The AI dooooooom thread

[Jacob]

Snapchat apparently has introduced a ChatGPT "friend" in friend-groups, including to children.

[HVC]

Kids use snap chat? 5hought.it was a 20s flirting app  

[viper37]

Lawyer uses ChatGPT to prepare its case

It didn't go well... 

Apparently, the "AI" has invented cases out of thin air and the lawyer's verification was to simply ask ChatGPT if they were real.

[Syt]

I fortunately learned this with something harmless like book recommendations.

[viper37]

I fortunately learned this with something harmless like book recommendations.

Yes, it's better to start small

Quebec bar put it to the test this week, submitting the AI to their bar test.  It got 2/10, this time too, inventing things that weren't true, and failing miserably on lawyer-client priviledges.

[Maladict]

The AI is a terrible liar, which probably is a good thing.

[Sheilbh]

This is me being lawyerly but surely feeding it with the details it would need to make a submission would be a breach of client confidentiality?

[crazy canuck]

This is me being lawyerly but surely feeding it with the details it would need to make a submission would be a breach of client confidentiality?

Yes, not only a breach but a waiver.

If you have not yet listened to Runciman's podcast on AI, you should.  The main takeaway - AI is dumb but we are easily fooled into thinking it is intelligent.  The biggest risk is humans trusting the AI to do things that require judgment.

[The Minsky Moment]

My understanding is that there have been cycles of AI Springs and Winters.  Each "Spring" period begins with some technical breakthrough and new applications that appear very impressive on first impact, but progressing significantly past that point proves difficult and the shortcomings of the new techniques become increasingly apparent.

We are clearly in a Spring period now and it is materially different from past ones in that the potential commercial applications are significantly broader than e.g. a really good AI chess opponent.  However, the outlines of the winter to come are coming into focus.  There is a ton of money flowing into AI research and applications now, but it all involves refining and building upon the basic techniques of using large masses of existing data to generate "new" data based on existing patterns.  It is basically a very sophisticated way of rearranging Titanic desk chairs. As in past iterations of AI, the "intelligence" part is arguably a misnomer because these system are just manipulating data patterns without any insight or understanding into its meaning or content. 

There are clearly enough uses cases here to justify a lot of investment, but probably not as much as is going to flow.  Thus my predictions that there will be lot of AI fortunes made and a lot of investor capital wasted.

[crazy canuck]

My understanding is that there have been cycles of AI Springs and Winters.  Each "Spring" period begins with some technical breakthrough and new applications that appear very impressive on first impact, but progressing significantly past that point proves difficult and the shortcomings of the new techniques become increasingly apparent.

We are clearly in a Spring period now and it is materially different from past ones in that the potential commercial applications are significantly broader than e.g. a really good AI chess opponent.  However, the outlines of the winter to come are coming into focus.  There is a ton of money flowing into AI research and applications now, but it all involves refining and building upon the basic techniques of using large masses of existing data to generate "new" data based on existing patterns.  It is basically a very sophisticated way of rearranging Titanic desk chairs. As in past iterations of AI, the "intelligence" part is arguably a misnomer because these system are just manipulating data patterns without any insight or understanding into its meaning or content. 

There are clearly enough uses cases here to justify a lot of investment, but probably not as much as is going to flow.  Thus my predictions that there will be lot of AI fortunes made and a lot of investor capital wasted.

I am not so sure about the Spring analogy.  AI can do things that are repetitive and churn out objective data to be analyzed.  For example, bench research is being sped up considerably by AI/robotics running respective experiments, analyzing the outcome, and tweaking the inputs for the next experiment.

But where it falls down is judgment.  There it is being described as a deadend or an offramp to something that might come in the future, decades from now.

[DGuller]

There is a ton of money flowing into AI research and applications now, but it all involves refining and building upon the basic techniques of using large masses of existing data to generate "new" data based on existing patterns.

Isn't that what intelligence is in a nutshell?

[Jacob]

There is a ton of money flowing into AI research and applications now, but it all involves refining and building upon the basic techniques of using large masses of existing data to generate "new" data based on existing patterns.

Isn't that what intelligence is in a nutshell?

Intelligence is knowing when to think outside the nutshell.

[DGuller]

There is a ton of money flowing into AI research and applications now, but it all involves refining and building upon the basic techniques of using large masses of existing data to generate "new" data based on existing patterns.

Isn't that what intelligence is in a nutshell?

Intelligence is knowing when to think outside the nutshell.

Seriously, though, I think what Minsky described is exactly what intelligence is, when you strip away the heuristics specific to humans.  Intelligence is the ability to generalize from prior experience and education in order to understand new situations that you haven't experienced before.

[Jacob]

Why would you want to strip away the heuristics specific to humans?

[The Minsky Moment]

Seriously, though, I think what Minsky described is exactly what intelligence is, when you strip away the heuristics specific to humans.  Intelligence is the ability to generalize from prior experience and education in order to understand new situations that you haven't experienced before.

Generative AI doesn't understand new situations (or indeed anything).  It doesn't have experiences and it doesn't recognize new situations.