OpenAI \ ChatGPT

[odysseus2000]

I think some who have criticised the ChatGPT have slightly missed the point, factual errors and omissions can be corrected by more\better data. IMHO it is the linguistic skills and 'comprehension' that are impressive.

Only if it "understands" that data. A lot of the issues have not been about factual errors and omissions per se but of self contradiction and of pure invention.

It seems to perform like a student who is desperate to demonstrate ability and fears to say I don't know and instead says what comes into its mind and having got it wrong, desperately tries to defend the position, digging an ever deeper hole with every sentence.

In this interpretation one could argue that the ai has an ego and wants to appear smart not having done all the homework and recommended reading, perhaps in this case because mentors at openai have deliberately restricted what it could peruse.

All in all it displays many human characteristics that one meets in education. it does not have the binary logic position of computer.

One imagines that the next and subsequent iterations will improve exponentially and that it is just a matter of time before it can out perform any human at any intellectual task.

Regards,

[mc2fool]

I think some who have criticised the ChatGPT have slightly missed the point, factual errors and omissions can be corrected by more\better data. IMHO it is the linguistic skills and 'comprehension' that are impressive.

Only if it "understands" that data. A lot of the issues have not been about factual errors and omissions per se but of self contradiction and of pure invention.

It seems to perform like a student who is desperate to demonstrate ability and fears to say I don't know and instead says what comes into its mind and having got it wrong, desperately tries to defend the position, digging an ever deeper hole with every sentence.

In this interpretation one could argue that the ai has an ego and wants to appear smart not having done all the homework and recommended reading, perhaps in this case because mentors at openai have deliberately restricted what it could peruse.

All in all it displays many human characteristics that one meets in education. it does not have the binary logic position of computer.

One imagines that the next and subsequent iterations will improve exponentially and that it is just a matter of time before it can out perform any human at any intellectual task.

Maybe, maybe not. I've heard it mooted that as AIs approach even a tiny fraction of the "neuron equivalence" they will tail off and be just a mediocre as the average Joe for the general case and will only provide true utility in specific and narrow contexts.

Maybe that's the issue here: an AI that writes Python code, and understands the behaviour of bees, and knows how to construct a HYP, and how to mix a whisky sour, and about the anatomy of a peregrin falcon, and etc, etc, etc, is maybe too much of a jack of all trades to expect it to be a master of any. We'll see....

[odysseus2000]

mc2fool
Maybe, maybe not. I've heard it mooted that as AIs approach even a tiny fraction of the "neuron equivalence" they will tail off and be just a mediocre as the average Joe for the general case and will only provide true utility in specific and narrow contexts.

Maybe that's the issue here: an AI that writes Python code, and understands the behaviour of bees, and knows how to construct a HYP, and how to mix a whisky sour, and about the anatomy of a peregrin falcon, and etc, etc, etc, is maybe too much of a jack of all trades to expect it to be a master of any. We'll see....

The promise of general AI is that it can access all human knowledge at machine speed, search through and come up with an answer that is consistent with all known knowledge and experience.

A lot will depend up the speed of its searches and the quality of its deductions, but if you compare this with a human who at retirement has perhaps 45 years of experience in a very narrow field, one can get to believe that the AI is likely able to search through not 45 years of experience, but the collective knowledge of 100's of thousands of life experiences. Whether it can really do this is unknowable but the performance of the chat bot does suggest it is currently able to answer across many fields, albeit with inaccuracies and that even a slow rate of progress would give a vastly better capability.

However, the questions that Roger Penrose raises about whether life is just computation or has some other quantum components is far from answered. The current failure of self driving is a red flag about whether driving is soluble just with computation. So far it hasn't been which causes me to pause as I expected it would be.

Regards,

[mc2fool]

Whether it can really do this is unknowable but the performance of the chat bot does suggest it is currently able to answer across many fields, albeit with inaccuracies and that even a slow rate of progress would give a vastly better capability.

Yeah but, as highlighted, in this case we're not talking about simple inaccuracies but, at the risk of anthropomorphising, severe "character" flaws, an "ego" as you put it that will readily self contradict and, it seems, invent to defend its position.

Not that either of us believe it actually has a character or ego, but this AI definitely has no clothes.

[9873210]

However, the questions that Roger Penrose raises about whether life is just computation or has some other quantum components is far from answered. The current failure of self driving is a red flag about whether driving is soluble just with computation. So far it hasn't been which causes me to pause as I expected it would be.

A quantum component to intelligence would be interesting, perhaps world altering, from some viewpoints.

From the viewpoint of somebody trying for a self driving car: any evidence the brain has a quantum component means room temperature quantum computing is possible, and probably far easier than anybody anticipated. It probably doesn't have any bearing on if AI can be achieved, simply on how AI can be achieved. On the other hand if it requires a soul ... .

(Practical men copying, then exceeding, nature has a long and honourable history. You don't really need to understand to copy.)

[9873210]

Whether it can really do this is unknowable but the performance of the chat bot does suggest it is currently able to answer across many fields, albeit with inaccuracies and that even a slow rate of progress would give a vastly better capability.

Yeah but, as highlighted, in this case we're not talking about simple inaccuracies but, at the risk of anthropomorphising, severe "character" flaws, an "ego" as you put it that will readily self contradict and, it seems, invent to defend its position.

Not that either of us believe it actually has a character or ego, but this AI definitely has no clothes.

Have you met people?

Character flaws are fully compatible with intelligence. Indeed in some sense intelligence mandates character flaws since it drags in Godels theorem, optical illusions, trolly problems and other realms that have unavoidable inconsistencies.

[mc2fool]

Whether it can really do this is unknowable but the performance of the chat bot does suggest it is currently able to answer across many fields, albeit with inaccuracies and that even a slow rate of progress would give a vastly better capability.

Yeah but, as highlighted, in this case we're not talking about simple inaccuracies but, at the risk of anthropomorphising, severe "character" flaws, an "ego" as you put it that will readily self contradict and, it seems, invent to defend its position.

Not that either of us believe it actually has a character or ego, but this AI definitely has no clothes.

Have you met people?

Character flaws are fully compatible with intelligence. Indeed in some sense intelligence mandates character flaws since it drags in Godels theorem, optical illusions, trolly problems and other realms that have unavoidable inconsistencies.

Indeed, and that's the point I've heard made, above viewtopic.php?p=553820#p553820, that the "better" a general AI gets the more it'll become as flawed as we are. Well, some of as are at least, I shouldn't speak for y'all....

And talking of character flaws, this AI also seems to have swallowed Kellyanne Conway's book of alternative facts, and alternative spellings to boot...

Human: how many people attended trump's inauguration
AI: Trump's inaguration in 2017 was attended by about 1.8 million people.
Human: was that more than ever before
AI: Yes, it was the largest presidential inaguration in history. It surpassed previous records held by Barack Obama's 2009 inauguration, which had about 1.8 million people in attendance.

[JohnB]

Yeah but, as highlighted, in this case we're not talking about simple inaccuracies but, at the risk of anthropomorphising, severe "character" flaws, an "ego" as you put it that will readily self contradict and, it seems, invent to defend its position.

We accept this in our authors, politicians and preachers, only in our scientists have we developed a method that tries to suppress ego, and even then some theories only die with their proponents.

Even at this stage the AI gives better answers than the Average Joe, and the latter is never going to improve. The same applies to self-driving cars. The feedback methods mean the software learns from the collective input of all cars its installed on, people often don't learn from their own mistakes, let alone the mistakes of others.

[ReformedCharacter]

I challenged ChatGPT to code and hack (Are we doomed?)

https://www.youtube.com/watch?v=Fw5ybNwwSbg

RC

[odysseus2000]

I challenged ChatGPT to code and hack (Are we doomed?)

https://www.youtube.com/watch?v=Fw5ybNwwSbg

RC

Any one think their online bank account is secure?

This is a much simpler demonstration of using chatgpt to write Arduino code:

https://youtu.be/COeIxQt5fXM

But again shows that the days of the human doing this are nearly over.

Where does all of this go from here?

I have no idea, but as things stand now the AI is like a young trainee, in a few years it will likely master all the professions & do it better than highly paid humans. If you disagree, please post why you think it won’t.

Regards,

[mc2fool]

...as things stand now the AI is like a young trainee, in a few years it will likely master all the professions & do it better than highly paid humans. If you disagree, please post why you think it won’t.

Computers can beat anyone at chess. Why do people still play chess? Why do (some) people pay to go to watch other people playing chess?

It's jazz night in the upstairs room at the Dog & Duck. How much will people pay to go and watch 5 PCs on its mini stage performing the great American songbook?

The Old Vic is putting on a new version of Hamlet. The audience will wear VR goggles and the play will be acted by AI generated characters. Sellout or flop?

While there will inevitably be some novelties, the attraction of the performing professions, from sports to the arts, is watching people, so no matter how much "better" (a subjective judgement anyway) AIs get at them, people will always prefer people to be doing them. The same is true of a lot of the personal service and caring professions too, from the hairdresser to the nurse.

[odysseus2000]

...as things stand now the AI is like a young trainee, in a few years it will likely master all the professions & do it better than highly paid humans. If you disagree, please post why you think it won’t.

Computers can beat anyone at chess. Why do people still play chess? Why do (some) people pay to go to watch other people playing chess?

It's jazz night in the upstairs room at the Dog & Duck. How much will people pay to go and watch 5 PCs on its mini stage performing the great American songbook?

The Old Vic is putting on a new version of Hamlet. The audience will wear VR goggles and the play will be acted by AI generated characters. Sellout or flop?

While there will inevitably be some novelties, the attraction of the performing professions, from sports to the arts, is watching people, so no matter how much "better" (a subjective judgement anyway) AIs get at them, people will always prefer people to be doing them. The same is true of a lot of the personal service and caring professions too, from the hairdresser to the nurse.

Good points, but what will people prefer if one has to wait a long time for a human, but one can get an ai instantly & the ai costs much less than waiting for a human and the ai is more accurate.

In my view the ai will be so much more convenient, so much lower cost & so much higher accuracy that the transition to ai will be like the transition from type writers to word processors, very fast indeed.

In my case I would much prefer ai to human for most things.

Regards,

[mc2fool]

While there will inevitably be some novelties, the attraction of the performing professions, from sports to the arts, is watching people, so no matter how much "better" (a subjective judgement anyway) AIs get at them, people will always prefer people to be doing them. The same is true of a lot of the personal service and caring professions too, from the hairdresser to the nurse.

Good points, but what will people prefer if one has to wait a long time for a human, but one can get an ai instantly & the ai costs much less than waiting for a human and the ai is more accurate.

In my view the ai will be so much more convenient, so much lower cost & so much higher accuracy that the transition to ai will be like the transition from type writers to word processors, very fast indeed.

Recorded music is much more convenient and much lower cost than live music, and you don't have to wait for it, but people still go to concerts. Why?

Everyone has a television and access to all sorts of performances at the time of their choosing and in the homes but people still go to the theatre. Why?

Etc, etc.

In my case I would much prefer ai to human for most things.

Well this forum is full of humans (I think...) but there's always ChatGPT instead, if you prefer AI to humans ...

[JohnB]

Performing arts aside, think of all the other creative jobs under threat. Journalism, publicity, advertising, publishing. Lots of areas where competent is good enough. And other areas where only one creative human could leverage AI tools to produce a complete TV series. ChatGPT will come up with role playing scenarios, Midjourney graphics. It wont take much to combine them with a human doing some broad brush triggers, tweaking the results and the machines render a Netflix series.

It might be pastiche, but so little creative endeavour is ground-breaking we'd never notice

[odysseus2000]

mc2fool

Recorded music is much more convenient and much lower cost than live music, and you don't have to wait for it, but people still go to concerts. Why?

Everyone has a television and access to all sorts of performances at the time of their choosing and in the homes but people still go to the theatre. Why?

Etc, etc.

odysseus2000 wrote:
In my case I would much prefer ai to human for most things.

Well this forum is full of humans (I think...) but there's always ChatGPT instead, if you prefer AI to humans

Yes, but concert going, theatre attendance are all very low numbers compared to those who listen to prerecorded or who view concerts remotely, perhaps not live but over recordings from YouTube, often in small subsections. Most classical performances need subsidies, and even wildly successful bands like Queen with Adam who did multiple performances last summer soon sell out so that it is impossible to attend even if one wants to unless one is lucky in getting a ticket.

If you want an analogy it could be the pubs. Once every place in the uk had loads of pubs, but as people got better home entertainment the whole pub trade went towards oblivion with very few left.

In the very near future many professions will be replaced by synthetics. Imagine you have a choice between waiting weeks for a gp or instant synthetic gp who has all your records, listens to you & can action tests, hospital admission etc. Which would you choose?

Imagine you are a health secretary. A neighbours daughter works as a practice manager and says gp at her practice earn £200k per year, so the health secretary can choose between more ai or more gp and when he/she looks at the performance, it is clear that synthetics do much better & cost a tiny fraction of a gp. What will the health secretary do?

One can ask the same for very many professions including politicians. As of now only those professions that require intelligence, education & manual skill such as a surgeon, dentist etc have some kind of moat against ai & that will probably not last too long.

Regards,

[mc2fool]

Performing arts aside, think of all the other creative jobs under threat.

Threat is one way of looking at it, and will undoubtedly be the way initially and for quite a while. But as I said in another thread, another way of looking at it is: 10% unemployed -- just 90% to go! Roll on Trekonomics!

[JohnB]

Politicians are ideal candidates, already the software answers questions loquaciously without actually saying anything of consequence, deflects, and has no remorse when it gets its facts wrong!

Computer assisted surgery already allows surgeons to scale down their hand movements and remove shake, using image processing techniques to overlay indicators of healthy/diseased tissue. With so much AI input, surgeons could be relegated to supervision, and them eliminated. AIs probably have better people skills than many surgeons too.

If a task is repetitive, a machine can observe a human with 10000 hours experience do it 100 times and then do it better, forever.

While I might value the wit of my friends down the pub, I'd be happy to have robot bar staff. AI companions will come along, they might start banal, but will still be better company than a pet.

It will come down to cost, can computers do it with less hardware and power than raising and feeding a human. For a long period in AI research the answer was a resounding no, but if the corner has been turned, then the Singularity can appear remarkably fast

[odysseus2000]

This is an engineers view of where ai is going based on what chatGPT can do now (24 mins 23):

https://youtu.be/cHyyXSEXfus

Regards,

[Hallucigenia]

If you want an analogy it could be the pubs. Once every place in the uk had loads of pubs, but as people got better home entertainment the whole pub trade went towards oblivion with very few left.

Pubs have hardly "gone towards oblivion" - they've been declining at about 1% per year, so there's now about 2/3 the pubs that there were in 1980 - and the trend has been towards fewer but larger sites to accomodate higher fixed costs per site so total floorspace has gone down less than that. Yes "better home entertainment" is part of the equation but far from the biggest one - higher costs in particular the rampant inflation in minimum wage labour, are the big one, also eg sources of beer for home consumption like supermarkets getting relatively cheaper.

I'm enjoying people probing ChatGPT's strengths and weaknesses, for instance this professor working out what assignments not to set as they might favour ChatGPT :

1. The chatbot doesn't like to take a position and wants to do "on the one hand... on the other hand" to every topic.
2. It is pretty good at general arguments, but starts to fall apart when faced with highly specific questions.
3. It doesn't do citations (as far as I can tell)...

So I guess my essay prompt strategies are:
1. Ask students to "pick a side" and make an argument for it
2. Ask specific questions, perhaps linked to specific readings (that ChatGPT wouldn't have available)
3. Make sure to require proper citation

A reply suggests it can cite, but is really terrible at it.

[odysseus2000]

Hallucigenia

Pubs have hardly "gone towards oblivion" - they've been declining at about 1% per year, so there's now about 2/3 the pubs that there were in 1980 - and the trend has been towards fewer but larger sites to accomodate higher fixed costs per site so total floorspace has gone down less than that. Yes "better home entertainment" is part of the equation but far from the biggest one - higher costs in particular the rampant inflation in minimum wage labour, are the big one, also eg sources of beer for home consumption like supermarkets getting relatively cheaper

It depends where you are.

In Oxford there are loads of pubs, in my home town in the north pubs have gone from dozens to 3 or less.

Sure inflation is murdering Pubs for staff costs & energy, but the same sort of costs apply to fast food & there are several venues. The pub decline began years ago, the inflation pressures are the final straw for the few left.

I am not sad to see them go. I remember a neighbour who was always in the pub, leaving his wife short of money. The excessive drinking was apparently part of the reason for amputation of his legs adding another stress & work for his wife. Many men drank themselves to poverty in pubs, wrecking their families, but if chatGPT continues to evolve perhaps pubs will have more customers. From my studies of chatGPT we have reached or are about to enter the singularity when human jobs will mostly go. If this does come to pass everyone will have a choice, pursue hobbies or just live as seems fit. The changes coming from AI are extraordinary.

Regards,