AI! AI! A ChatGPT is come!

[Frelga]

Tech has finally delivered a language model that can predict the sequence of words that are most likely to go together in response to a natural language query and produce a coherent text. This is Quite Cool and has a number of genuinely helpful applications.

Unfortunately, they called it an Artificial Intelligence. It isn't. It's not designed to reason.

Even more unfortunately, a lot of people are unclear on its potential and limitations. Worst of all, businesses decided they can use this text generator and fire actual human writers, journalists, and tech support staff.

In today's example of what the AI is not - it's not a legal search engine.

[Voronwë the Faithful]

It gets worse.

[Frelga]

What does "let him show cause" mean?

[Voronwë the Faithful]

It's just a legalistic way of saying that he needs to explain why he should not be sanctioned. Which he will not be able to do.

[N.E. Brigand]

Great subject line!

(Sily stories like this aside, AI makes me nervous.)

[N.E. Brigand]

The lawyer in question specifically asked the AI program if the cases were real, and the program lied and said yes. This is not the first case of AI programs lying. Are there theories about what AI programs do that? Has anyone tried punishing the programs for lying?

[Frelga]

It's not lying. It's doing exactly what it's supposed to be doing - generating text based on the prompt that it was given and the body of information used to train it. If you ask it to generate text about something that doesn't exist, then you can't blame it for doing just that.

Someone (I'll find the source) called it Spicy Autocomplete, and it's a helpful reminder that it's not a search engine.

Courtney Milan points out that the public version is designed to avoid repeating the texts it's been trained on verbatim, to avoid charges of plagiarism. This makes it even more likely to "hallucinate" and provide inaccurate information or a non-existent quote.

This blog post is a good summary of the story. It also seems to suggest that maybe the lawyer is to blame for more than the naive use of a new technology, as it's not easy to make ChatGPT hallucinate in that specific way.

[Maria]

Our sales manager got ChatGPT to suggest names for a new product we were launching- and we humans ultimately picked one of the computer's suggestions. So even small manufacturers are using it nowadays.

My programmer daughter uses it to look up obscure bits of coding so she doesn't have to waste time on something that's already been done.

[Alatar]

Its definitely a useful tool. When used correctly.

[N.E. Brigand]

The Center for AI Safety has issued the following statement signed by many scientists who work on artificial intelligence:

"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Last week I read that some futurists are concerned less about artificial intelligences causing human extinction and more about artificial intelligences causing endless human suffering.

[N.E. Brigand]

The Center for AI Safety has issued the following statement signed by many scientists who work on artificial intelligence:

"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Last week I read that some futurists are concerned less about artificial intelligences causing human extinction and more about artificial intelligences causing endless human suffering.

That said, there's something to this:

[Frelga]

Its definitely a useful tool. When used correctly.

Indeed but unfortunately it's a big when, shading into an if.

I imagine searching legal databases for cases is a very good application for a machine learning model. But, it would have to be trained on those databases and instructed to stick to quoting them verbatim, which ChatGPT et al are explicitly instructed not to.

It's also helpful with coding. Once I explained to the Bing AI what I don't want, it generates a correct code snippet, and even supplied a link to Stack Overflow where I would have normally copied that snippet from. Others managed to make it write a simple game without editing any code, but with multiple rounds of feeding back errors and corrections.

It's not qualified as a therapist (what were they thinking?)

Screenshot_20230531-091202.png (296.71 KiB) Viewed 7405 times

Here's the second article.

[RoseMorninStar]

Daughter told me she successfully used used ChatGPT the other day for some code and it saved her hours.

[narya]

The eating disorder chat line is closing down, and the clients are being forwarded to those of us phone volunteers who man the national 988 hotline/text-line. I'm brushing up on my understanding of that issue, so I'll be ready when a call comes in, and hopefully a little less clueless than Tessa.

https://www.nationaleatingdisorders.org ... t-helpline

[Inanna]

Daughter told me she successfully used used ChatGPT the other day for some code and it saved her hours.

I’ve started doing the same. It’s great for these quick lookups.

[Túrin Turambar]

AI-generated entertainment has been having rough start.

AI-generated ‘Family Guy’ livestream banned after making a bomb threat

An AI-generated livestream version of Family Guy has been banned after it appeared to make a bomb threat.

The parody version of the animated sitcom was being hosted by a channel called ai_peter before streaming platform Twitch ended the broadcast.

In the stream — which has no affiliation to the show and its creators — the character of Peter Griffin started talking about how to plant a bomb in a venue in Washington DC.

Which shows a limitation with anything AI-generated - it's limited by its inputs.

[Voronwë the Faithful]

The judge ordered that the attorneys responsible for citing and quoting the false cases have to pay a $5,000 sanction and send copies of the sanction order to their client and to each of the judges that were falsely claimed to have written the non-existent opinions. He let them off very lightly.

[Dave_LF]

(I just got the joke today)

SlowpokeRFVH-4123970859.png (637 Bytes) Viewed 7232 times

[Alatar]

The title sequence for MCU's Secret Invasion TV Series on Disney+ was created using AI Art. Legitimate artists are naturally up in arms, but frankly, they should be cause it demonstrates just how valuable a real artist is

[Dave_LF]

I know AI is going to kill us all and everything, but at least it will entertain us on the way out (no I don't know that the origin story is really what it says it is):
.