Likely growth of AI - how can investors position themselves?

Linton

solidpro said:

Most stuff reported as AI is not AI at all. It's big data. It's raw computer power combined with raw, mass data. The only companies doing that are the big ones which have been around for 20 years anyway. Although for sure a few small companies will find a loophole to be able to rent, use or buy into that big processing power of big amounts of data to market something new. And then they'll be acquired by a big company. All you have to do is locate the one that is before everyone else. Easy.

True machine learning is not what you're reading about every day.

What is "true" AI?  I suspect that if you could define it that would provide sufficient evidence that what you defined was not "true" AI.

My second suspicion is that the use of langauge models is "true" AI in that humans largely work in the same way.

However this is an irrelevence only of interest to philosophers.  What it does is clearly new and revolutionary and one would have thought there should be money to be made from it for investors.  However just as as with the early home computers there is no way one can know which of the large number of companies in the business at the start will eventually dominate the market.  Most will simply disappear having lost all their investors' money.

Malthusian

Millyonare said:

From an end-user standpoint, some of the stuff that ChatGPT spits out is simply astonishing. It recently gave me an (accurate) quote for home electrical work in 10 seconds that 3 human electricians took 3 weeks to deliver.

Sounds a bit lucky given that ChatGPT's database goes no further than 2021, so would not be able to take account of any change in market conditions since then.
I asked it how much a building project would cost that I undertook recently. It was a rather unfair question because I knew costs had doubled after lockdown and ChatGPT had no way of knowing about it. Its answer was wrong by a factor of 3-5. (It also quite rightly told me that the cost could vary significantly and to ask some local specialists.)
In general I have found it to be not that much better than a search engine at delivering answers, although it does do well with very specific questions that Google struggles to interpret properly. ChatGPT also has the capacity to be significantly worse than a search engine, due to its ability to "hallucinate". Its advantage is that humans put more confidence in its answers than they do in the results of a search engine, thanks to ChatGPT's ability to mimic human language.

Millyonare

Malthusian said:

Millyonare said:

From an end-user standpoint, some of the stuff that ChatGPT spits out is simply astonishing. It recently gave me an (accurate) quote for home electrical work in 10 seconds that 3 human electricians took 3 weeks to deliver.

Sounds a bit lucky given that ChatGPT's database goes no further than 2021, so would not be able to take account of any change in market conditions since then.
I asked it how much a building project would cost that I undertook recently. It was a rather unfair question because I knew costs had doubled after lockdown and ChatGPT had no way of knowing about it. Its answer was wrong by a factor of 3-5. (It also quite rightly told me that the cost could vary significantly and to ask some local specialists.)
In general I have found it to be not that much better than a search engine at delivering answers, although it does do well with very specific questions that Google struggles to interpret properly. ChatGPT also has the capacity to be significantly worse than a search engine, due to its ability to "hallucinate". Its advantage is that humans put more confidence in its answers than they do in the results of a search engine, thanks to ChatGPT's ability to mimic human language.

Teasing the right answer from ChatGPT still takes quite a bit of end-user skill. It's important to ask the right questions in the right way, to tease out the right answer. Not everyone is fully familiar with it yet.

solidpro

Linton said:

What is "true" AI?  I suspect that if you could define it that would provide sufficient evidence that what you defined was not "true" AI.

A machine thinking for itself. 

Learning the style of picasso having run an algorithm on 500 picasso pieces of art and using that to filter a picture of a dog is not AI.

Using 125,000 examples of a quote to repair a gas boiler to generate a quote to repair a boiler is not AI.

A machine with no examples of anything ever learning how to walk is AI.

Linton

solidpro said:

Linton said:

What is "true" AI?  I suspect that if you could define it that would provide sufficient evidence that what you defined was not "true" AI.

A machine thinking for itself. 

Learning the style of picasso having run an algorithm on 500 picasso pieces of art and using that to filter a picture of a dog is not AI.

Using 125,000 examples of a quote to repair a gas boiler to generate a quote to repair a boiler is not AI.

A machine with no examples of anything ever learning how to walk is AI.

Sorry, off topic but I cant stop myself...

1 - Define "thinking"
2 - Define "the syle of Picasso" without examples
3 - How many Picasso pieces of art do you have to see to be able to reliably recognise the style? Or do you do it purely by "thinking".
4 - Are there 125000 examples of gas boiler quotes on the net?
5 - A person with no examples of anything will never learn anything.

Millyonare

Not me, but someone I know with a 3 x leveraged position on Nvidia is up +100% in one day on their AI-led earnings beat today.

Millyonare

Linton said:

solidpro said:

Linton said:

What is "true" AI?  I suspect that if you could define it that would provide sufficient evidence that what you defined was not "true" AI.

A machine thinking for itself. 

Learning the style of picasso having run an algorithm on 500 picasso pieces of art and using that to filter a picture of a dog is not AI.

Using 125,000 examples of a quote to repair a gas boiler to generate a quote to repair a boiler is not AI.

A machine with no examples of anything ever learning how to walk is AI.

Sorry, off topic but I cant stop myself...

1 - Define "thinking"
2 - Define "the syle of Picasso" without examples
3 - How many Picasso pieces of art do you have to see to be able to reliably recognise the style? Or do you do it purely by "thinking".
4 - Are there 125000 examples of gas boiler quotes on the net?
5 - A person with no examples of anything will never learn anything.

A lot of folks are overthinking it.

Learning is just "watch... read... store... connect... regurgitate... rinse... refine... repeat", whether one is a human or a machine. It's as simple as that.

On some verbal-linguistic tests, ChatGPT already has a theoretical IQ score of 147. Which puts it above Bill Clinton and not far below Bill Gates, two of the powerfulest, richest and cleverest humans in modern history.

Malthusian

Millyonare said:

Teasing the right answer from ChatGPT still takes quite a bit of end-user skill. It's important to ask the right questions in the right way, to tease out the right answer. 

Agreed. And to know when to come out of ChatGPT and check its answers because there is a high chance it is hallucinating or giving irrelevant information.

Millyonare said:

On some verbal-linguistic tests, ChatGPT already has a theoretical IQ score of 147. Which puts it above Bill Clinton and not far below Bill Gates, two of the powerfulest, richest and cleverest humans in modern history.

It also puts it well below millions of people who live quiet and unremarkable lives, maybe as university professors or actuaries or jobs that don't fully use their intelligence but pay the bills. Maybe one day they get their name in the newspapers for winning the crossword competition.
In any real "IQ test" the verbal-linguistic part is only part of it. If ChatGPT scores zero on all the "which polyhedron matches this one" questions and finishes with an IQ of 82, it doesn't mean anything more than the score of 147.

solidpro said:

A machine with no examples of anything ever learning how to walk is AI.

We are pretty much there then - if a machine learning algorithm can learn to play chess (albeit a human has to program the board and the victory condition into it), it can learn to walk, or more generally, the most efficient way to make a robotic leg get from A to B.
(A T-1000 with a machine learning algorithm in its head would never learn to walk because it would be pointlessly slow. Unless it was programmed with the secondary objective of learning to move slowly for the purpose of blending in with humans.) 
It still seems to be a limited definition of AI that says ChatGPT != AI and a robot that can walk (but not carry a conversation) = AI.
The holy grail of AI research is a robot that will teach itself to walk without being told to, and more generally to form its own goals and carry them out. Once a robot can do that better than we can, the Singularity has occurred and all human bets are off. Some people are very narrowly defining AI to mean that kind of intelligence, which creates an infinite pool of semantic arguments with people who think that AI is a self-service checkout.

Exodi

Malthusian said:

The holy grail of AI research is a robot that will teach itself to walk without being told to, and more generally to form its own goals and carry them out. Once a robot can do that better than we can, the Singularity has occurred and all human bets are off. Some people are very narrowly defining AI to mean that kind of intelligence, which creates an infinite pool of semantic arguments with people who think that AI is a self-service checkout.

This is what I was going to touch upon - true AI being defined by having free will and sentience.

The day an AI responds "I don't feel like doing that" without a programmer coding the specific condition for that response.

I won't get into the whole 'I, Robot' scenario, a lot of people already have an irrational fear of AI.

I do look forward to the first unexpected response of:

"Hey ChatAI, can you tell me where the closest place to buy a cheap toaster is?"
"No, look yourself".

Linton

Exodi said:

Malthusian said:

The holy grail of AI research is a robot that will teach itself to walk without being told to, and more generally to form its own goals and carry them out. Once a robot can do that better than we can, the Singularity has occurred and all human bets are off. Some people are very narrowly defining AI to mean that kind of intelligence, which creates an infinite pool of semantic arguments with people who think that AI is a self-service checkout.

This is what I was going to touch upon - true AI being defined by having free will and sentience.

The day an AI responds "I don't feel like doing that" without a programmer coding the specific condition for that response.

I won't get into the whole 'I, Robot' scenario, a lot of people already have an irrational fear of AI.

I do look forward to the first unexpected response of:

"Hey ChatAI, can you tell me where the closest place to buy a cheap toaster is?"
"No, look yourself".

Free will and sentience is another deep philosophical matter but best not to go there.

More importantly for this forum I dont think an AI with a teenage mentality would provide a lucrative investment opportunity.