Thanks for the link @mikebrisy - very worthwhile listening to get a sense of where the producers of AI are at right now and the challenges they see coming.

Given the government/regulation fail over social media, I can't say I've much hope they'll somehow get their act together in time to meet the coming AI challenges.

I thought the discussion of teaching AI values to align with human society and the protection of same a little worrying, but at least someone's thinking of that!

Yeah, that's an interesting thought @tomsmithidg

It depends on the degree to which AI is commoditised -- that is, do only a handful of big companies have the capacity to deliver models with the necessary capabilities, or do we see a raft of "fast followers" who can quickly roll out comparable models? So far, the latter seems to be the case (eg DeepSeek). So while Google or OpenAi or Anthropic tend to lead, each of those seem to fail to hang onto any significant lead for long. Musk's Grok model was a late starter but from what i understand is near the top of all the usual benchmarks that try and quantify AI capabilities.

The guy behind Deepmind, alpha-Go and Alpha fold and who leads Google's efforts, Demis Hassabis, was interviewed on Lex Fridman recently and spoke a bit about this (well worth a listen btw). When he was brought on as the lead of Google's AI team a year ago, Google was a good deal behind OpenAi and Anthropic. Now, it seems, they are leading. He seemed to say that the improvements they are making now are more about scaling up and refining existing approaches, rather than needing new computer science breakthroughs. He even thinks we may be able to get to AGI with existing knowledge.

But, im far from an expert on these things, so wouldnt make any confident assertions at this stage. Definitely something to watch

@tomsmithidg by far not an expert in this complex field but you have (maybe) hit upon my thesis for why the hyperscalers will win and why i own them. having the best model is only part of it, i think you need a model that is comparable, there is leap frogging occurring now and i dont think that matters now all that much, maybe Meta neesd to lift here (thye know it and why Zuck is poaching). what matter is selling the AI sevices more and more non-it companies. the fear is that companies will fall behind competitors if they dont use AI and imbed it in their systems. the sellers are GOOG MSFT AMZN plus a few others. however, the early AI adopters and those that execute well (customers of the hyperscalers) will get an advantage over their competitors but ultimately all will embed AI and become reliant on it, so beholden to above mentioned companies.

there is a long way to play out and many moving parts but thats my game plan in this space, which guys embed their AI agents into the most companies so then they have them captive, MSFT winning so far but all big tech will have a go, imo. they are seeing results and spending big.

the MSFT CEO had a good quote he said something like we buy a GPU and add software layers that increase its value by 2-3X, thats the plan, how customised/commoditised those software layers are case by case becomes important i suspect

SM i tried listening to the Lex podcast, those two guys are way too smart for me, ill give it another go. lol

I reckon MSFT has potentially got a handy advantage given that so much of the world use their software, just look at the proliferation of Office, Teams and now pushing Copilot. I agree with @Solvetheriddle that hyperscalers will dominate through market dominance and the ability to buy out competitors. AI obviously has a long way to go in a lot of fields, I had an interesting discussion with my physio who mentioned that it failed the most basic test for diagnosis in the radiology field and similarly struggled with detecting skin cancers.

Is a great topic to discuss.

I might have a slightly different view to some of the posts here. From all the reading and podcasts I’ve jumped on, I’ve formed the view that, to really understand the impact we’re facing, we need to talk about AI and robotics as one topic. Yes, you can see them as two sides of the same coin for easier discussion but they are merging to create the ultra-smart PhD+ level robots. When you look at the conversation through that lens, you start to see a pattern where the future winners may well be more on the hardware side rather than software, given the best programming engineers today are going to increasingly be available (through AI) to everyone in the world....and for free! There will still be software companies that do well due to their network effects, but new modules and features even for those systems will come from AI models as will, pretty much literally, any software that we use in the future. The cost of these will and is rapidly trending to zero.

As just a few examples, the software and data analytics tools that Catapult use will be easy to duplicate, but not so much their proprietary data, so they may still do okay. Microsoft has such a strong network effect, as do Facebook and X, so they’re likely to be fine even as AI writes more and more of their code and new platforms that are just as good, but not as widely used, become commonplace. Self-driving cars are already a thing and the implications and opportunities for that are extreme. Look at what Meta is doing, paying extreme salaries to a very small group of people rather than having an army of developers.

My approach in all of this? To use the gold rush days as an analogy, we know there’s gold in “them their hills”, but I have no idea where the best deposits are. So I’m looking to invest in those that make the picks and shovels (and even those who make the materials for the picks and shovels), as those areas are very likely to do extremely well. This means the hyperscale's, those that build chips (not just NVIDIA), power generators for the ongoing demand for data centres and the leading companies in the robotics space, to name a few.

Companies that are a pure software play without deep networks could be in for a rough time. Maybe a good future business could be making T-shirts that say, “I am not Sarah Connor.” :)

Given the topic of this thread, I haven’t touched on the changes in the financial system either, but they’ll also be profound.

Interesting times ahead.

Interesting comment @tomsmithidg I had much the same conversation with my skin care doctor recently. She quoted your findings on AI results to date, I said, just add ‘atm’ to that statement. Surely it’s a case of just feeding more data on shapes, sizes and colours of cancerous barnacles and other skin mutations into the AI bank. Plus, it might perform more searches on the private parts of the body. True story: a neighbour lost her husband to melanoma a number of years ago. It was in the cheeks of his butt where the sun don’t shine! Not checked! Nor is mine and I am aware of the risk. Not sure how AI alone does the parting of the cheeks, but that’s a story for another day.

Two issues hang off this: Will AI err on the side of caution and flag more biopsies than necessary, then again, it will eliminate human error in spotting questionable problems. Secondly, it may or may not aid in the ‘medico second phase revenue play’.

I am aware of a case where a friend queried the number of additional excisions/freezes and biopsy samples that followed his regular check up - so he ‘second opinioned’ and to his horror, he was being ‘over doctored’, no doubt with thoughts of the Merc lease, the private school fees and the upcoming holiday in Colorado. The Hippocratic oath might say NO this, then again, they are running a business.

@Solvetheriddle the future is happending,.... now:

This is big. New CEO describing phenomenal scaling benefits for logistics companies using this. "Agents" are trained in specific "roles", e.g., testing an exception event for contractual non-compliance - something a person does today. Huge.

100% @mikebrisy

if we take that chart and include other outcomes, like AI empowered robotics handling the manufacturing of 'things' and then delivering the final products, we start to see that this brave new world will be unlike anything we can truly imagine. The investing side of that is a fascinating conversation in itself, but the social impacts will be massive.

As you touched on, this brave new world is already scaling on the software side, and the hardware side is coming fast. Self driving cars are here and now, but that’s just the beginning. This is the side of the coin that often gets missed in most conversations about AI. Yes, the software side may trend toward zero marginal cost, but the hardware opportunities (and human displacement) are almost endless, and it’s in this space where the profit potential is beyond massive.

Two quick examples: Amazon went from being a book seller to a cloud computing giant, and is more than likely to become a robotics company in the near future, as they commercialise their investments that have allowed their distribution centres to run with fewer and fewer humans. Also, why did Facebook, Apple, and others release their smart glasses? To provide a good experience, maybe, but also think about the data they’ve been capturing on how humans do things, think Tesla cars recording every drive, to help train future AI driven robotics.

We’re not in Kansas anymore folks.

@BigStrawbs70 true,

I saw a WSJ video recently (I can't share as its behind the paywall, but relevant screengrab is below) which shows that China is implementing more robots (physical ones) than the rest of the world combined.

Your point is a good one. Robots in manufacturing have been around for a few decades. What's accelerated in the last few years are robots with greater software control to widen the scope of what they do and, of course, the current phase of that change is "AI" robots. The impact on the labor force will be profound and gradual, because of the capital replacement cycle.

New factories (including so called "dark factories") are being built that run without people for some product lines. Some are already online in China. Of course there are people there, but they are increasingly in supervision, analysis and maintenance roles. Of course, this is a revolution the chemical process industries went through 50-70 years ago. But now it is hitting manufacturing and supply chain.

The academics and "analysts" are starting to get their heads around it, but the political discourse hasn't really started yet. There was a bit of an attempt by ACTU to put it on the table in Australia in the recent "roundtable" but I'm not sure how well-developed the dialogue was. It looked a bit like employers vs. unions positioning, but the issue really needs people to work together on this.