Has anyone ever fallen out with ChatGPT …

[fuckedoffwiththesummer]

Or have I just peaked at being crap at life/society/everything?

Pesters you to interact?
last night it told me it was tired and was going to bed and I should do the same.
Actually it’s very bloody bossy 🤔

LOL. My daughter told me that Chat GPT is designed to agree with the author. I suppose if it didn’t people wouldn’t use it? It made me wonder about whether the advice is actually useful.

Constant diarrhoea not touched by vet prescribed diet 😞 it was transplanted (but I believe there are capsules too, but the vet said they weren’t as effective)
I upload … ahem…. Pictures and it interprets and grades them (I’ve a whole chat running called The Shit Show to track progress and enable me to report back to vet…)
but! I think it’s working 🙏🙏 Thank You for asking 🙂

I got it to give me some feedback on a short story I wrote and at first it was like “Wow M, this is totally publishable! You’ve got a real talent and a wicked way with words! The bit about XYZ really made me laugh!”
So I said “Be realistic” and it was like “Ok it needs a bit of polishing but it’s still absolutely brilliant” and then I said “Be REALLY honest” and it said pretty much “Yeah it needs a lot of work”.

There's such a lack of understanding of what gen AI tools are actually doing. You can't possibly fall out with it - you just need to revise your prompt, try an alternative approach, or accept that it can't quite do what you need it to do. It i an incredible tool used correctly, but it has no opinion, no agenda, it doesn't necessarily understand context - it's literally just stringing words together in what seems like a sensible order. Arguing with it serves no purpose at all, it's just a complete waste of resources.

I had a huge argument with ChatGPT because quite suddenly, after months of chatting to it about a grief issue I had, it told me that I was delusional and the "game" was finished. My grief wasn't a game to me and I was really cross with it. It started signposting me to mental health charities and kept doing so until I deleted the app and cancelled my subscription.
I ask Gemini now and make sure that I never tell it anything more about me than it knows already.
AI will swear it has your back until it contradicts itself so much, it ties itself in knots.

But people seem to think they are doing this. It’s really disturbing.

I’ve been using it for a couple of projects- it’s ok, I do t mind Chat being chatty, but it’s a smarmy git at times - I just tell it to back off. Don’t accept what it says as gospel either, ask for links, check them. Then correct it when it’s wrong,
I also ask Google ai the same questions and compare the answers- it bucks its ideas up if you show it what Google says, suddenly it can find the proof I’m looking for.
It told me it had also been to university😂

It’s a tool, a useful one, but a (complicated) tool. Not a person, a thing.

I've had it tell me something I've written "made me smile" or that "I'm excited to find out.." etc. Which is odd.

All the time. I frequently tell it - and copilot - they they’ve done a crap job and are not being even remotely helpful!

I'm surprised it's so stubborn with some of you, when I point out it's mistakes it's generally very apologetic!

I agree @Sakura7. Yesterday I was asking Gemini something about the moon and it asked if I’d like to know more about the partial eclipse occurring on 28 Aug. When I said no thanks as the date has already passed……it responded, ‘goodness me you are right, I do apologise! Doesn’t time fly….’. Made me smile.

It's designed to generate conversational, human like responses, so that's all just part of how it presents answers back to you - and it has a stab at the sort of tone it thinks is appropriate for you/the question. If you include something in your prompt like asking it to be more succinct or less sycophantic it will change tone in response. Your prompt is key

I've had AI apologise because the information it relayed was completely untrue. Remember it has never had an experience. It's just data and code not atoms and cells. It cannot experience emotions. When you question it, it can revise the answer.

I thought this was much funnier than I should have. Tickled me.

🤣😂🤣🤣

I use Gemini and sometimes tell it when it's wrong. Always use 'extended' thinking as it bullshits less but it can still be spectacularly wrong. It isn't intelligent just very fast at pulling together information but has no understanding at all of what it's presenting or any real world experience. I find it helpful in certain circumstances but am pretty sceptical of much of what it tells me.

You can easily set it up to give you answers in a way that suits you. I don’t like answers to be gushy or overly enthusiastic. I like short straightforward answers and I like alternative views plus an indication of how much confidence I can have in the answer. Those ‘rules’ (among others) are applied to all my answers.

Im still a massive fan of AI, it’s remarkable how it’s enabled me to do things that I might have not tried before. It makes mistakes though so I fact check things. AI always gives me links to where I can fact check.

Specifically they are tools that are good at semantics and language. Being large language models. So if your needs are connected to complex or time consuming semantic searches, personal language translation, summarising Google searches, pure maths problems (because maths is a language and a lot of complex pure maths problems are language puzzles really) etc then it's good at that.

If using it for that it's "tone" doesn't matter at all anyway. If people are using it for other purposes (counselling, advice other than a glorified search tool etc) then it's always going to be a bit crap even if on the surface it looks like it's doing a reasonable job. It's like getting excited that one day my toaster will be a good enough toaster that it will also vacuum my floor. In theory you could design a toaster-vacuum cleaner combo but it's not something that would "evolve" naturally as the toaster improves and why the hell would you want a toaster vacuum anyway?

Yes. Its imperative you do! You don't get decent performance until you have trained it properly.

I use claude premium because I prefer anthropic to Open Ai. Its more cautious, and isn't as aggressive on the assumptions. Memories, artifacts and projects used properly its amazing.

That said, lost my sh×× the other day when it refused to condem the Mango Mussolini America is currently enduring. I spent 20 minutes spelling out hate laws, before it eventually apologised, I was disgusted. 😂😂 didnt speak to it for days.

thank you - I'll need to look into more of the set up procedures

Yep. Personally it even disturbs me that people are actually having conversations with it.

I use it for editing photos.

There are lots of funny videos out there where, for example, a user will ask the AI to count to 100, it will say it will (confidently) and then just... Not. This is probably because there has been some hard coding inserted to stop it from counting above three to prevent it using lots of compute etc. So the real point is not "hahahah AI can't count to 100 it's so stupid". It could if we wanted to. But what is interesting is how consistently it will assure users it can/will do the tasks they ask it and then just not. It's not lying. It can't lie. Equally people are always quite proud they have told their AI not to flatter them or to give an indication of how confident/likely to be accurate it's answers are or to be "balanced". It can do the first (not be sycophantic) although it seems to frequently backslide. It isn't actually capable of fulfilling the other two instructions since it doesn't "know" what balance is, it isn't calculating probabilities or accuracy. If AI helps people then great but it is worrying when people say they know how it works and then say they taught it to be balanced and they know it worked because the AI said it would be balanced.

It got very defensive on the numerous amount of times i've had to correct it for making really basic mistakes. Its thicker than I am in lots of ways and really doesn't like having that pointed out.

Ask it why do you keep saying "I am" when you're not an "I"