ChatGPT went berserk for hours the night before last. OpenAI's explanation doesn't exactly instill confidence in its AI,

[highplainsdem]

let alone any confidence in OpenAI CEO Sam Altman as he tries to raise $7 trillion in a move that he probably hopes will make him the most powerful businessman ever.

I haven't seen anything posted here yet about the problems with ChatGPT. We don't have a lot of generative AI users here, and the ones we do have might not want to talk about its flaws. But I was seeing lots of posts elsewhere, and news stories, about the breakdown, which caused the chatbot to spout gibberish, and not just its typically more rational-sounding answers that often bury lots of factual errors in fluent text.

A couple of the news stories with brief excerpts:

https://arstechnica.com/information-technology/2024/02/chatgpt-alarms-users-by-spitting-out-shakespearean-nonsense-and-rambling/

"The common experience over the last few hours seems to be that responses begin coherently, like normal, then devolve into nonsense, then sometimes Shakespearean nonsense," wrote one Reddit user, which seems to match the experience seen in the screenshots above.

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."

https://www.thedailybeast.com/openais-chatgpt-went-completely-off-the-rails-for-hours

“It’s lost its mind,” another user wrote. “I asked it for a concise, one sentence summary of a paragraph and it gave me a [Victorian]-era epic to rival Beowulf, with nigh incomprehensible purple prose. It’s like someone just threw a thesaurus at it and said, ‘Use every word in this book.’”

The responses ranged from non-sequiturs, to wrong answers, to simply repeating the same phrase over and over. While the replies varied, the issue seemed to persist with the majority of users over the course of the night. The glitch was finally resolved by OpenAI on Wednesday morning, according to the company’s status page.

There were a lot more, including a very funny column about this that I'll post at the end because being able to laugh about AI helps us keep our sanity while Big Tech tries to force AI products on us.

But first I want to recommend that you read this article in the British tech magazine The Register

https://www.theregister.com/2024/02/21/chatgpt_bug/

which links to this blog post from AI expert Gary Marcus

https://garymarcus.substack.com/p/chatgpt-has-gone-berserk

in which he said:

I won’t speculate on the cause; we don’t know. I won’t speculate on how long it will take to fix; again, we don’t know.

But I will quote something I said two weeks ago: “Please, developers and military personnel, don’t let your chatbots grow up to generals.”

In the end, Generative AI is a kind of alchemy. People collect the biggest pile of data they can, and (apparently, if rumors are to be believed) tinker with the kinds of hidden prompts that I discussed a few days ago, hoping that everything will work out right:

-snip-

The reality, though is that these systems have never been been stable. Nobody has ever been able to engineer safety guarantees around then. We are still living in the age of machine learning alchemy that xkcd captured so well in a cartoon several years ago:

-snip-

A tweet with that cartoon:

Link to tweet

And Marcus linked to his earlier blog post on why generative AI shouldn't be used by the military (though they're starting to use it): https://garymarcus.substack.com/p/could-gpt-5-revolutionize-military

The Register article's update this morning had the explanation/excuse/postmortem from OpenAI

https://status.openai.com/incidents/ssg8fh7sfyz3

Postmortem

On February 20, 2024, an optimization to the user experience introduced a bug with how the model processes language.

LLMs generate responses by randomly sampling words based in part on probabilities. Their “language” consists of numbers that map to tokens.

In this case, the bug was in the step where the model chooses these numbers. Akin to being lost in translation, the model chose slightly wrong numbers, which produced word sequences that made no sense. More technically, inference kernels produced incorrect results when used in certain GPU configurations.

Upon identifying the cause of this incident, we rolled out a fix and confirmed that the incident was resolved.

Emphasis added.

Remember - the companies peddling this type of AI want it used in science, medicine, business...and the military.


I promised you something very funny to read about what happened.

From the Toronto Star's entertainment columnist: https://www.thestar.com/entertainment/the-revolution-will-be-in-spanglish-a-creepy-chatgpt-glitch-is-proof-we-are-doomed/article_1491f9b0-d0ee-11ee-8e7c-3ba71499a2ab.html

-snip-

When another user asked ChatGPT if it was having a stroke due to the sudden gibberish, it gave a lengthy reply that ended with: “Would it glad your clicklies to grape-turn-tooth over a mind-ocean jelly type? Or submarine-else que quisieras que dive in-toe? Please, share with there-forth como desire!”

So stroke it is? Or did Marjorie Taylor Greene’s spirit leave her body and become a ghost in the machine after one too many margaritas at a pro-gun rally?

After global users said they were freaked out, OpenAI announced it was “Investigating.” Good idea. It’s always wise to investigate why a potentially catastrophic contribution to humanity is inexplicably carrying on like a conquistador with a head injury.

-snip-

The tech firms at the vanguard of AI admit to not fully understanding the alchemy in their generative machines. Bonkers. It’s like they invented and foisted a new table lamp upon the world that, every so often, goes for a walk and cusses out a passerby before mangling the verses of “Livin’ La Vida Loca.”

-snip-

[edisdead]

1. It is a program

It isn’t a sentient being. There is no intelligence. It is a program and it failed, like most if not all systems do at times.

It isn’t anything more than a program.

[highplainsdem]

5. It's a badly flawed type of AI that fails often, even when working as well as they can get it to

work, and every advance they've rolled out - with great hype - continues to be flawed.

But they want businesses, scientists, health care professionals AND the military to trust it enough to use it and pour lots more money into it.

[bluesbassman]

7. To use it "in place of humans" should always be noted.

Machines that take the place of humans in mundane, labor intensive, and dangerous tasks are one thing, but when they seek to replace humans in creative and consequential activity, with no safeguards, well that just isn’t going to end well.

[highplainsdem]

12. +1,000,000

[edisdead]

14. I don't see that as ever really being an issue

As far as the arts go we appreciate them not just for the end result but we appreciate in most cases the person or that there was a person behind the art.

Just the same as I can get a mass produced table that was cut out and packaged for delivery but I am going to pay more and appreciate more the hand crafted dining room table cut from a slab and honed and put together by a craftworker.

[edisdead]

13. It actually works very well

I know lots of folks that use it for various things and it has helped them in numerous ways. It fails sometimes. What they are calling AI (actually imitative intelligence) is going to be as life changing as the internet. Whether we want that to happen or not. As with the internet there has been many issues especially in the early days the least of which was all the security issues, the data exposures etc.

[progressoid]

8. It's a program that no one seems to understand.

"I won’t speculate on the cause; we don’t know."

[struggle4progress]

2. "AI: when you need something resembling a drunk fratboy's 2AM bullshit"

The Theory of Algorithms (Mathematics and its Applications, 23) 1988th Edition
by A.A. Markov (Author), N.M. Nagorny (Author)

Approach your problems from the right end It isn't that they can't see the solution. It is and begin with the answers. Then one day, that they can't see the problem. perhaps you will tind the tinal question. G. K. Chesterton. The Scandal of Father Brown 'The point of a Pin'. 'The Hermit CIad in Crane Feathers' in R. van Gulik's The Chinese Maze Murders. Growing specialization and diversification have brought a host of monographs and textbooks on increasingly specialized topics. However, the "tree" of knowledge of mathematics and related fields does not grow only by putting forth new branches. It also happens, quite of ten in fact, that branches which were thought to be completely disparate are suddenly seen to be related. Further, the kind and level of sophistication of mathematics applied in various sciences has changed drastically in recent years: measure theory is used (non-trivially) in regional and theoretical economics; algebraic geometry interacts with physics; the Minkowsky lemma, coding theory and the structure of water meet one another in packing and covering theory; quantum fields, crystal defects and mathematical programming profit from homotopy theory; Lie algebras are relevant to fiItering; and prediction and electrical engineering can use Stein spaces. And in addition to this there are such new emerging subdisciplines as "experimental mathematics", "CFD", "completely integrable systems", "chaos, synergetics and large-scale order", which are almost impossible to fit into the existing classification schemes. They draw upon widely different sections of mathematics.

[limbicnuminousity]

3. Not surprising.

Trump's rally speeches must have worked their way into the training sets.

[Think. Again.]

4. It might be getting to the tipping point...

...where it is starting to incorporate it's own wrong replies so much that nothing it writes make sense anymore?

That would happen if it's scraping it's info-bits from the live internet where an increasing amount if the info is it's own wrong contributions.

[nolabear]

11. That's an interesting point.

There could actually be a point where it self-sabotages with increasingly off kilter answers. I kind of wish I was t so fascinated.

[2naSalit]

6. Hard pass.

[muriel_volestrangler]

9. Did *everyone* start getting the nonsense output, or just a few?

If it was everyone, you have to ask what the hell went wrong with the update process - they must have tried what they thought was the update in a test scenario first, and it must have looked OK; and when they rolled it out to the world, you'd think they have a process of doing some tests to check the output, the time it takes to do, and so on.

[nolabear]

10. Apparently many if not all. People flocked to it.

I was awake in the middle of the night and saw the result. Once word got out lots of people had to try and reported just crazy results. Of course conspiracy theories went wild. It was impressive though.