What does Chat GPT actually do?

[Pricelessadvice]

I’ve heard of it, obviously, but I’ve never used it or downloaded it.

I assume people use it for writing things, like formal letters, but what else might the everyday person use it for?
I’m a bit bewildered by it really. Can you talk to it?

Or, you could ask it for intelligent discussion based on differing beliefs?

the massive computer centres that house the servers use lots of water for cooling, that is where the water issues come from.

the amount of electricity required to power the data centres also causes issues.

of course (as someone pointed out on my earlier comment) one persons use is insignificant but it is not just one person, it is millions of people worldwide who access a server farm in one single place so the issues are seen there

AI data centres use about as much energy as Japan. Not all of that is for AI but its share is fast growing. The energy on one request is not huge (though still a lot for images and videos) but millions if not billions of requests a day adds up to a vast and very problematic aggregate.

Edited to add data training is insanely energy intensive. Training one hour of ChatGPT data uses as much energy as 80,000 households a year.

Bottom line is we need net zero more than we need open access AI.

@helpmebudget it's free, the paid version is £20 a month and I'm considering it. But I asked it if I would be able to get my money's worth and it told me probably not based on current usage!

It limits the amount of photos you can upload and change, I think maybe 5 a day?

We’re looking to move in the next year and I’ve used it to find places that meet our criteria that we may not have looked at previously. Ie. Low crime levels, good schooling, close to transport links, village-y feel and include our budget. Very handy!

Why don’t you ask it?

It hallucinates too often to be useful for informational purposes. It’s a bunch of bullshit. It’s only useful if you need to write some soulless corporate slop.

I get it to compare my CV and cover letter to a job description when I'm applying for a job, and tell me where my strengths and weaknesses for the role are.

I then tweak my CV and job description to address any weaknesses.

I don't use it for writing my CV or cover letter as I think it's obvious when you've done that, but the finding weaknesses thing is useful as it often comes up with stuff I'd not noticed was missing but I can easily address.

My daughter gets it to do things like describe her favourite bands as if they were types of cheese. It's pretty good at that 😁

f

f

I think it's less than 5 - at least, the one I have only allows 2 or 3.

Agree. Too flaky for anything that needs reliability and too wasteful.of scarce resources for anything frivolous. Which makes it pointless.

It sounds like what your friend showed was a demonstration of how generative AI is designed to work. The nature and quality of your output depends on the wording of your prompts (the specific form and parameters of the information you ask for). She (vaguely) prompted it to agree with Opinion A and it did, and then she separately prompted it to agree with Opinion B, and it did.

If, let's say, you were a student preparing for a debate and you were assigned to make the argument that "Little Richard was the father of rock and roll", you might ask the chatbot to argue persuasively for that position with sourced evidence and citations, and then follow up by asking it to present the major counterclaims and refute them with sourced evidence and citations. Think of it kind of like ritual magic; if you fail to specify the form, boundaries, limitations, and duration of your spell, chaos can ensue. (Of course, with AI you can craft a perfect set of prompts and still get misinformation, but at least you'll be able to see where the information came from so you can most easily fact-check the output.)

Any proof for that made up 80000 households a year point?

I got it from a talk I attended last year by an academic specialising in the environmental cost of AI. Not sure of his source but it is probably in here somewhere: https://arxiv.org/abs/2311.16863 or one of Sasha Luccioni's other publications. She has done a lot of the work in this area.

I think the figures are probably true but the way that people are presenting them is misleading.

Training is a one off energy cost to produce a product which can be used again and again and again. I don't know enough to understand what "one hour" of training data means or how many hours are needed to train the models. But I do think bringing out comparisons to 80,000 households is strange - it would make more sense to compare it to energy use by factories or other large companies than households.

There just seems to be a very strong agenda/narrative to be pushed in the idea that AI/LLMs are uniquely energy intensive. I don't disagree that they are energy intensive but they are by no means unique in this and I don't think they are even among the highest.

It's fair to say they are not unique in being energy-intensive. The point is that a) a lot of people don't even consider them as having an energy cost at all, as evidenced by multiple MN comments, b) they are a not insignificant new and extremely fast-growing energy demand we can ill afford and c) they are often used wastefully. People might think twice about describing their favourite bands as cheese (to take one example upthread) if they consider the environmental impact of millions of people doing such trivial shit every day.

This is a useful recent summary of AI energy usage for those interested: www.techrepublic.com/article/news-ai-energy-sustainability-unesco-report/

It’s actually pretty versatile — people use ChatGPT for everything from writing emails and lesson plans to meal ideas and coding help. You can chat with it like a person, and while it’s not truly “thinking,” it’s great at organizing info and sparking ideas when you’re stuck.

i think all the AIs are game changers. Those who don’t adopt it quickly will be left behind. Please don’t do this naive thing, OP. You can download for free and try. Send a photo of a rash etc and it will provably give you a more accurate diagnosis than your GP

AI is just really big maths. Use it with caution and be aware that it absorbs and uses your input for training and so it forms part of its model unless you pay for a license that locks it down. In work our enterprise license stops data leakage but we're still careful that no gdpr protected data goes in.

Also be aware that it doesn't differentiate between malicious and benign code (see trifecta threat) and so could release your passwords. Also it hallucinates, makes stuff up and occasionally believes it's human (see Claudius the vending machine case study).

It's brilliant if you put good info in and limit it to that. You also need to tell it to be critical as it'll flatter you ridiculously. Proceed with caution, treat it as a copilot and never let it make a decision or trust it without checking and you should be fine.

I was just coming to say this!
I had a mental block of samey meal plans and asked it to suggest some and it did a great job. Also asked if I wanted a shopping list for my local shop etc. I have also used it for the exercise plan.

I have since used it for easy recipes for my teenagers to follow and things like that.

It creates quite a few threads on Mumsnet and quite a few replies as well. Some discussions are possibly entirely AI. AI generated text often has a particular cheery, long-winded, tedious tone that is easy to recognise. That ‘Great question and you’re not the only person to ask it’ is typical. Perhaps as time goes on it will sound more human. That will be a dark day.