I do Microsoft Dynamics 365 ERP systems. As a test, I've been asking Copilot to generate code for me. I am yet to see a piece of code that runs correctly - and I'm talking about something that the AI system should know, since it has feasted on Microsoft documentation. Even the smallest piece of code - generating an event handler for a table insert event - is wrong. It might be 90% correct, but that 10% kills it.

Next you'll be saying the anthropic principle is true!

It's an attractive idea, but reality bites.

I do Microsoft ERP systems; specifically the Dynamics range. We have customers still on the last on-prem version that was sold, Microsoft Dynamics AX 2012. These people are starting to have titanic problems keeping a vintage system like this running. Not because to OS requirement, not because of database requirements, but because the world has moved on in the last 13 years. The tax solutions for this system are starting to no longer work. Integrations for systems like FedEx and UP are starting to fail because the APIs at the carrier end have advanced so much maintaining older protocols is that much harder. I've just finished mapping FedEx V20 (the current one) to the older one (V4) in the AX2012 code base, and it cost the company a packet, since FedEx depreciated V4.

Many interface technologies have gone away, and better one have appeared in the intervening time. Reporting requirements are always changing and evolving. Automatic bank transactions (eg. ACH) are evolving more and more security. Payment processors (like Braintree) are upping their security. And, security certificates in these systems are starting to expire, with no real way of updating them.

Today, no ERP system is on-prem; they are all cloud hosted. Keeping your data on-prem is a nice idea, but in reality it's almost impossible if you want to use a modern, fully featured ERP system

Let's enter the wayback machine, and transport ourselves to NSW, Australia, circa 1988. Back then, power generation was a state government responsibility. Electricity generation costs grew, of course, over time, but it was reasonable.

Then, in the late 1980's, the NSW state government had this bright idea: sell off electricity generation to private industry, who would compete (ha!) amongst each other and deliver lower electricity costs as a result of this competition to consumers. They would also benefit from the sale of publicly-owned assets as a one-off boost to the state coffers.

Predictably, the cost of electricity generation went up, the quality of service went down, and the people of NSW were kinda screwed over. We saw the same pattern with every other NSW asset that was sold off.

So, I personally am very much of the opinion that social necessities, such as water and electricity, should be supplied by socially-owned utilities.

Well, we can say goodbye to this version of the AI industry.

And that would actually be a good thing. The GenAI industry is basically a dead-end as far as artificial intelligence is concerned, because it's leading research time and money away from achieving General Artificial Intelligence. How can we know this? Because: in order to write this sentence, I didn't need to ingest the entirety of humanity's written word. I'v read a couple of thousand book so far in my life (if that); I've had a high school education to Year 12; I've done a trade course in electronics, and watched a crap-ton of TV. So my exposure to the entire written corpus of humanity is maybe .000001%. Yet I can still create a paragraph that has never been written before; I can hav original thoughts when crafting code (or recovering data from a smashed database) at my day job; I can hold a conversation with people.

ChatGPT (and its ilk) can only partially achieve that by transforming and parameterizing everything ever written. The AI industry is heading down a dead end. They know it; they are just milking every last investment dollar they can until the bubble bursts. Sure, there are use cases for GenAI: something that feasts entirely on, say, Microsoft documentation might be in a unique position to actually provide useful answers. The cancer screening AIs are doing a remarkable job of sorting through images and picking out images for further study. But GPT-based AI? It's a dead end.

When we understand how the brain actually works and can mimic that, then we will have true general AI. And every day, the AI industry is getting further from that point.

Unfortunately, this isn't the case in Australia. The Australian Research Council recommends to the Federal Minister for Education projects that should be funded. The Minister ultimately decides who and what to fund.

Sadly, it's not that simple.

Consider this case from Australia: The cancer drug, the faked data and the superstar scientist. A top research, backed by many prestigious institutions, faked his data and his mouse tests. Did people know this? Yes. Did they risk (and in many cases, end) their careers by blowing the whistle? Yes. Did the institutions do anything? No; they protected him because he brought in so many research dollars.

Yes, paper mills are a curse. But forcing an accreditation agency to vet all papers isn't the way. That's called peer review, and Mark Smyth passed all peer review.

Yet he still pumped out fake papers; hundreds of them, polluting scientific knowledge with fake data.

I am no fan of big pharma, but I realize they are the ones who take the commercial risk to create new drugs. Drugs such as Nelistotug, an anti-cancer drug that turns the bodies immune system against cancer. Or is supposed to; Nelistotug is build directly upon research done by Smyth, and all the underlying data is faked; the papers have been pulled. GSK is left holding the bag on this one, because they have invested so much into bringing the drug to market and starting human trials. The trials are showing that there are no ill-effects from the drug, because - surprise! - it does nothing. Millions and millions of dollars wasted. Over 42 million dollars has been poured into this fraudsters research, money that other researchers won't see.

Sure, we could prevent publication by the institutions Smyth has worked at... but that would end the careers of so many legitimate researchers, too.

It's a massive problem, and simple solutions unfortunately are not up to the challenge.

No; a weak password did not kill this company.

Management not investing in the most basic of backup systems is what killed the company.

Companies get their systems wiped out everyday nowadays by ransomeware hackers. Then, they pull the plug on the internet, scrub the computers, and restore from a recent backup. That is management.

This is stupidity.

Back in the mid 80's there was a show called "Knight Rider". You may know of it, from the iconic TransAm it used as the hero car. Guessing by the size of your user id, you should know it well.

Anyway, the music. The show didn't have the budget to license contemporary music, but they did have the budget to hire a sound-alike band to mimic the songs and skirt copyright and licensing restrictions (especially overseas restrictions). Sometimes, it was obvious it was a cover band. Other times, you had to listen really carefully to hear the differences. All this is documented in the book "The Knight Rider Companion", if you care to see my source.

To the point: The Knight Rider audience didn't much care who played and sung the songs; we wanted to see spectacular stunts and Michael Knight win the day. Was the music misrepresented? Maybe; can't remember that much. Was it fraud? By your definition, yes it was.

Curious about your thoughts on this.

Unlikely.

Far more likely, it was a Friday, and some poor sod of a programmer had been ordered to push code to PROD. The code, of course, was barely tested, and had a number of nasty bugs that were only apparent in the PROD environment because management was too cheap to properly build out a UAT environment for testing. Anyway, they push to PROD, reboot the system and BOOM! What looks like two black holes merging was actually a division by zero in a critical system.

That's my theory, anyway.

Sometimes, stuff is just hard to get right. Look at AirPower: announced in 2017; cancelled in 2019 because they just couldn't;t make it work right.

I'd personally rather Apple take their time and get AI done right, than jam it into every single product, irrespective of if it makes sense or not. A Copilot-powered Notepad, for example, is just silly; yet here we are. The case management built into Dynamics CRM has Copilot summaries now; they are less than useful and make page loading take so much longer.

I bought an iPhone 16 on the basis of the AI capabilities; it was our designated upgrade year anyway so the fact these features haven't materialized is no great loo; in fact I'm happy not to have them when I look at the ungodly mess Microsoft land has become. And Google's AI summaries... there is wrong, and there is not even close to being wrong.

Ah, another satisfied Microsoft 365 user, describing their wondrous interactions with Copilot!

Nope; just pre-empting tomorrow's internet carnage. Today's mess is something else.

The book "Infinite Loop" had a take on this, too. The whole Power line had issues, and increasing clock speeds would be problematic for it. Andy Grove paid Apple a visit, and showed them their roadmap. Apple took them seriously (for a change), seeing that Intel's CPUs would leave the Power chips in the dust, both in pure GHz performance and power consumption. It was a no-brainer for then to switch to Intel, whilst leaning how to design their own chips, with the A4 being released in 2010.

Remember, the whole PowerPC thing was an Apple - IBM - Motorola alliance to counter Intel. Apple was interested in their own chips even then. The alliance fell apart because Motorola couldn't hold it's end of the bargain up; specifically making faster CPUs.

Hopefully, the first title available will be Half Life 3.