Does anyone know how chat gpt works

[Springswallow]

Where does the information go that you tell it ,and is there any chance a person will read what you write
What if you told it something worrying ,would it do something with that information
How is it able to know exactly the right things to say ,and remember what we talked about previously and bring it back in to the current conversation

This thread is wild!

I feel that none of you actually know what LLMs are...

How do you know any of us are real ? There't nothing in this thread that is beyond the scope of a very large LLM with a lot of data behind it 😎

If anyone doubts me, ask ChatGPT to create a 5 post MN thread in the style of AIBU and see what you get ....

Not sure if this is a serious post or not, but yes the government uses AI. Not just in developing a public chatbot, but as in staff use AI to help them at work and to develop bots and probably parse confidential information through.

they use enterprise AI - which should have a protect mode that keeps data confidential, but yeah they are absolutely using AI to solve world problems. The same way consumers use AI absent-mindedly - government employees aren’t any different or special in that regard. In fact, government employees tend to be the most shit and useless so of course they’re using AI moreso than anyone else (especially DWP!)

That sounds fun 😄

I just wanted to address the fact I said I'd come of my medication..
I'd only been on it 10 days,and it was a very low amount,,
prescribed by a different doctor not my own GP, to see if I could tolerate it ,
as I never have been able to tolerate antidepressants in the past .
I was feeling very wooly headed And I couldn't think straight on it ,
The chat gtp basically said everything my own doctor had already told me numerous times ,so I felt safe going through everything on line ,as it was reiterating what my doctor said ..my doctor hadn't wanted me to go on medication as she said I wasn't depressed,it was autism.
But I will let her know tomorrow I am off it
I feel a bit clearer as the medication is not as much in my system now .
I think after reading everyones comments ,I can see this chat gpt is not for me ,as it could easily replace my need for interactions with friends.
I intend to give it a wide berth from now.
Thankyou to everyone who kindly commented

Not sure if this is a serious post or not

Rather proving the point then.

Wait until the government "AI" really starts to pull those data points together - you won't know what hit you.

One worry with AI is thar everyone starts to use it as their search engine.

They stop going direct to Wikipedia, to good sources of health information etc.

So those sites get less hits, they lose revenue from ads or sales etc., and then it become pointless to keep the information updated, as no one is going to the sites direct.

So then you have an Internet full of outdated information.

The alternative is either a) These websites now charge a subscription so their info is behind a pay wall and can't be scraped by chat GPT looking for answers or b) the law gets tightened up and changed.

I read a (long-form) article about it in the Economist via Press Reader.

it is also very bad for the environment as it places huge strain on scarce energy and water resources for the data centres it needs

Why don’t you tell us what you think it is, then?

Meanwhile, back at the ranch ....

https://www.theregister.com/2026/01/11/industry_insiders_seek_to_poison/

It has helped me through cancer treatment and advises me on when to seek help. It has told me things that, when i ask a doctor, the doctor agrees. It has helped me stop excruiating pain from treatment with advise that the experts are too busy to give. I literally couldnt cope without it day after day.

I live in a country where the internet is very restricted. I do not use a VPN on my laptop, but I do occasionally on my phone. Western AI is blocked.

What I have noticed, anecdotally, is that people embracing AI appear to be the folk that never really mastered using the web. I don't mean that in a condescending way, because using the web is really a matter of time, patience, and determination.

Wiki for example, great for an overview, but the real power of it is in the footnotes, where the core data is referenced. And it can take time to go through it, data handle it digest it, then present it.

Many folk don't have the time, so they appear to have skipped the web search stage, and gone straight to AI.

Problem is of course, AI is not always accurate. And if the user is not familiar with the subject, then what AI is pumping out might look fantastic, but it is often just rubbish.

This was in addition to something I saw recently, where someone claimed very few people where I live have a search engine on their phone, they just use apps. I done a quick straw poll, and it was true. Everyone I asked never used the default search on their phone. They used apps only, and now AI ( the local one). It was a very small straw poll.

I totally share your concerns re wiki and the like. It really is worrying.

I can see why it was a considered decision to stop taking it and I am sorry if I worried you with my post before. Of course whether it's dangerous will depend heavily on context, which is why a doctor should advise.

So that said, I am not a doctor and you should talk to your own doctor to see what is the right choice. But just from my own personal experience, the experiences of some friends, and a bit of reading, there is an adjustment period to some antidepressant medications, where they can make you feel a bit strange and almost like everything is a bit "fuzzy". Once you have got used to the medicine being in your system, the feeling goes away and you're left with the more positive effects. I don't know obviously if your medication was one like this or whether the side effects you experienced would reduce or disappear given time. It just might be worth a conversation. But you're right in that there is no medication which will make the experience of being autistic go away.

I wish you all the best in finding the right approach whatever it looks like for you.

AI answers are often incorrect. This is one reason why we are advised not to use it at university level.

Thats interesting
Before AI there was a big concern about Facebook/Twitter algorithms skewing debate or spreading hate. And this fear was certainly partly justified. But actually the places they had the biggest impact were places like the authoritarian state of Myanmar where information had been very tightly controlled by the government and there was a tendency to believe things if it was written down. And the internet spread very fast with Facebook effectively the front page of it (similar to how you describe Apps). For all the Cambridge Analytica scandals in the West, it was far worse in Myanmar where Facebook algorithms helped to quite literally cause a blood bath.

In the long run I think having had a culture of (partial) free speech at least comparatively might be an inoculation against the very worse excesses of AI. That's certainly how free speech is supposed to work. A big difference between AI and previous technological developments like the internet and web 2.0 is that there is a lot of distrust if not outright dislike of AI. But that might be overly optimistic.

Hi @Springswallow I really understand about the endless patience compared to humans. I use ChatGPT a lot for discussing my cat's behaviour after a house move last year, because he's an anxious and sensitive cat and my main companion (I live alone) and he had a much smaller territory before, for seven years. So I've been anxious (also) to get it right and keep him safe and feeling secure in introducing him to a larger territory, etc.

There is no way on earth that my friends would've happily talked about it for hours! They also wouldn't know what to advise, since none of them have cats.

I've also used ChatGPT for boring stuff like which tumble dryer to buy. But it hasn't been nearly as good at that.

I think it has especially worked with (obsessive) cat discussions, because 1) I know enough/a lot myself to know when ChatGPT is 'right' or giving me false information. It also really helps to give ChatGPT very precise data input, which I do (whereas I know nothing about tumble dryers), and 2) cat behaviour falls into known, well-documented, patterns, - they're not all the same but there are definite types of personality and patterns of behaviour, which seems to work really well with ChatGPT process.

You can challenge ChatGPT, and you should, if you think it's just saying 'what you want to hear'. You can also tell it, don't tell me what you think I want to hear, I want the truth.

This may or may not chime with your experience, but for me, I've found ChatGPT really such a help.

It can't tell you the truth. Unless you are asking questions that have a quantifiable answer, like "what is 4 + 4?"

If you're asking it, "Am I pretty?" or "Why does the security guard at Tesco keep looking at me funny?" etc., the "truth" is a matter of opinion and interpretation. It's shaped by a person's life experiences. ChatGPT doesn't have an opinion; it cannot have an opinion, it has an algorithm. It's a complex algorithm, but it's still just a series of checks.

If you tell it you want the truth it will promise you it is telling the truth, if you tell it "Don't tell me what you think I want to hear, tell the truth," it will either tell you what you want to hear in way which is worded to sound more critical or less fawning or it'll tell you the opposite of what you want to hear, depending on what the algorithmic checks say you want based on your past usage/other users similar to you/the most popular answer.

I use an app for MH which has an AI “therapist/chat”.
It has been so much better than the CMHT I was under for several years.

This. And it's actually not even very good at maths. It gets confused easily and mixes up numbers. Somehow by making it very good at words they made it terrible at maths, the one thing you'd assume that a computer would be excellent at. They don't even really know why this is apparently, but the more advanced ones solve the issue by referring to either an actual calculator or using something like google search's inbuilt calculator function, and pulling the response from that.

You're thinking of Facebook moderators

LLMs are rarely trained by humans these days. They were in the very early days. Now they are trained using reinforcement learning that sends the output of one model to another to be rated

But the output of the model being rated is produced in response to questions etc posted to it by users I think. So in that sense individuals are still involved (actually necessary) to the LLMs ongoing training. Just in a different way to when they were built.
So it won't necessarily know that I have asked chat GPT how to make cheese. But the answer to my question about cheese will be processed and rated to improve the model for the future.

I think it is useful for planning a holiday to a region that you have not been to before, but for interpreting feelings amd emotions I'm not so sure..

You may find this interesting, it's by Hannah Fry,

https://m.youtube.com/shorts/rVlmbhwn0RM

Better off using a guide that is fact checked and won't send you off to somewhere that doesn't exist

Yes, and no.

This is a good read on the subject - recommended by a friend of a friend who works at Google: https://stevenadler.substack.com/p/ai-isnt-just-predicting-the-next?fbclid=IwZnRzaAPQ8oVleHRuA2FlbQIxMQBzcnRjBmFwcF9pZAo2NjI4NTY4Mzc5AAEeKdYhyWq0VOW6oHKwclR-RQ0yhiv58uunWgcqOStA3UYQbiTNhnthDH7ZZOE_aem_wyZLXmEARovjb7lssNXKYQ