Why are we hearing more about AI now?

[Citybreak1]

Seemingly Siri has been AI for years and we have had picture editors for years able to make people look smooth with special effects? And even tools on eBay to edit quick auctions nothing new so why the big fuss? Why was this never spoken about years ago?

 

[nlogax]

ChatGpt and other implementations of generative AI have made some significant advances over the last twelve to eighteen months and have become far more accessible to people outside of scientific and academic circles. This is less of a sudden moment, more of a threshold about which many in the industry have definitely been talking up for years.

 

[dosxuk]

Same reason there was all the talk about NFT's a while ago - there's easy money to be made by the lucky few who back the right horse, provided they can get a line of mugs investors to keep shovelling money in.

Sure, there's been some actual technical achievement in the last 18 months, but it no way justifies the current hype.

 

[greatkingrat]

Siri / Alexa isn't really AI in the same sense that ChatGPT is. If you ask Siri to tell you a joke it will just pick from a pre scripted list, while ChatGPT can come up with original jokes by itself.

 

[jon81uk]

Seemingly Siri has been AI for years

Siri isn't AI, it offers set answers.

 

[RichJF]

The mainstream media is more aware of it. They have also got hold (both rightly & wrongly on varying examples) of buzzwords & are slinging them into stories, therefore the general public absorbs it more.

 

[birchesgreen]

Next year something else will be cool and AI will revert back to interesting and useful background tech and less useful foreground tech.

 

[boiledbeans2]

Siri isn't AI, it offers set answers.

Siri is AI, but not in the sense of ChatGPT.

Siri has been trained to understand and interpret what someone is saying, e.g. different accents, voices, inflection, etc. Then subsequently fetches answers from relevant topics/lists or performs a search on the Internet for the answer.

On the other hand, ChatGPT is a "smarter" AI, able to pull answers from the Internet as well as generate its own (which might be incorrect).

 

[DoubleLemon]

There are many types of ai. It's a catch all term. the ghosts in the original pac man are considered ai.

The computing power available and creation of large and really large models and having the Internet to draw on has changed things.

At my work we are really going in ai in a big way. The problem is finding a use for it. Now it's been about fir a while people are using it for practical er... Uses.

There is analytical ai and predictive ai (which is what chat gpt is at heart)

Im putting a proposal together to use ai to first analyse a change in our software or a change in the help to see if it reduces support calls. Then a model to predict what changes will have the biggest positive impact on our customers.


We can do this by hand, and we do but automating this will allow us to produce a lot more effective changes to make the software easier to use.

Next year something else will be cool and AI will revert back to interesting and useful background tech and less useful foreground tech.

I don't thinkso. I think it's going to have the same impact as the Internet and smartphones had to the world.

 

[Puffing Devil]

The technology has now developed and moved into the mainstream - like penicillin moving from the Petri dish to a production line. Tasks that required much skilled human effort are now being completed by machines with limited human instruction.

A consumer-level view is MS Copilot - it's worth reviewing the now available capabilities. Excel was the "killer app" for the PC, Co-Pilot may well be the killer app for personal productivity.

This is just the beginning - so many applications are being developed to take advantage of improved machine learning and the now available massive computing power required to drive it. I asked Google Bard to show the increase in computing power from the 1980s to now. It said this:

Computing power has increased exponentially over the past four decades. In 1980, the most powerful computer in the world, the Cray-1, could perform about 133 megaflops (millions of floating-point operations per second). Today, the most powerful computer in the world, the Fugaku, can perform over 415 petaflops (quadrillions of floating-point operations per second). This is an increase of over a million times in computing power.

I also used Grammarly to check the language I used in this post.....

 

[Bantamzen]

Because it wants to be heard....??

 

[najaB]

Seemingly Siri has been AI for years and we have had picture editors for years able to make people look smooth with special effects? And even tools on eBay to edit quick auctions nothing new so why the big fuss?

None of those examples are generative AI. They're all just examples of a machine following a script.

Why was this never spoken about years ago?

Because generative AI is completely different. Machine learning, large language models and massive data sets mean that modern AI such as Dal-E and ChatGPT are able to extrapolate from the data sets they're given and produce* completely new things in answer to questions/prompts.

*Call me a luddite, but I'm still not ready to use the word "create".

 

[McRhu]

How do I know that I'm not the only human here and you're all AIs luring me in to the digital netherworld?

 

[AdamWW]

How do I know that I'm not the only human here and you're all AIs luring me in to the digital netherworld?

We don't know that you're human, do we?

 

[McRhu]

We don't know that you're human, do we?

But... but.. I must be. I can do those little CAPTCHA tests and pick out all the squares with umbrellas and things.

Actually this reminds me of that film "The Thing" where nobody knows who's the alien and everybody's suspiciously eyeing up everyone else. A taste of things to come maybe.

 

[AdamWW]

But... but.. I must be. I can do those little CAPTCHA tests and pick out all the squares with umbrellas and things.

And that's what sets us apart from the machines.

Not the ability to be creative, to feel, or to fall in love.

No....we can find all the squares with a bicycle in.

I'm a bit fed up with bicycles. And American pedestrian crossings and traffic lights. I did get trains once. That was better.

 

[birchesgreen]

I don't thinkso. I think it's going to have the same impact as the Internet and smartphones had to the world.

Oh its having a big impact but not so much in the areas getting the hype which are more gimmicky.

 

[Mcr Warrior]

On the other hand, ChatGPT is a "smarter" AI, able to pull answers from the Internet as well as generate its own (which might be incorrect).

Think we've had a separate thread on this previously, but any AI system that routinely suggests incorrect answers, to legitimately asked questions, doesn't seem to me to be all that intelligent, and so, might the conclusion be that someone, somewhere, is taking users for mugs?

 

[DoubleLemon]

But... but.. I must be. I can do those little CAPTCHA tests and pick out all the squares with umbrellas and things.

Actually this reminds me of that film "The Thing" where nobody knows who's the alien and everybody's suspiciously eyeing up everyone else. A taste of things to come maybe.

There's a report of a test ai with access to the Internet try and hire someone from fivr to complete the CAPTCHA

Oh its having a big impact but not so much in the areas getting the hype which are more gimmicky.

People expected it to take on more process driven tasks but it seems to be moving to creative and predictive tasks.

The public usies of it are gimmicky and people struggle to find a use for it. I'm involved in building a custom learning model with our company ai person based on internal data. It's really interesting and very over my head.

Think we've had a separate thread on this previously, but any AI system that routinely suggests incorrect answers, to legitimately asked questions, doesn't seem to me to be all that intelligent, and so, might the conclusion be that someone, somewhere, is taking users for mugs?

Hallucinations are an issue and more so with chat gpt4. It's not an easy one to solve for a general predictive ai with the whole Internet as its source. Most companies using ai use specific data sets from my understanding. It depends on the purpose.

 

[najaB]

Think we've had a separate thread on this previously, but any AI system that routinely suggests incorrect answers, to legitimately asked questions, doesn't seem to me to be all that intelligent

Have you seen some of the answers posted on this forum by allegedly intelligent humans?

 

[Mcr Warrior]

Have you seen some of the answers posted on this forum by allegedly intelligent humans?

Some of the duff postings occasionally coming from me, but at least on this forum, it normally doesn't take too long for such "hallucinations" (or mistruths, or complete and utter fabrications, or whatever they are) to get called out as being B.S.

 

[najaB]

Some of the duff postings occasionally coming from me, but at least on this forum, it normally doesn't take too long for such "hallucinations" (or mistruths, or complete and utter fabrications, or whatever they are) to get called out as being B.S.

And AI inventions are usually called out just as quickly by people in the know. The danger - with both AI-generated and human created - is when the duff info is taken as gospel. I suppose the risk is that some people are less likely to exercise scepticism on answers they get from a computer, vs a "guy on the Internet".

 

[AdamWW]

Hallucinations are an issue and more so with chat gpt4. It's not an easy one to solve for a general predictive ai with the whole Internet as its source. Most companies using ai use specific data sets from my understanding. It depends on the purpose.

But so far as I can see the hallucinations don't come from errors in the datasets used - they're an intrinsic part of how the current models work by throwing together words that look reasonable together (vast oversimplifcation).

ChatGPT spent quite a bit of time trying to convince me that you can do wheelies on a unicycle. I don't think that's because it "read" it somewhere. I think it had unicycles lumped in with two wheeled vehicles and wasn't distinguishing the critical difference.

 

[Stephen42]

But so far as I can see the hallucinations don't come from errors in the datasets used - they're an intrinsic part of how the current models work by throwing together words that look reasonable together (vast oversimplifcation).

ChatGPT spent quite a bit of time trying to convince me that you can do wheelies on a unicycle. I don't think that's because it "read" it somewhere. I think it had unicycles lumped in with two wheeled vehicles and wasn't distinguishing the critical difference.

Yes, even with far more trimmed down data sets generative AI can hallucinate. I've seen examples where given a library of case studies requests have ended up fusing different case studies into a combined effort that doesn't reflect anything like what the company has previously done.

That doesn't mean it isn't powerful and a huge step forward. Many more organisations can now use tooling like that productively to reduce human effort required for various tasks. Going from a sketch with key points to fully formed email/article/proposal is near immediate requiring only verification/tweaking rather than writing from scratch. For searching vast arrays of data generative AI when given the right commands can identify a broader range of relevant results than keyword searching alone ever could. It's already pretty good at refactoring code. There will be a period of people getting used to its capabilities and understanding its weaknesses, but I can't see it going away.

 

[DoubleLemon]

But so far as I can see the hallucinations don't come from errors in the datasets used - they're an intrinsic part of how the current models work by throwing together words that look reasonable together (vast oversimplifcation).

ChatGPT spent quite a bit of time trying to convince me that you can do wheelies on a unicycle. I don't think that's because it "read" it somewhere. I think it had unicycles lumped in with two wheeled vehicles and wasn't distinguishing the critical difference.

We are having to take this into account for a test chat bot using chat gpt. You cant delete feature X in the app. You ask it the question and it will tell you no. You then ask it to double check a few times. I think what is happening is its looking at something similar. You can delete feature Y & Z so it seems to think because Y&Z can do it - eventually it thinks you can do it in X.

I think the longer the conversation thread the the more it hallucinates. Its something you don't want to happen with software that contains legal stuff. Its all very new and there are ways to do it, I think we may need to think about how we write content to take it into account.

 

[AdamWW]

We are having to take this into account for a test chat bot using chat gpt. You cant delete feature X in the app. You ask it the question and it will tell you no. You then ask it to double check a few times. I think what is happening is its looking at something similar. You can delete feature Y & Z so it seems to think because Y&Z can do it - eventually it thinks you can do it in X.

I think the longer the conversation thread the the more it hallucinates. Its something you don't want to happen with software that contains legal stuff. Its all very new and there are ways to do it, I think we may need to think about how we write content to take it into account.

My unicycle experience was the other way round. It suggested wheelies in the first place. It took me a while to get it do admit it was talking rubbish.

 

[DoubleLemon]

My unicycle experience was the other way round. It suggested wheelies in the first place. It took me a while to get it do admit it was talking rubbish.

That's interesting - was it part of a longer chat thread?

 

[MasterSpenny]

An answer to the thread title is because of more technology being created when it comes to digital chats with you and a computer of some sort and generating responses to what you asked.

 

[Purple Train]

My unicycle experience was the other way round. It suggested wheelies in the first place. It took me a while to get it do admit it was talking rubbish.

I'm surprised it didn't give a rambling and overly saccharine conclusion. It does that to me even when I'm asking it to define cheese.

 

[DoubleLemon]

I'm surprised it didn't give a rambling and overly saccharine conclusion. It does that to me even when I'm asking it to define cheese.

To be fair an overly sweet response could also be considered a cheesy response... Therefore it's technically giving a right answer. Hahaha