AIBU to think ChatGPT is not fit for purpose?

[Lilifer]

As of right now ChatGPT is insisting that Joe Biden is the democrat presidential nominee. And that is both on the free version and the expensive paid for version. I cannot believe that it continues with out of date information on something so hugely topical and important in the world, it's really made me rethink my using it for anything else.

Well that at least is an appropriate answer.

Any AI can only work with what it's trained on, training takes time and so they're generally not useful for current affairs. You need to use the right tool for the job, and it's told you to do so. What's your problem with that?

You can't complain something isn't fit for purpose if you're using it for something inappropriate.

The 'hallucinations' and incorrect results and lack of ability to provide accurate references mean it's actually not fit for many serious purposes and shouldn't be used without then fact checking it. It can clean up grammar and provide summaries which may be useful starting points.

I would consider that a good answer. It's directing you to where you might get answers rather than categorically giving you an answer that might not be true.

I thikn expectations of AI are often unrealistic. It's a TOOL. Like any other tool. to be used with purpose and skill.

I asked the same and got this result:

The 2024 U.S. presidential election, scheduled for November 5, features several candidates:

• Kamala Harris (Democratic Party): The current Vice President, Harris became the Democratic nominee after President Joe Biden withdrew from the race in July 2024. She selected Minnesota Governor Tim Walz as her running mate.
• CNN
• Donald Trump (Republican Party): The former President is seeking a second, non-consecutive term. He secured the Republican nomination and chose Ohio Senator J.D. Vance as his vice-presidential candidate.
• Wikipedia
• Jill Stein (Green Party): A physician and activist, Stein is the Green Party's nominee. She previously ran for president in 2012 and 2016.
• Ballotpedia
• Chase Oliver (Libertarian Party): An activist and politician, Oliver is representing the Libertarian Party in the presidential race.
• Ballotpedia

-

Interesting. - thanks all for your responses, i wlll keep all that in mind when using it in future

I wouldn't say it's not fit for purpose, because its intended purpose is not to give factual answers, as would make sense as anyone can post any old shite on the internet and the AI may use that to build its answer to you.

When I googled "what is ChatGPT used for" I got this:

ChatGPT is primarily used for natural language understanding and generation, making it valuable for tasks like content creation, chatbot development, language translation, and more. It can be used for a variety of tasks, and largely depends on how each user chooses to use it.

So while you can use it for factual answers, I personally wouldn't.

It's not a search engine for up to date info, but it CAN work as a search engine for stuff you want to know how to do. I use it to teach me stuff I need for my job that I might be doing for the first time or I'm not sure how to execute. Prior to CHatGPT I'd be googling for articles and reading then putting the info togehter to form a pathf forward myself. Now, I ask ChatGPT to teach me how to do something step by step instead, and it collates the information for me. So for example today, I've asked it to explain regression modelling to me which it did, and asked it for the excel formula for R squares etc etc. It saves me a lot of time, I can ask questions if there's any part of it I don't understand immediately etc. I can ask it to re-word or rephrase parts. It saves me hours.

The purpose of ChatGPT is to have “human like” conversations - It certainly does, and it will argue the point when it’s knowledge is out of date - just like some humans who think that they are right
(that’s the Chat part)

GPT is a “Generative Pre-trained Transformer”
It is a pre-trained model (ChatGPT4o was ‘trained’ as at October 2023) and uses the ‘knowledge’ at that time, but that only forms the basis of its knowledge, the purpose of ChatGPT is to express that knowledge in conversational terms. It can ‘learn’ more, but how it does that will vary

With ChatGPT 3 we were demonstrating that with the sovereign and prime minister. When telling it that the answers Queen Elizabeth the Second and Boris Johnson it would ‘double check’ but keep coming back and arguing the same answers as being right. This is because they were absolute facts which were correct as per its ‘training’. The ‘checking’ it would do would be the interpretation of the right way to answer questions (going back to the language model) but as far as it was concerned Elizabeth and Boris were absolute facts

It's a large language model. All it's supposed to do is return responses that are the most statistically likely combination of words to sound natural in response to your prompt.

It doesn't understand your question and it doesn't understand the answer it gives you.

LLMs were developed for foreign language translation.

Does it cite its sources and do you verify those sources aren't invented or incorrect?

ChatGPT is great at making stuff up, including fake citations.

I asked the exact same question and got the correct answer 🤷‍♀️

ChatGPT isn’t a finished product and is frequently incorrect. The press sometimes reports on funny inaccuracies on other platforms like Alexa too so it’s not limited to ChatGPT.

It is strange that asking a question slightly differently yields a very different result.

The belief that AI will take over baffles me. It can’t tell identical twins apart in the airport and resulted in a manager plus security and an officer to come and work out how one person had boarded twice. She hadn’t but AI thought dtd1 was dtd2 so logged dtd2 had boarded twice. Twins aren’t that rare.

Chat gpt free version is using data that is months out of date. You should be aware of that before using it.

ChatGPT is great at making stuff up, including fake citations.

Yes, that's very revealing.
I think it doesn't really 'understand' what it's being asked, it just knows what form the answers should take. It may be able to produce something that looks like an authoritative document with correctly formatted reference but it doesn't understand what is really required, and afaik doesn't know which particular multiple inputs went into its output.

And I'm pretty sure it doesn't have access to many proper academic journals which require login or pay per view.

As @TheSilkWorm says depends a lot on what model you’re using on ChatGPT, when I get downgraded for a while when ice been maxing it out too much I’m always gutted to drop to 3 as it’s such a huge (worse) difference

Garbage in, garbage out.

It is learning from all the drivel posted on X, Mumsnet and the Daily Mail as well as more respected sources like Viz Magazine and Internet erotica.

It is going to be the most delightfully moronic thing in time

I actually really like Claude (Anthropic) - it's really good at writing tasks - I find it a lot better than ChatGPT.

I've also started using Perplexity as a substitute for Google search and been pretty impressed with it!

Its usually excel based stuff i use it for, so the proof is if it works when i execute it, or not. If i ask it for something like an explanation then yes, I ask it for the source and then check the source exists and chatGPT is summarising it properly via a skim read. It's still quicker!

@TheSilkWorm this wasn't the free version though, it was the £20 per month one.

My husband used it at the weekend for filling it a work document and it saved him lots of time and did a good job

It’s very important to recognise that one of the limitations of ChatGPT is that it is incapable of knowing or caring if the information it produces is truthful. It is a predictive language model, trained on data which may or may not be accurate. It cannot independently verify or assess the data on which it is trained, or the answers it produces. It creates plausible-sounding responses to prompts based on the information it gleans from the internet. That’s all.

It is very useful for some things (generating administrative letters or emails for example, or condensing specific blocks of text into short summaries). It should never be used for anything intended to provide advice or information unless its responses have been thoroughly independently verified by a human mind.

My chatGPT answered: The word "strawberry" contains three R's.

People really do fundamentally misunderstand what chatGPT is.

It's a machine learning algorithm, trained on whatever data is fed to it with an admitedly very good natural language processor sitting on top.

It is not, in any way, AI.

It really depends on what your imagined purpose for it is.

I mean, if you think something that takes months to train on a static dataset at enormous expense is the right place to get your current affairs news...well, you're just logically wrong.

If you think it's useful as a supervised wingman for getting stuff done, then you're much closer to being able to use it as a productivity tool.

As with all technology...when you understand a bit of how it works and how it's built, you can use it more effectively.

I told mine that there was a thread on Mumsnet stating it had given the wrong answer about presidential candidates, and it replied "I understand there's been some confusion regarding my previous statements about the U.S. presidential candidates. As of November 4, 2024, the major candidates for the U.S. presidential election are:

• Democratic Party: Vice President Kamala Harris
• Republican Party: Former President Donald Trump

President Joe Biden announced his decision not to seek re-election earlier this year, endorsing Vice President Harris as the Democratic nominee. The Sun

I apologize for any earlier inaccuracies and appreciate your understanding."

So, at least it is sorry for what it has done. 😄