Urban75 Home About Offline BrixtonBuzz Contact

Artificial Intelligence Developments (ChatGPT etc)

FTC investigates OpenAI over data leak and ChatGPT’s inaccuracy

The Federal Trade Commission has opened an expansive investigation into OpenAI, probing whether the maker of the popular ChatGPT bot has run afoul of consumer protection laws by putting personal reputations and data at risk.

The agency this week sent the San Francisco company a 20-page demand for records about how it addresses risks related to its AI models, according to a document reviewed by The Washington Post.
[...]
The FTC called on OpenAI to provide detailed descriptions of all complaints it had received of its products making “false, misleading, disparaging or harmful” statements about people. The FTC is investigating whether the company engaged in unfair or deceptive practices that resulted in “reputational harm” to consumers, according to the document.
[...]
The agency also demanded a detailed description of the data that OpenAI uses to train its products, which mimic humanlike speech by ingesting text, mostly scraped from Wikipedia, Scribd and other sites across the open web.
 
Anyone used Firefly yet?

It's fucking scary. Will seriously put graphic designers out of a job. I mean it's amazing but. Hey. I need a job.

 
Not sure rebuttal is quite right. The reality actually sounds worse.
Seems like a rebuttal to me:

Claim - ChatGPT is getting worse at prime number recognition
Truth - It was never good to begin with

Nuance - the dataset chosen highlights the differences between the models, but not their capabilities. If they'd given ChatGPT 100 non-prime numbers, would they have written the headline "ChatGPT can recognise primes with 100% accuracy"?

The 'rebuttal' article suggests it'd get them all right if you asked it about only non-primes.
 
this seems important - another new development


by my understanding:
explains how researchers are getting GPT4 to be self reflexive and learn in a way that IMO is the way humans do.

Basically when it gives an answer it reflects and asks itself if it got it right, recognises its own mistakes, and then gives an improved answer. Elsewhere I've read about how this method is being deployed at all times, so giving the computer a task it will continue to interrogate its conclusion and improve on them, at lightning speed.

What is incredible is that it is now at a point where it can recognise its mistakes and improve on them. GPT3.5 could not do this.
This process also upends earlier predictions about rates of improvement, as they were based on more data input. This suggests more data isn't necessary

You would expect a computer to broadly give the same output based on input. Limitations would require human upgrading to overcome. This transcends that.

...GPT5 + self-reflection and reasoning could well be a significant step towards something that could reasonably be called artificial general intelligence

Coming late to the thread but reminds me of something I read 30 years ago. Researchers in the 70s? cut the corpus callosum (bundle of nerves that connects the two halves of the brain) of people with epilepsy in an attempt for the subjects to control episodes. They then put a card between their eyes and showed their left eye an object and asked them what they saw.

The left eye connects to the right half of the brain, but it's the left half of the brain that controls speech, so the right half of the brain knew what they'd seen but couldn't talk and the left half of the brain could speak but didn't know what they'd seen. The subjects made a guess at what they'd seen and got it wrong, shook their head and made another guess until they got it right. :eek:
 
Cleaning up a baby peacock sullied by a non-information spill

“Adorable!” I thought, and clicked share. Two friends of mine very quickly pointed out that this is a synthetic (generated) image. (This is quite plain when looking at the screencap above on my laptop; it was less so when scrolling past it on my phone.) I thanked them and removed my share of the post.

I then wanted to find out what baby peacocks really look like. Here’s a screencap of the current Google Image search results: [a mixture of real and generated images]
[...]
In the grand scheme of things, this particular experience (regarding peachicks) is just a minor inconvenience. But at the same time, I found it to be a vivid example of the effects of the synthetic media spill — and it’s frightening to think about similar experiences accumulating for many people across many topics.

I tried this and DuckDuckGo did much better than google.
 
Probably one of my most mundane yet invaluable applications of it yet. As anyone who has read my previous posts might have noticed, I struggle with dyslexia. I frequently omit words and commit other silly brain-related typos, which I even catch onto when I reread something, but not straight away (I extensively review work emails before sending them, yet some errors still elude me, only to become glaringly obvious upon receiving a reply).

In any case, I am currently completing a lengthy job application form and decided to employ ChatGPT's assistance. Not to artificially enhance its quality, as everyone can detect AI-generated text, but rather to enhance its grammar and structure. This has truly revolutionized my approach.

*this post might have had the same treatment :oops:
 
Probably one of my most mundane yet invaluable applications of it yet. As anyone who has read my previous posts might have noticed, I struggle with dyslexia. I frequently omit words and commit other silly brain-related typos, which I even catch onto when I reread something, but not straight away (I extensively review work emails before sending them, yet some errors still elude me, only to become glaringly obvious upon receiving a reply).

In any case, I am currently completing a lengthy job application form and decided to employ ChatGPT's assistance. Not to artificially enhance its quality, as everyone can detect AI-generated text, but rather to enhance its grammar and structure. This has truly revolutionized my approach.

*this post might have had the same treatment :oops:

Well it is a language model, so if ChatGPT should be good at anything, then it should be helpful in smartening up grammar and structure.

So yeah, no shit that ChatGPT doesn't really know what a prime number is. It's not a mathematics model.
 

That's not my experience of ChatGPT at all. Outputs relating to anything that could be remotely controversial or contentious are always hedged at the end with at least one paragraph of tedious fence-sitting and an obstinate refusal to come down firmly on a position. When I've used I've sometimes had to give it some serious prodding to give me something that isn't wishy-washy liberal centrist bollocks.
 
Read an interesting thread about how chat gpt made a typo when trying to write ancient in a response. The replies got technical quickly! I will dig the link out if anyone is interested.
 
Proper eyeball psychopath.
I mean, what could possibly go wrong?
4740616-4090865107-17003.jpg
 
Mushroom pickers urged to avoid foraging books on Amazon that appear to be written by AI

Amazon has become a marketplace for AI-produced tomes that are being passed off as having been written by humans, with travel books among the popular categories for fake work.

Now a number of books have appeared on the online retailer’s site offering guides to wild mushroom foraging that also seem to be written by chatbots.
[...]
Four samples from the books were examined for the Guardian by Originality.ai, a US firm that detects AI content. The company said every sample had a rating of 100% on its AI detection score
 

“These kinds of arguments around existential risk or the idea that AI would develop super-intelligence, that was very much on the fringes of credible discussion,” says Mhairi Aitken, an AI ethics researcher at the Alan Turing Institute. “That's really dramatically shifted in the last six months.”

The EA community credited Hogarth’s FT article with telegraphing these ideas to a mainstream audience, and hailed his appointment as chair of the U.K.’s Foundation Model Taskforce as a significant moment.

Under Hogarth, who has previously invested in AI labs Anthropic, Faculty, Helsing, and AI safety firm Conjecture, the taskforce announced a new set of partners last week – a number of whom have ties to EA.
 
That article seems to set up a false dichotomy. It is possible to be concerned about long term future risks and also be concerned about immediate problems. And the alignment problem is a problem that applies to both long and near-term issues. The issue seems to be one of what regulation should focus on, but I would have thought that it isn’t actually necessary to pick one or the other.
 
I've been using it a bit recently to make scenarios for classwork.

I'm goin to experiment with it a bit to see how it does with marking. I don't expect much for the bulk of the decisions but I'm hoping it will help with marking spelling and grammar which is a bit of a blind spot for me due to my dyslexia.
 
Interest in generative AI is fading, according to this report

i think day to day average consumer excitement has dropped but in all areas where it is most transformative, say for example this kind of thing
theres only one direction of travel
 
According to some webmasters I follow some search engines are becoming clogged up with masses of obviously AI generated content.

This is important as the large language AI systems train themselves on internet content, and soon they as a species could be learning off content they themselves created.

I doubt that will end well.
 
..
In any case, I am currently completing a lengthy job application form and decided to employ ChatGPT's assistance. Not to artificially enhance its quality, as everyone can detect AI-generated text, but rather to enhance its grammar and structure. This has truly revolutionized my approach.

*this post might have had the same treatment :oops:
Hi UOS, I know how to prompt ChatGPT to write things based on asking it increasingly detailed questions, but how would I get it to improve the grammar of a piece that I had written?
 
Back
Top Bottom