, we haven't had a convincing explanation for what happened here, but one of the theories is that is a combination of Dividing By The Wrong Thing and changing the number of GPUs. Code gets the logit right (by, kind of, coincidence) on initial config. Change the number of GPUs. Oh dear, we are Dividing By The Wrong Thing, and now the number is wrong.
nbm said,
February 23, 2024 @ 4:30 pm
Pretty soon AI won't be making these mistakes anymore, and then the apocalypse will truly be on us, but meanwhile, I was shown one of these earlier today in which it went wild and apparently got infected with Scots vocabulary in response to an innocent query about giving your dog Cheerios, and let me tell you I have not laughed so hard in months. Per this example, I plan to ghostline the flux and ethel the books at the earliest opportunity.
Marc said,
February 27, 2024 @ 10:34 pm
Dot your j’s and cross your x’s?