ChatGPT - artificial intelligence.

[jetsetwilly]

It is all to do with shape recognition, think about reading. We look at the page and interpret the shapes so as to form words and these are pre stored in our memory from when we learnt to read. We can connect the words to form meaningful sentences so we can understand, AI can easily have a camera that would do the same and end up with a word but no understanding. Not a problem if you are just comparing the word to a table of words where each results in some action as in a digital system but that is not AI, to be AI it would need to just make the decision based on knowledge which is a very long way off. AI would need to read and understand in order to learn which I think is still very sci-fi at the moment.

Taking your example step by step, there are a few operations which I believe are common between human and machine text interpretation. I work on Siri, so I'll use examples from the Assistant domain.

1. Given the shapes, what are the words? Humans and machines do this in functionally-equivalent ways - essentially it's as you say, we compare the squiggles on the page to lots of learned examples and make a decision. Both humans and machines are biased towards 'common' sequences of words eg 'I need some groceries so I'm off to the shops' is much more likely than 'I need some groceries so I'm off to the ships' - both humans and AI use their previous experience of language to bias their predictions about the squiggles.
2. Given the words, what is the meaning? This is where humans have almost unlimited advantages, and where systems like chatGPT are moving the needle for machines. Does 'hold my beer' really mean that there is a beverage nearby which requires immobilisation? Sometimes yes, but usually, no. So, how do you learn what 'hold my beer' means? You read and understand the context. If it's a paragraph where somebody is literally drinking a beer and needs to relinquish it temporarily, it might be literal. More likely, you see 'hold my beer' in a meme or other high-frequency graphical motif, or at the end of a social media post. You learn it's actually a metaphor. AI can learn that too, and from the same signals, and the clever bit about GPT and other transformer-based large language model systems, is that the AI can learn it without humans having to label all of the training data - just like humans can.
3. What should be the response? Here the AI is often on more solid ground, simply because its range of responses is smaller. 'Hey X, be a a love and turn on the lights while I find my glasses' has a lot of content in it, and if you are X, you can probably construct a whole mental scene around it suggesting how the person feels about you, where they are, what's the likely event context, where they left their glasses etc etc, and therefore what is the most salient response. But if X=Siri, the most salient response is smart_lights.TurnOn(). As assistants mature, more of that sentence might be important eg 'turning the lights on, also I can see your glasses on the hall table' (creepy!).

You mentioned meaning, but also understanding, and learning. Tricky stuff! ChatGPT can learn concepts from textual data, and resynthesize and explain those concepts in new text in response to questions. Does it 'understand'? Well, first define 'understanding'

 

[Tris]

'hey Alexa, define pi'

'Do you want that by decimal places or filling?'

 

[Dibs-h]

@rwillett,

I'm not expecting an undergrad to re-write the SMTP (SMPT was a typo - LOL ) world but along IP\etc - a somewhat more than a very basic (or no) understanding would be good\useful. Basic things like how IP addresses work, the classes, sub-netting etc.

Similar with databases - I'm not expecting a fresh grad to be a data guru but to be able to design a smallish (basic) database (be it relational or otherwise) and interrogate it - but many don't know one end from another.

[Very large event driven data - the relational databases have no issue dealing with it at eye-watering scale (if it is structured - and most data in use\on the move is structured, admittedly not all). A key question is whether the data is structured or unstructured and that will shape what you use. But I digress.]

Yep - Typescript (React etc) along with it's "parent" Javascript, the demand isn't going away any time soon.

.Net - we'll have to disagree, C# is far from dead and the demand for it isn't to deal with legacy stuff.

https://www.devjobsscanner.com/blog/top-8-most-demanded-languages-in-2022/
If you talked to a fresh (other subject) graduate and you mentioned a basic principle\system (say foundations for any subject related to the built environment or some element of the human body for a medical related subject, etc.) and the person said they never studied it - I'd be shocked.

I'm not expecting a fresh computer science graduate to realistically be off and away on day #1 - but mentioning mainstream current world stuff and getting a "we didn't study that" response is far from encouraging on the quality of courses.

Undergrad courses should be preparing students for the real world. Cutting edge stuff is the realm of Postgrad courses. And the stuff that gets academic staff off - that's for PhD's LOL.

 

[Dibs-h]

ChatGPT can learn concepts from textual data, and resynthesize and explain those concepts in new text in response to questions. Does it 'understand'? Well, first define 'understanding'

Thomson Reuters - developed something (Open Calais) 10 or so years ago. When given a load of text (say an article\essay) it could identify the semantic information, i.e. who, what, etc.

Impressive for it's time and even by today's standard. More a tool for commercial use rather than public - i.e. allow companies to acquire meta-data that relates to their content (probably more those in the publishing world).

 

[selectortone]

It's interesting the level of responsibility put on teachers these days. Pretty hard to produce decent outputs when you have kids and parents that have no desire to engage with school and often actively disrupt others. Perhaps if parents put in the effort the kids might follow. Seems like schools are expected to raise peoples kids but without any level of authority over them.

My other half talking our son to school the other day saw a parent smoking a spliff on the way to school with their kid trailing behind. What hope does that kid have? My other half spoke to the school and they said they could do nothing about it and the problem is so bad with some kids that they have to leave their book bags outside as they stink so much. These are 5/6 yr olds. Likely to be a little bit baked and unlikely to have had a proper breakfast and you expect them to learn and it's the teachers fault if they don't?

I live in a rural area and it's a pretty good school, so can't imagine how much worse a bad school is.

Maybe a lot of people should ask ChatGPT how to be a better parent

Absolutely. I have a daughter who teaches A level English at a large school in Surrey. Most parents are resposible and take an interest but some families are several generations deep into an alernative universe of weed and benefits.

 

[Jacob]

Absolutely. I have a daughter who teaches A level English at a large school in Surrey. Most parents are resposible and take an interest but some families are several generations deep into an alernative universe of weed and benefits.

Have you got a bit of a chip about people on benefits? First it was tattoos and Netflix, now weed and an alternative universe?
Are you a little jealous and feeling left out?
Whatever it is you've completely missed the point - which is that it is surely a good thing that inadequate parents are seen to be bringing their kids to school, which, all being well, could improve their kids lives, and maybe even the parents' too.
n.b. there are plenty of other forms of bad parentage which don't involve tattoos, netflix or weed, and bad parentage is not a modern phenomenon.

 

[Jacob]

..... Seems like schools are expected to raise peoples kids

Well yes education is a big proportion of their upbringing. This is what schools are for and they can transform a persons prospects.

but without any level of authority over them.

My other half talking our son to school the other day saw a parent smoking a spliff on the way to school with their kid trailing behind. What hope does that kid have?

Well he was being taken to school - so that's a good thing. Education is one major route out of deprivation and cycles of poverty.

 

[selectortone]

Have you got a bit of a chip about people on benefits? First it was tattoos and Netflix, now weed and an alternative universe?

You know nothing about me Jacob so that's quite an assumption. My Dad's family goes back generations of Co Durham coalminers and my Mum is from Irish subsistence farming stock. You can't get much more working class than that.

We've been through this over and over in the naughty forum. I have no problem with benefits for people truly in need. My Grandad and Great-uncle marched and fought for and won the welfare state we have today. I'm hugely proud of them.

What I do have a problem with is when it becomes a life style. That's taking the p*iss, it's not what it's for, and it wasn't what they fought for.

That's all I have to say to you.

 

[Jacob]

You know nothing about me Jacob so that's quite an assumption. My Dad's family goes back generations of Co Durham coalminers and my Mum is from Irish subsistence farming stock. You can't get much more working class than that.

We've been through this over and over in the naughty forum. I have no problem with benefits for people truly in need. My Grandad and Great-uncle marched and fought for and won the welfare state we have today. I'm hugely proud of them.

What I do have a problem with is when it becomes a life style.

I'd say that when it becomes a lifestyle that's when they really have problems and could most end up needing the welfare state. Same with alcohol or obesity etc. etc.
I don't see the point of disapproving of tw-ats. They are not the problem - but the mega rich really are.

 

[Agent_zed]

Well yes education is a big proportion of their upbringing. This is what schools are for and they can transform a persons prospects.

Well he was being taken to school - so that's a good thing. Education is one major route out of deprivation and cycles of poverty.

hahahah well that is looking for a silver lining. I guess they might learn some maths too whilst they watch their parents buying an 8th. Drugs are one major route into deprivation and cycles of poverty and I'd argue it's a lot easier to fall into drug use than accidentally learn maths.

Schools are there to educate not for upbringing. Education forms a role in upbringing but there is only so much a teacher can do in 60mins with 30 kids. They would literally have 2mins per child per day. not a lot they can do if the kid doesn't give a monkeys about being there. I worked in a secondary school for 18months doing the IT. The amount of equipment I changed that certain kids damaged because they were bored or whatever was unreal. I have witnessed some truely awful behaviour, kids literally screaming obscenities at the teachers disrupting the entire class. They literally do not care about learning and don't want to be there. How much effort can you put into that kid vs one that wants to learn? How much effort can you put in when the parent doesn't show for parents evening or give a monkeys about the kids behaviour?

I don't advocate giving up on them entirely but what hope does a teacher have. Personally I think they should bring back metal work/wood work for the kids that clearly arent academic.

Bringing it back on topic, AI and computerising everything so there is even less jobs for people to do does concern me slightly. I guess not a lot different to people trashing the threshing machines of old as they took their manual job.

 

[Jacob]

hahahah well that is looking for a silver lining. I guess they might learn some maths too whilst they watch their parents buying an 8th. Drugs are one major route into deprivation and cycles of poverty and I'd argue it's a lot easier to fall into drug use than accidentally learn maths.

You are completely wrong. There are endless stories told of how people escaped their background disadvantages by education and other forms of help along the way. To some extent that is the whole purpose of state education.

I don't advocate giving up on them entirely but what hope does a teacher have.

It's the teachers job. If they don't like the kitchen etc.....

Personally I think they should bring back metal work/wood work for the kids that clearly arent academic.

and for kids who are academic if they choose.

Bringing it back on topic, AI and computerising everything so there is even less jobs for people to do does concern me slightly. I guess not a lot different to people trashing the threshing machines of old as they took their manual job.

The whole point of technology is to increase productivity i.e. to reduce the amount of work needed.

 

[Spectric]

All of the problems will be resolved once we do get AI because then we can replace the useless woke police force with robocops and they can police and clean up our streets.

The whole point of technology is to increase productivity i.e. to reduce the amount of work needed.

But a point will be reached where there is very little if any work for people, we have already seen the impact of industrialisation and then throw in AI workers and we have made ourselves redundant, but overall production will be way down because AI will not need a huge amount of what us humans need and so a real positive for the planet.

 

[Jacob]

All of the problems will be resolved once we do get AI because then we can replace the useless woke police force with robocops and they can police and clean up our streets.


But a point will be reached where there is very little if any work for people, we have already seen the impact of industrialisation and then throw in AI workers and we have made ourselves redundant, but overall production will be way down because AI will not need a huge amount of what us humans need and so a real positive for the planet.

This is way brainless politicians keep going on about growth. They can't see a way to accommodate the redundancy we are deliberately creating, without simply making and consuming more stuff - including making ever more labour saving devices!

 

[Peri]

It's the teachers job. If they don't like the kitchen etc.....

I think a lot of people are of the same sentiment. Nearly a third of the teachers who qualified in the last decade have left the profession, and last year teacher recruitment was 41% below target for secondary schools (here & here)

Recruitment for some secondary subjects is consistently much lower than the average. In particular physics (83% below target in 2022/23), design and technology (75% below target), and computing (70% below target).

 

[Jacob]

I think a lot of people are of the same sentiment. Nearly a third of the teachers who qualified in the last decade have left the profession, and last year teacher recruitment was 41% below target for secondary schools (here & here)

Hence strikes and the need for better wages/conditions and investment overall.
Same with NHS and other public services.
They say "we can't afford it" but the reality is we can't not afford it or it's a downwards spiral and years of recovery, if ever.

 

[Peri]

Back on track -

New research conducted by a professor at University of Pennsylvania’s Wharton School found that the artificial intelligence-driven chatbot GPT-3 was able to pass the final exam for the school's Master of Business Administration (MBA) program.

The bot's score, Terwiesch wrote, shows its "remarkable ability to automate some of the skills of highly compensated knowledge workers in general and specifically the knowledge workers in the jobs held by MBA graduates including analysts, managers, and consultants."

The bot did an "amazing job at basic operations management and process analysis questions including those that are based on case studies," Terwiesch wrote in the paper, which was published on Jan. 17. He also said the bot's explanations were "excellent."

Link

 

[Spectric]

They can't see a way to accommodate the redundancy we are deliberately creating, without simply making and consuming more stuff

Almost like perputual motion, just that this is a downward spiral that cannot support itself forever. The biggest problem is consumerism is eating away at the very planet we all live on so perhaps the future is AI but then if humans have no real purpose what would the purpose of AI be and at what point would they work this out.

 

[Molynoox]

Taking your example step by step, there are a few operations which I believe are common between human and machine text interpretation. I work on Siri, so I'll use examples from the Assistant domain.

1. Given the shapes, what are the words? Humans and machines do this in functionally-equivalent ways - essentially it's as you say, we compare the squiggles on the page to lots of learned examples and make a decision. Both humans and machines are biased towards 'common' sequences of words eg 'I need some groceries so I'm off to the shops' is much more likely than 'I need some groceries so I'm off to the ships' - both humans and AI use their previous experience of language to bias their predictions about the squiggles.
2. Given the words, what is the meaning? This is where humans have almost unlimited advantages, and where systems like chatGPT are moving the needle for machines. Does 'hold my beer' really mean that there is a beverage nearby which requires immobilisation? Sometimes yes, but usually, no. So, how do you learn what 'hold my beer' means? You read and understand the context. If it's a paragraph where somebody is literally drinking a beer and needs to relinquish it temporarily, it might be literal. More likely, you see 'hold my beer' in a meme or other high-frequency graphical motif, or at the end of a social media post. You learn it's actually a metaphor. AI can learn that too, and from the same signals, and the clever bit about GPT and other transformer-based large language model systems, is that the AI can learn it without humans having to label all of the training data - just like humans can.
3. What should be the response? Here the AI is often on more solid ground, simply because its range of responses is smaller. 'Hey X, be a a love and turn on the lights while I find my glasses' has a lot of content in it, and if you are X, you can probably construct a whole mental scene around it suggesting how the person feels about you, where they are, what's the likely event context, where they left their glasses etc etc, and therefore what is the most salient response. But if X=Siri, the most salient response is smart_lights.TurnOn(). As assistants mature, more of that sentence might be important eg 'turning the lights on, also I can see your glasses on the hall table' (creepy!).

You mentioned meaning, but also understanding, and learning. Tricky stuff! ChatGPT can learn concepts from textual data, and resynthesize and explain those concepts in new text in response to questions. Does it 'understand'? Well, first define 'understanding'

This is a really great post

I feel that many people see AI and human intelligence as two totally separate things, just because the AI is not AS clever as a human. Yet. The reality is that it's just at the very start of it's journey.

Also I feel like 'understanding' is not a threshold but more of a scale and there is nothing special about the human level datum, it's just OUR level and therefore we relate to it.... as a result we have a tendency to compare other things to it and poopoo things that don't meet what we consider to be this standard of intelligence.

Having said that artificial general intelligence is likely many years away and possibly something I will never see but there are many that disagree with me and some think it's just around the corner. Ben goertzel from Hanson is very optimistic for example saying timeframes like 20 or 30 years whereas many others are more like 80-100. But I think they all agree we need a breakthrough before it happens

Martin

 

[Flynnwood]

Appears that ChatGPT launched on 30th November 2022. By the morning of 5th December, it had 1m users.

I note the website is now more professional today - timing aligned with the MS investment announcement.
Example:
https://beta.openai.com/examples/

 

[Dibs-h]

I'd say that when it becomes a lifestyle that's when they really have problems and could most end up needing the welfare state.

In some\many cases - it's a spiral downwards and the people and many times their offspring are marginalized.

There is another definition of lifestyle - the one where a person does as little as possible and "milks" the state. They do exist and whilst it may not suit you (currently or ever) to agree, it doesn't make them any less real.

And they are probably more prevalent than we'd like to think - after all the days of seeing someone come out of the DSS office (and knowing they went to claim for something) are long gone.

I don't see the point of disapproving of tw-ats. They are not the problem - but the mega rich really are.

Sadly - both ends of the spectrum are part of the problem. Just to differing degrees depending on location and specifics.

Then you have the hybrid version: very very rich (99.999% through illegal means) and tw-ats.