AI endeavours

[odysseus2000]

CNBC do their take on Apple missing out on AI (9:17):

https://youtu.be/A2QhMsaZfgc?si=mMP-o4_1yjoxfLon

It is a clueless rambling piece, suggesting no one has a clue what is going on with Apple AI.

I still have this thesis that only NVDIA are making a killing out of AI. All the other runners tell me how wonderful & powerful AI is, but don’t seem to be covering the overheads of their AI chip buys from Nvidia.if as suggested Apple license Google AI, does this come a long side their existing licensing of Google search, or does it replace it? The current AI bubble looks like the internet bubble when prices rose, but earnings didn’t & eventually prices came off.

As of now there have been some wildly successful applications of AI, such as it becoming world GO champion & there are lots of folk saying that AI Wii replace Adobe, Search, writers, graphic artists, etc etc. These & many more possibilities seem very likely, but as of yet the general AI models are far from good enough & no specialist AI have gained traction towards becoming dominant in some specific discipline. As of now it looks to be a few spectacular successes that don’t scale into other areas & a shed load of hype.

Am I getting this wrong?

Regards,

[odysseus2000]

Nice overview video covering the latest AI chips including the new Nvidia with fp4, but still I don’t really get fp4 (21:49):

https://youtu.be/zXNUBFoNPX0?si=0w1TXVB0A0WVojtc

Regards,

[Tedx]

8 million jobs at risk

I've read elsewhere that the UK is the least automated major economy in the world and the least productive.

We're ripe for the picking it would seem

https://news.sky.com/story/ai-risks-up- ... s-13102214

[odysseus2000]

8 million jobs at risk

I've read elsewhere that the UK is the least automated major economy in the world and the least productive.

We're ripe for the picking it would seem

https://news.sky.com/story/ai-risks-up- ... s-13102214

This is badly researched.

AI is currently far more capable of doing skilled high paid jobs than low paid unskilled jobs.

AI companies are creating expert systems to replace humans in jobs were knowledge & high pay go together. For example a GP costing around £200k for 40 hours or less is a prime target where the fee for AI can be some fraction of the GP’s salary.

Replacing low skilled workers needs biped robots & the cost saving & fees are lower, so not yet a priority.

Regards,

[GrahamPlatt]

Not been keeping up with this thread & have skipped the past three pages of posts, so apologies if this has already been mentioned

“Startup Databrix has just released DBRX, the most powerful open source large language model yet - eclipsing Meta’s Llama2”

https://www.wired.com/story/dbrx-inside ... -ai-model/

[88V8]

Ebay are now offering sellers the option to have AI write the description.

So if you see a pretentious pile of woffle that tells you nothing about the item, that's the AI option.
I don't think AI is about to do away with us just yet

V8

[odysseus2000]

Ebay are now offering sellers the option to have AI write the description.

So if you see a pretentious pile of woffle that tells you nothing about the item, that's the AI option.
I don't think AI is about to do away with us just yet

V8

Interesting! If the article is correct with a cost of $10 million, this is extremely low & could not have involved many (any?) Nvidia chips.

There have been stories circulating for a while that AI will train AI, I.e. existing models created at huge cost will be used to train new models with out needing all the Nvidia chips saving a shed load of money.

Does anyone know about this?

Regards,

[Clitheroekid]

Ebay are now offering sellers the option to have AI write the description.

So if you see a pretentious pile of woffle that tells you nothing about the item, that's the AI option.
I don't think AI is about to do away with us just yet

V8

I think many people seriously underestimate the capability of AI. I've been astonished at the high quality of some of the answers it's provided to legal queries.

I just did a dummy sale of a Rolex watch on Ebay, and having provided virtually no information about it and using a stock photo, I asked AI to provide a description. It came up with this:

This elegant Rolex Datejust wristwatch for men features a silver bracelet made of stainless steel and a smooth bezel in the same colour. The watch has a round shape and an analogue display with baton indexes on a silver dial. It is a self-winding mechanical watch that is water-resistant up to 100 meters (10 ATM) and has a date indicator. The watch is made by the well-known Swiss brand Rolex and belongs to the Datejust model. The reference number is 126300, and the case and band are made of stainless steel. The watch is 41mm in size and has a solid caseback. The watch has a Swiss-made movement and comes with a Rolex Oyster bracelet.

OK, hardly a marketing man's dream, but perfectly accurate and acceptable, and I would imagine far better than many sellers on Ebay could have drafted themselves.

[odysseus2000]

From about 1min 25 to 2 minutes shows a Tesla on FSD (AI) negotiate stopped cross traffic & go through a space left by another driver:

https://youtu.be/-oAVFEDkJuk?si=LNjZdyZQ4tHBdL7n

Assuming this is not a fake, the AI is driving with intelligence like a human would, suggesting comparable to human cognition.

Regards,

[odysseus2000]

Ebay are now offering sellers the option to have AI write the description.

So if you see a pretentious pile of woffle that tells you nothing about the item, that's the AI option.
I don't think AI is about to do away with us just yet

V8

I think many people seriously underestimate the capability of AI. I've been astonished at the high quality of some of the answers it's provided to legal queries.

I just did a dummy sale of a Rolex watch on Ebay, and having provided virtually no information about it and using a stock photo, I asked AI to provide a description. It came up with this:

This elegant Rolex Datejust wristwatch for men features a silver bracelet made of stainless steel and a smooth bezel in the same colour. The watch has a round shape and an analogue display with baton indexes on a silver dial. It is a self-winding mechanical watch that is water-resistant up to 100 meters (10 ATM) and has a date indicator. The watch is made by the well-known Swiss brand Rolex and belongs to the Datejust model. The reference number is 126300, and the case and band are made of stainless steel. The watch is 41mm in size and has a solid caseback. The watch has a Swiss-made movement and comes with a Rolex Oyster bracelet.

OK, hardly a marketing man's dream, but perfectly accurate and acceptable, and I would imagine far better than many sellers on Ebay could have drafted themselves.

As far as I can tell AI is the first Industrial Revolution that will first target the most sophisticated & well paid jobs in society. People who are skilled in crafts like plumbing will not be replaceable in this first wave, but highly sophisticated well paid people will be replaced. The economics are overwhelming stacked against professional people. Not only can for example say a £200k GP be replaced for far less cost, but the AI can work 168 hours per week, equivalent to 4 doctors, so that a medical Ai would replace around £800k of clinician costs. Moreover, AI is demonstrating far better accuracy than humans in diagnosis. My own experience of GP’s at my own surgery is poor: No follow ups, reading notes during phone calls, rarely listening… on & on it goes. One potential problem is the quality of the training that an AI gets. If for example you try & study some medical aspect via web searches there is always contradictory information. For example I was looking at the relative iron content in various foods & I found that pork in the UK there has little, but a US web site said pork had plenty of iron. Perhaps US pork has iron added, but there are numerous examples of contradictory information. For these kinds of reason it looks necessary to develop specialist AI, rather than the general purpose models currently commonly used for each discipline, very like how humans get trained to specific professions.

Regards,

[Clitheroekid]

The economics are overwhelming stacked against professional people. Not only can for example say a £200k GP be replaced for far less cost, but the AI can work 168 hours per week, equivalent to 4 doctors, so that a medical Ai would replace around £800k of clinician costs. Moreover, AI is demonstrating far better accuracy than humans in diagnosis.

I tend to agree - medical diagnostics seems the perfect target for AI, as it consists of analysing data and arriving at a conclusion.

For all that GP's are (generally!) well-intentioned, they are human beings, with all the failings of human beings. I have met quite a few over my career, and they vary immensely, as would be expected. However, there are many who are fed up and disillusioned with general practice, and can't wait to retire. They have long since lost any enthusiasm for the job, and it always amazes me that there aren't many more with this attitude, bearing in mind the number of idiots they have to contend with, who simply want the GP to give them a sick note so they can skive off work, or who have imaginary illnesses.

It's inherently likely that such GP's are, having qualified decades ago, unlikely to keep up to date any more than their CPD regulations require. They are also overworked and likely to be bored with the same old problems day after day, so that on the odd occasion a patient is genuinely ill they may easily overlook it. This is the sort of problem that keeps hitting the headlines, where a child with meningitis or similar has been prescribed Calpol and sent home to die.

Another major issue is that GP's are almost invariably from middle-class backgrounds, and have unconsciously absorbed middle-class culture and values. This includes a judgemental attitude, making them less inclined to provide a high quality service to people who they see (probably correctly!) as feckless and largely responsible for their own problems.

AI doesn't suffer from these human failings. It's more up to date than even the most diligent GP could hope to be, and it makes no assessment of the patient's character or social standing unless it's relevant to the diagnosis.

GP's or their patients may counter by saying that AI doesn't have the empathetic insight that humans have between each other, but whilst that's obviously true, at least at present, how much of that really exists in the average rushed GP appointment? This is particularly the case since patients nowadays rarely see the same GP regularly, so that the personal relationship that patients once had with their GP - the rather paternalistic but essentially benevolent Dr Finlay type of relationship - has largely gone, and is, no doubt, another reason why so many GP's want out.

I agree with odysseus2000 that the jobs most at risk from AI are the highly skilled professional ones, if only because the financial savings are far greater than those from replacing lowly paid workers. In particular I would think that accountants, actuaries and architects are at risk. And in view of the recent furore regarding exorbitant vets' fees I'm sure that many people in the future will prefer to trust their pooches to an AI diagnosis at a fraction of the price that a vet would charge.

As is said in my last post I really believe that the vast majority of people are incredibly ignorant and/or complacent about AI, and have no conception of just how much it's likely to change things. I think it's likely to cause just as much, if not more, disruption and change as the mobile phone or even the internet itself.

[gnawsome]

... but the AI can work 168 hours per week, equivalent to 4 doctors, so that a medical Ai would replace around £800k of clinician costs...

I have printed out a copy of "Which days does my GP work"
It give the details of the 'shifts' each Dr works.
It names 9 doctors and indicates each day or half-day worked.
The total of 20 days by 8hrs per day (maybe an underestimate) would indicate 160 hrs.

Only one works 4 days. Another works one and a half days
I wonder how many of them do shifts as locums at other places at higher rates of pay.

[Hallucigenia]

As of now there have been some wildly successful applications of AI, such as it becoming world GO champion & there are lots of folk saying that AI Wii replace Adobe, Search, writers, graphic artists, etc etc. These & many more possibilities seem very likely, but as of yet the general AI models are far from good enough & no specialist AI have gained traction towards becoming dominant in some specific discipline. As of now it looks to be a few spectacular successes that don’t scale into other areas & a shed load of hype.

Am I getting this wrong?

I can throw two specific personal examples of someone losing real-world £££ to "AI". One was a friend of a friend who does voiceover work, had a job booked but then got kicked off it because they'd decided to use an AI replacement voice, leaving them wondering what their long-term future looks like. Another was getting some new branding for a company recently. A couple of years ago I had some branding for a new consumer-facing company plus some other bits from a local graphic designer for a few £100 and a couple of weeks of to-and-fro. The branding for the recent company was less critical - it was one of those where we just needed something to look the part rather than a critical part of the company's future - but we got it done in an evening with AI. And it's fine - not the greatest ever, but a solid 7 or 8 out of 10, which is all it needed to be. And that wasn't even with one of the good image AIs like Midjourney.

And some of those "visual" AIs - in still and now video photography, are good enough to make real inroads now. To get an idea of what's going on, these are some of the companies that one of the leading Silicon Valley incubators is funding :
https://twitter.com/snowmaker/status/17 ... 4332530953

And it will probably be some of the more obscure ones like optimising chemistry conditions that will make the most difference, in the same way that arguably currently the most successful AI is one that most people have barely heard of, namely what Deep Mind have achieved in protein folding.

I get the impression that we are heading for a repeat of what happened with Intel x86 - that the general models enter a virtuous circle where they pull in all the money so end up being so much better than the specialist models that they mostly kill off the smaller specialists - at best you'll see OpenAI-for-oncology rather than Onco-AI from a smaller company, but we'll see.

[odysseus2000]

As of now there have been some wildly successful applications of AI, such as it becoming world GO champion & there are lots of folk saying that AI Wii replace Adobe, Search, writers, graphic artists, etc etc. These & many more possibilities seem very likely, but as of yet the general AI models are far from good enough & no specialist AI have gained traction towards becoming dominant in some specific discipline. As of now it looks to be a few spectacular successes that don’t scale into other areas & a shed load of hype.

Am I getting this wrong?

I can throw two specific personal examples of someone losing real-world £££ to "AI". One was a friend of a friend who does voiceover work, had a job booked but then got kicked off it because they'd decided to use an AI replacement voice, leaving them wondering what their long-term future looks like. Another was getting some new branding for a company recently. A couple of years ago I had some branding for a new consumer-facing company plus some other bits from a local graphic designer for a few £100 and a couple of weeks of to-and-fro. The branding for the recent company was less critical - it was one of those where we just needed something to look the part rather than a critical part of the company's future - but we got it done in an evening with AI. And it's fine - not the greatest ever, but a solid 7 or 8 out of 10, which is all it needed to be. And that wasn't even with one of the good image AIs like Midjourney.

And some of those "visual" AIs - in still and now video photography, are good enough to make real inroads now. To get an idea of what's going on, these are some of the companies that one of the leading Silicon Valley incubators is funding :
https://twitter.com/snowmaker/status/17 ... 4332530953

And it will probably be some of the more obscure ones like optimising chemistry conditions that will make the most difference, in the same way that arguably currently the most successful AI is one that most people have barely heard of, namely what Deep Mind have achieved in protein folding.

I get the impression that we are heading for a repeat of what happened with Intel x86 - that the general models enter a virtuous circle where they pull in all the money so end up being so much better than the specialist models that they mostly kill off the smaller specialists - at best you'll see OpenAI-for-oncology rather than Onco-AI from a smaller company, but we'll see.

Some very good points, but the Intel x86 example now seems dated. Much of the success of x86 was due to Microsoft and Bill Gates making everything closed source so that to use computers effectively most users had to enter the duopoly of Wintel. This made Bill Gates a fortune but the closed source model became so greedy and top heavy that users left wintel and either moved to Apple or unleashed all the power of open source unix.

It may be possible for someone to achieve a wintel like dominance but there is a, not unrealistic, fear of such monopolies based on the Gates coercion of users to use one operating system and all the negative impacts in terms of price and lack of development that culminated in the App revolution where countless small developers were granted a slice of the pie and users got low cost tailored software for almost any interest.

Achieving wintel like dominance in terms of AI looks currently unlikely as we have several powerful and well funded alternatives: Google, Microsoft/open AI, Grok, Facebook and a host of smaller players who are concerned with building expert systems rather than general purpose AI. There are issues with the big players but as of yet they all have cash to develop models suggesting that at least 4 big AI players will be important with none able to choose the closed source model of Microsoft as users can move between them if one group get too aggressive with pricing.

However, I still wonder if any of these AI titans can make enough profit to cover what they are paying Nivida, but that may not be necessary as each can absorb the costs if needed and remain profitable from other income sources. Perhaps Grok is the weakest as its not clear if X can grow enough to cover the expense although Elon may continue to fund it anyway. There is also a question over Microsoft/Open Ai with the Musk law suit against Open AI. The defence that Altman has put up looks extremely weak to me, but until it goes to trial it is hard to know. Google and Facebook are extremely well funded.

Regards,

[Hallucigenia]

Much of the success of x86 was due to Microsoft and Bill Gates making everything closed source so that to use computers effectively most users had to enter the duopoly of Wintel. This made Bill Gates a fortune but the closed source model became so greedy and top heavy that users left wintel and either moved to Apple

As so often, you are so wrapped up in your own world that you ignore what other people are actually saying. I didn't mention Windows or Bill Gates at all, because it's not relevant to my point. I'm making a point about generalists versus specialists - in the 1980s and 1990s there were a whole bunch of specialist chips like transputers and i860 for supercomputers, specialist chips for game consoles, all sorts. But one by one they fell by the wayside as x86 was good enough and equally importantly was making so much revenue that Intel could afford to spend a fortune on R&D to make it even better, so that even Apple ended up with Intel inside.

Nor was I trying to say anything about monopolies - whilst I think generalists will mostly prevail over specialists, I think it's quite likely that we will have several generalists, in the same way we've ended up with several providers of mobile phone CPUs. It really would help if you actually read what people wrote, it gets quite tiresome when you don't.

[odysseus2000]

Much of the success of x86 was due to Microsoft and Bill Gates making everything closed source so that to use computers effectively most users had to enter the duopoly of Wintel. This made Bill Gates a fortune but the closed source model became so greedy and top heavy that users left wintel and either moved to Apple

As so often, you are so wrapped up in your own world that you ignore what other people are actually saying. I didn't mention Windows or Bill Gates at all, because it's not relevant to my point. I'm making a point about generalists versus specialists - in the 1980s and 1990s there were a whole bunch of specialist chips like transputers and i860 for supercomputers, specialist chips for game consoles, all sorts. But one by one they fell by the wayside as x86 was good enough and equally importantly was making so much revenue that Intel could afford to spend a fortune on R&D to make it even better, so that even Apple ended up with Intel inside.

Nor was I trying to say anything about monopolies - whilst I think generalists will mostly prevail over specialists, I think it's quite likely that we will have several generalists, in the same way we've ended up with several providers of mobile phone CPUs. It really would help if you actually read what people wrote, it gets quite tiresome when you don't.

Yes, I know you did not mention windows or Bill Gates. I am saying that without the wintel duopoly Intel would not have become the power house it was under Andy Grove. What I was arguing was that the situation now is unlike what happened previously because of the points I raised. To argue that x86 became popular with out mentioning the duopoly is to ignore an important point of history as related to giving a steer to the current situation. The point of discussion boards is to present varying views allowing the reader to decide for themselves what is correct.

Regards,

[GrahamPlatt]

The point of discussion boards is to present varying views allowing the reader to decide for themselves what is correct.

I’d have stopped after “views” (or perhaps “themselves”).

[odysseus2000]

The point of discussion boards is to present varying views allowing the reader to decide for themselves what is correct.

I’d have stopped after “views” (or perhaps “themselves”).

Yes, but at some level folk will often jump on one side or the other.

I have no problem with people correcting me or deciding my ideas are wrong. Some times it is more useful to see opposing rather than supporting views.

Regards,

[TUK020]

I get the impression that we are heading for a repeat of what happened with Intel x86 - that the general models enter a virtuous circle where they pull in all the money so end up being so much better than the specialist models that they mostly kill off the smaller specialists - at best you'll see OpenAI-for-oncology rather than Onco-AI from a smaller company, but we'll see.

Interesting. I had a slightly different take on this - I was working at Intel in the early 80's when there was the battle for dominance between Motorola 68000 and Intel 8086/88.
I think Motorola had the better architecture at the time, but Intel grasped sooner that the key investment issue for their customers was in software development, and they put much more effort into design tools Development Systems, In Circuit Emulators etc.
And once a company had made significant investment in development tools, and then more critically in engineers training on compilers etc, then the design wins went down the path of least resistance

[Sorcery]

I get the impression that we are heading for a repeat of what happened with Intel x86 - that the general models enter a virtuous circle where they pull in all the money so end up being so much better than the specialist models that they mostly kill off the smaller specialists - at best you'll see OpenAI-for-oncology rather than Onco-AI from a smaller company, but we'll see.

Interesting. I had a slightly different take on this - I was working at Intel in the early 80's when there was the battle for dominance between Motorola 68000 and Intel 8086/88.
I think Motorola had the better architecture at the time, but Intel grasped sooner that the key investment issue for their customers was in software development, and they put much more effort into design tools Development Systems, In Circuit Emulators etc.
And once a company had made significant investment in development tools, and then more critically in engineers training on compilers etc, then the design wins went down the path of least resistance

I remember the Motorola 68000 too but from a user programmer's perspective. In the days of using DEC PDP/11s and VAXs the 68000 looked more like a VAX apart from it only having 16 registers and split into 2*8, one set for addressing and one set for (arithmetic & data manipulation). In retrospect perhaps an 80386 with 32 bit addressing and 32 bit registers, would be a better comparison. The 8086/88 should never have become the chip of choice for IBM.

Turning to Microsoft, I seem to recall that an IBM purchasing manager visited the company that made the CP/M operating system and expecting to find an office full of head-down nerds, found that the CEO was playing golf. So the IBM person left and went to visit Microsoft also unannounced and gave them the operating system development which turned out to be MSDOS. Look at them now.