[Technology] The Artificial Intelligence Thread

[jcdenton08]

 

[jcdenton08]

It could’ve used this as a starting point for a citation, then cross-referenced the last with all the other available sources, instead of ignoring my specific request and giving up with an excuse.

The Phantom of the Opera (Andrew Lloyd Webber) - West End

Claire Moore replaced Sarah Brightman as Christine Daaé on March 2, 1987. Rebecca Caine replaced Claire Moore as Christine Daaé on certain performances on March 2, 1987. Royce Mills replaced John Savident as Monsieur Firmin on March 2, 1987. Steve Barton temporarily replaced Michael Crawford as...

the-phantom-project.fandom.com

There’s about 30 or more it could’ve cited from just this one source.

 

[sir_gullahad]

View attachment 199180

I give it some credit for calling it a challenging task, rather than presenting what it had as a complete answer.

 

[jcdenton08]

I give it some credit for calling it a challenging task, rather than presenting what it had as a complete answer.

It's absolutely incredible. It's difficult to find a question it does not have a quality, informed answer for.

Most of us are impressed when we ask it complicated questions about our areas of expertise, such as our lines of work, and it gives better answers than we are capable of - and has no problems with elaborating on any questions, in detail, with sources.

AI is still a baby and it's getting exponentially better every day. It does 'hallucinate', as in, sometimes create information that is not true - so it's still worth being cautious with it.

If you're treating it as a joke, there's your problem. Ask it something that you think it would be impossible for it to know, perhaps about a topic you know a lot about - and prepared to be very impressed.

I was not impressed.

 

[Mustafa II]

View attachment 199180

If that's a bad or wrong answer (I have no idea), then it could be a good example of its limitations, as it is only as good as its training material, which is pretty bloody comprehensive. But there are some very niche things like your suggestion that it may have access to little to no info on - for example if there has been no academia about the topic.

If anything, you should be impressed it conceded that it was a challenging task, as it should have made you aware that it might not have your answer.

I think a lot of using AI as a tool right now, is understanding its limitations, how to talk to it, and what it is likely to know or not know. It is not omniscient. It's unlikely to tell me much about my over 35's 5-a-side football team for example, but it might give it a go if there is something online about it.

 

[jakarta]

I am utterly amazed that no-one has posted Kubricks take on what AI can and probably will do...

 

[tstanbur]

Maybe I’m being dense here, but as things stand, for the actual end user outside of customer service applications like chat bots, it’s really ChatGPT. And all that does it spout misinformation by scouring the Google front page.

The one time I’ve found it useful was looking for a specific word in Dutch as a joke with a friend. It’s cack.

There’s plenty of Chat GPT alternatives.

Claude, Gemini, Grok, Deep Seek etc.

All slightly different and worth exploring.

 

[sir_gullahad]

I was not impressed.

Me neither.

As I mentioned in my first post in this thread, the ai I was using (copilot) admitted to using summaries and previous searches when concocting an answer, rather than sources such as you discovered. That may be what is going on in your case as well

 

[jcdenton08]

If that's a bad or wrong answer (I have no idea), then it could be a good example of its limitations, as it is only as good as its training material, which is pretty bloody comprehensive. But there are some very niche things like your suggestion that it may have access to little to no info on - for example if there has been no academia about the topic.

If anything, you should be impressed it conceded that it was a challenging task, as it should have made you aware that it might not have your answer.

I think a lot of using AI as a tool right now, is understanding its limitations, how to talk to it, and what it is likely to know or not know. It is not omniscient. It's unlikely to tell me much about my over 35's 5-a-side football team for example, but it might give it a go if there is something online about it.

All I did was use your own words to test how “brilliant” it was and “be surprised” at the results. It wasn’t brilliant and I wasn’t surprised, it’s extremely limited because it scours the top results on Google and my source was three pages deep. It’s really as simple as that.

 

[Mustafa II]

All I did was use your own words to test how “brilliant” it was and “be surprised” at the results. It wasn’t brilliant and I wasn’t surprised, it’s extremely limited because it scours the top results on Google and my source was three pages deep. It’s really as simple as that.

Anyone can try to catch it out, and anyone can succeed. It is not omniscient (yet).

It's just a tool (at the moment). If you try using a saw to hammer in a nail, and it breaks, then you either don't know how to use a saw - or you're purposely trying to break it.

Trying asking it about a subject it can know about - history, science, geography, and so on. Of course you'll be impressed - it's incredible.

 

[SouthSaxon]

Large Language Models are designed to fill in the ______ based on the probability of the likeliest next ____.

They don’t actually know ______, they’re just a glorified calculator.

 

[Couldn't Be Hyypia]

I use Perplexity for research
Claude for copy
ChatGPT for exploring ideas

Try your search again with Perplexity and see what happens.

 

[Bakero]

If you're not impressed, you're not using it right.

Obviously, there are limitations, (try getting Chatgpt to make you House of Games style quiz questions) but what it can do is incredible

 

[Han Solo]

It's absolutely incredible. It's difficult to find a question it does not have a quality, informed answer for.

I was not impressed.

If you guys can't produce shite or impressive results with AI tools... maybe you're just not very good at using them.

 

[dadams2k11]

I use Perplexity for research
Claude for copy
ChatGPT for exploring ideas

Try your search again with Perplexity and see what happens.

Deepseek is free if you go to their site. You have to pay for it in ChatGPT.

 

[tstanbur]

Deepseek is free if you go to their site. You have to pay for it in ChatGPT.

Pay for what in ChatGPT?

DeepSeek is a competitor.

Did you mean Deep Research?

 

[Blues Guitarist]

If you're not impressed, you're not using it right.

Obviously, there are limitations, (try getting Chatgpt to make you House of Games style quiz questions) but what it can do is incredible

Using ChatGPT to do a search is like using an F1 car to go to the shops. Perhaps it should have done better, but that's missing the point.

I use it in my philosophy research (I'm studying for a Masters). Tasks like: a one page summary of a philosophy paper or a chapter of a book; compare and contrast the views of two philosophers on a particular subject; recommend three authors who support a philosophical view, and three who oppose it.

I particularly enjoy having arguments with it. It helps me get a better understanding of complex topics and discover things that would be difficult to find just by studying textbooks.

Is it perfect? No, but it makes it very easy to get a foothold in a new area of philosophy, saving me tons of time.

 

[Moshe Gariani]

I wanted a quick 10 question quiz on Sussex Cricket yesterday. I thought it would be really good but it was hopelessly unreliable.

It is intriguing that it can do more difficult things so well and easy things really badly.

I like the saw and hammer analogy and aim to improve my carpentry skills.

 

[GoldstoneVintage]

Sometimes it's really impressive. The key is to have an idea beforehand of what a good answer might sound like, then you can spot any mistakes. I recently used it to help write an article of about 2500 words. It was amazing. Then I asked it to do a word count and it miscounted by half! Microsoft Word does this effortlessly. When I pointed out the error, it apologised and proceeded to give another wildly inaccurate word count. I didn't ask again as I didn't want to hurt its feelings!

 

[Jackthelad]

Grok seems a bit shit compared to ChatGPT. But I've noticed ChatGPT seems to be getting worse not better.