Artificial Intelligence Developments (ChatGPT etc)

[tommers]

Pop Culture

Hi there. Do you like this post? Did you know that I also do a podcast called Better Offline? If not, please immediately download it on your podcast app. Follow the show. Download every episode. Share with your friends, and demand they do the same. I did an entire episode

www.wheresyoured.at

A quite long, but interesting, blog about AI and its prospects over the next 12-18 months.

 

[kabbes]

Pop Culture

Hi there. Do you like this post? Did you know that I also do a podcast called Better Offline? If not, please immediately download it on your podcast app. Follow the show. Download every episode. Share with your friends, and demand they do the same. I did an entire episode

www.wheresyoured.at

A quite long, but interesting, blog about AI and its prospects over the next 12-18 months.

A good article

What’s frustrating for me is that I was saying literally this same stuff over a year ago:

It’s fine until it’s not. It works at doing your job until it blithely recommends or produces something that is totally inappropriate, or illegal, or offensive, or just plain incorrect, and it has no idea that what it has produced can’t be used because it has no referent, no intentionality, no originality and no understanding. Because it’s just complicated predictive text. As an expert in risk management, the amount of alarm bells this sets off is deafening, the idea that it can replace genuine expertise. Yes, automate the donkey work. But don’t assume that the result is necessarily meaningful.

And I was saying years ago that it’s just not possible for an LLM to be a reliable actor or to become an AGI because that’s not how intelligence works. And if know this, despite not having a PhD or being a researcher in the area or anything, I’m forced to conclude that either those working in it are somehow deluding themselves or they are intentionally lying for the grift.

 

[tommers]

And I was saying years ago that it’s just not possible for an LLM to be a reliable actor or to become an AGI because that’s not how intelligence works. And if know this, despite not having a PhD or being a researcher in the area or anything, I’m forced to conclude that either those working in it are somehow deluding themselves or they are intentionally lying for the grift.

It's absolutely "sales & marketing". It does a very narrow range of things reasonably well, unless you go into any sort of depth, but it's been sold as HAL, the minds from the culture and the Terminator all rolled into one. I think there's a growing number of people who actually know about it who are saying "hang on, this isn't true" but people have already swallowed a lot of the nonsense.

I used it recently to plan a road trip, it was quite impressive how it altered things based on me adding more conditions, but I'm not sure that's worth building new energy infrastructure for.

TBH I don't think you can really overestimate how gullible a lot of CEOs are and how scared they are of being the one person who didn't take advantage of this world changing tech.

 

[pbsmooth]

Sam Altman is a grifter par excellence and everyone else has followed his lead and the money.
The Nvidia valuation - and rapid recent rise - is completely mental.

 

[Signal 11]

AI Finds That AI Is Great In New Garbage Research From Tony Blair Institute

A new paper from the Tony Blair Institute for Global Change, presented yesterday by the former Prime Minister himself, predicts that more than 40 percent of tasks performed by public-sector workers could be partly automated, saving a fifth of their time in aggregate, and potentially leading to a huge reduction in workforce and costs for the government.

The problem with this prediction, which was picked up by Politico, Techradar, Forbes, and others, is that it was made by ChatGPT after the authors of the paper admitted that making a prediction based on interviews with experts would be too hard..

 

[magneze]

That's amazing. Sums up both the hubris of both Blair and the AI industry in one swoop.

 

[Fez909]

This is the closest thing to a 'real' short I've seen done with AI so far.

It's still pretty rudimentary, but I liked it. And it's clear there is still a lot of skillful human work required to make something good.



He did a breakdown of how he made it here, which is interesting in itself:



Apparently it took around 50 hours to create this.

 

[NoXion]

AI Finds That AI Is Great In New Garbage Research From Tony Blair Institute

"This post is for paid members only"


It's on archive.ph but the paywall is there too.

 

[tommers]

So I don't know if you saw the trailer for Megalopolis that kicked off with a load of negative reviews of previous Coppola masterpieces?

Well......



That is just fucking amazing. If you ever needed something to sum up this bullshit. Fucking hell.

 

[tommers]

And, I repeat, we're integrating this nonsense into every search engine.

It's a great UI. I'll give it that.

 

[mango5]

I just listened to a wild interview with mild-mannered librarian who ran for mayor in Cheyenne, Wyoming as a self-professed "meat avatar" who would outsource official decisions to a custom ChatGPT. NYTimes background article archived.

https://archive.is/jHpnE

On phone so can't screenshot the Xitter post

 

[Signal 11]

AI worse than humans in every way at summarising information, government trial finds

Reviewers were given summaries produced by both humans and so-called AI, without being told which was which.

These reviewers overwhelmingly found that the human summaries beat out their AI competitors on every criteria and on every submission, scoring an 81% on an internal rubric compared with the machine’s 47%.

Reviewers told the report’s authors that AI summaries often missed emphasis, nuance and context; included incorrect information or missed relevant information; and sometimes focused on auxiliary points or introduced irrelevant information. Three of the five reviewers said they guessed that they were reviewing AI content.

The reviewers’ overall feedback was that they felt AI summaries may be counterproductive and create further work because of the need to fact-check and refer to original submissions which communicated the message better and more concisely.

 

[ska invita]

More hype or a next level becoming available?
OpenAI launches AI models it says are capable of reasoning

https://archive.ph/MqZKQ#selection-1701.0-1701.58

OpenAI will launch an AI product it claims is capable of reasoning, allowing it to solve hard problems in maths, coding and science in a critical step towards achieving humanlike cognition in machines.

The AI models, known as o1, are touted as a sign of the progression of technological capabilities over the past few years as companies race to create ever more sophisticated AI systems. In particular, there is a fresh scramble under way among tech groups, including Google DeepMind, OpenAI and Anthropic, to create software that can act independently as so-called agents — personalised bots that are supposed to help people work, create or communicate better and interface with the digital world.

According to OpenAI, the models will be integrated into ChatGPT Plus starting on Thursday. They are designed to be useful for scientists and developers, rather than general users. The company said the o1 models far outperformed existing models such as GPT-4o in a qualifying exam for the International Mathematics Olympiad, where it scored 83 per cent compared with 13 per cent for the latter.

Mira Murati, the company’s chief technology officer, said the models also opened up avenues in understanding how AI works. “We get visibility into the model’s thinking . . . we can observe its thought process, step by step,” she told the Financial Times.

The new models use a technique called reinforcement learning to approach problems. They take a longer time to analyse queries, which makes them more costly than GPT models, but are more consistent and sophisticated in their responses.

“What it’s doing during that time is . . . exploring different strategies for answering your query,” said Mark Chen, the lead researcher on the project. “If it realises it’s made mistakes, it can go and correct those things.”

 

[Fez909]

I tried out Google's latest thing: Notebook LM.

It's pretty clever. You can feed it documents (text, pdf, website URLs) and ask it questions about the source material. I've not tested it extensively yet, but it has a feature where you can make a podcast from the source material.

I asked it for a podcast of Urban75's homepage and it created this:

AI Podcast - Urban75

Listen to AI Podcast - Urban75 by Michael Ferry / f3z #np on #SoundCloud

on.soundcloud.com

It doesn't go into the forums, or threads, so everything it says it based on the forum homepage alone. It's quite convincing, even though it gets some things embarassingly wrong, and it's extremely shallow. Maybe if it had more info it would be better.

Though, tbf, lots of 'podcasters' are as uninformed, so perhaps it's all too real.

 

[pbsmooth]

we're in an age of "good enough", which is why we have all this hype about generative AI, which can deliver exactly that.

 

[skyscraper101]

I tried out Google's latest thing: Notebook LM.

It's pretty clever. You can feed it documents (text, pdf, website URLs) and ask it questions about the source material. I've not tested it extensively yet, but it has a feature where you can make a podcast from the source material.

I asked it for a podcast of Urban75's homepage and it created this:

AI Podcast - Urban75

Listen to AI Podcast - Urban75 by Michael Ferry / f3z #np on #SoundCloud

on.soundcloud.com

It doesn't go into the forums, or threads, so everything it says it based on the forum homepage alone. It's quite convincing, even though it gets some things embarassingly wrong, and it's extremely shallow. Maybe if it had more info it would be better.

Though, tbf, lots of 'podcasters' are as uninformed, so perhaps it's all too real.

The voices are incredibly realistic - if a bit overly saccharine and generic 'American podcaster' - but it's long way from the sort of Stephen Hawkin style text to voice which has only seen modest improvements for a long while while.

I could've been fooled into thinking these were real people being a tad overly analytical about urban.

 

[mwgdrwg]

I tried out Google's latest thing: Notebook LM.

It's pretty clever. You can feed it documents (text, pdf, website URLs) and ask it questions about the source material. I've not tested it extensively yet, but it has a feature where you can make a podcast from the source material.

I asked it for a podcast of Urban75's homepage and it created this:

AI Podcast - Urban75

Listen to AI Podcast - Urban75 by Michael Ferry / f3z #np on #SoundCloud

on.soundcloud.com

It doesn't go into the forums, or threads, so everything it says it based on the forum homepage alone. It's quite convincing, even though it gets some things embarassingly wrong, and it's extremely shallow. Maybe if it had more info it would be better.

Though, tbf, lots of 'podcasters' are as uninformed, so perhaps it's all too real.

Fascinating!

As an aside, I instantly turn off any real podcast that uses "the podcast voice" these days. I find it completely unbearable to my ears, even if the subject matter is interesting. I'm not sure why, it's like nails on a blackboard.

 

[tommers]

Turns out that chatgpt 4o-mini (which tbf is free) knows nothing about anything past 2021.

 

[keybored]

I'm sorry Dave, I'm afraid I can't do that...

 

[Fez909]

This is the closest thing to a 'real' short I've seen done with AI so far.

It's still pretty rudimentary, but I liked it. And it's clear there is still a lot of skillful human work required to make something good.



He did a breakdown of how he made it here, which is interesting in itself:



Apparently it took around 50 hours to create this.

OK, this is the best thing I've seen now:

 

[Petcha]

This has been blowing my mind this morning. Creating a video about quite complex financial stuff. Using James Dean and Marilyn Monroe... You would have no idea they weren't voiceover artists.

ElevenLabs: Free Text to Speech & AI Voice Generator | ElevenLabs

Create the most realistic speech with our AI audio in 1000s of voices and 32 languages. Pioneering research in Text to Speech and AI Voice Generation

elevenlabs.io

 

[yield]

The Pentagon Wants to Use AI to Create Deepfake Internet Users
Intercept. October 17 2024
The Department of Defense wants technology so it can fabricate online personas that are indistinguishable from real people.

The United States’ secretive Special Operations Command is looking for companies to help create deepfake internet users so convincing that neither humans nor computers will be able to detect they are fake, according to a procurement document reviewed by The Intercept.

The plan, mentioned in a new 76-page wish list by the Department of Defense’s Joint Special Operations Command, or JSOC, outlines advanced technologies desired for country’s most elite, clandestine military efforts. “Special Operations Forces (SOF) are interested in technologies that can generate convincing online personas for use on social media platforms, social networking sites, and other online content,” the entry reads.

The document specifies that JSOC wants the ability to create online user profiles that “appear to be a unique individual that is recognizable as human but does not exist in the real world,” with each featuring “multiple expressions” and “Government Identification quality photos.”

In addition to still images of faked people, the document notes that “the solution should include facial & background imagery, facial & background video, and audio layers,” and JSOC hopes to be able to generate “selfie video” from these fabricated humans. These videos will feature more than fake people: Each deepfake selfie will come with a matching faked background, “to create a virtual environment undetectable by social media algorithms.”

The Pentagon has already been caught using phony social media users to further its interests in recent years. In 2022, Meta and Twitter removed a propaganda network using faked accounts operated by U.S. Central Command, including some with profile pictures generated with methods similar to those outlined by JSOC. A 2024 Reuters investigation revealed a Special Operations Command campaign using fake social media users aimed at undermining foreign confidence in China’s Covid vaccine.

More of a wishlist, so far

 

[UnderOpenSky]

Turns out that chatgpt 4o-mini (which tbf is free) knows nothing about anything past 2021.

View attachment 443192

Yes. It's why I pay for it when studying for a Microsoft exam.

 

[Crispy]

Anything on the internet past 2021 has ever-increasing amounts of AI slop in it so is ever more useless for training.

 

[Artaxerxes]

 

[Signal 11]

Don’t believe the hype: AGI is far from inevitable

Creating artificial general intelligence (AGI) with human-level cognition is ‘impossible’, explains Iris van Rooij, lead author of the paper and professor of Computational Cognitive Science, who heads the Cognitive Science and AI department at Radboud University. ‘Some argue that AGI is possible in principle, that it’s only a matter of time before we have computers that can think like humans think. But principle isn’t enough to make it actually doable. Our paper explains why chasing this goal is a fool’s errand, and a waste of humanity’s resources.’

 

[kabbes]

I’ve been saying since the beginning of the hype that “AI” is not built anything like the way that human intelligence works (or any other animal, for that matter), and so there is simply no reason to think it can ever become an AGI. The whole thing has been based on wishful thinking from the start.

 

[magneze]

Say someone makes an AGI. What then? Why does it matter? What does it improve? What, in short, is point?

 

[kabbes]

Say someone makes an AGI. What then? Why does it matter? What does it improve? What, in short, is point?

It doesn’t improve things, it fucks them. Once you have an AGI, it can instruct itself to make a better AGI and 3 minutes later you’ve got Skynet…

 

[littlebabyjesus]

I’ve been saying since the beginning of the hype that “AI” is not built anything like the way that human intelligence works (or any other animal, for that matter), and so there is simply no reason to think it can ever become an AGI. The whole thing has been based on wishful thinking from the start.

I've been doing work on LLMs this year. Enormous effort (and money) is going into attempts to teach them how to reason. But an LLM isn't reasoning. When it gets a reasoning question right, that's because it is either familiar with the question already from its training or it has lucked out with its 'dive in' approach.

It can only 'dive in'. It does go back and correct itself sometimes, but more often, it ploughs on to a nonsensical solution that it presents with total confidence. It rarely has 'hang on that can't be right' moments.

Could you teach it to systematically check its answers by applying more general heuristics to check if an answer is even roughly right, the kind of thing a human would do? Maybe. But that would involve generalising from the specific to the general and recognising what a good heuristic is, and then we're straight away into 'stopping problem Turing-type' issues.

I can't go into too many specifics about what I've been involved with, but I can tell you that it's not working! And my judgement is that it can't work.

Re wishful thinking, there is a lot of that. There are voices within academia that have been saying from the start that the various problems of hallucination, etc, can't be fixed. Unsurprisingly with the amount of money invested in this stuff, they're ignored in favour of the person who is upbeat. 'Can Do' attitude and all that.