Generative AI, from LLMs, is just bullshit: no accuracy or truthfulness guaranteed. Great for generating bullshit where the content doesn't matter, but mostly useless for anything else.

"AI" already has a bad reputation amongst the general public, who have quite sophisticated bullshit-sense. When will government realise they're being conned by Big Business?

""AI" already has a bad reputation amongst the general public, who have quite sophisticated bullshit-sense."

Nope - there is no BS sense in most of the population, look at the rise of Nigel Garbage and co.

"When will government realise they're being conned by Big Business?"

Government don't care about wasted investment, because the waste isn't counted in growth figures. With the current desperation to find growth, government are looking to bump up GDP. Business investment and government spending contribute directly. Whilst the quickest way of bumping up GDP and finding some growth is government spending (eg build a national high speed road network, and a national high speed rail network) that's constrained by government borrowing costs. If credulous tech companies can be persuaded to invest a few billion in DCs, then it doesn't matter if they deliver any benefits, or even if they stay open, or the investing company goes bust.

Unfortunately, one of the reasons we have a growth problem is that previous governments have also disregarded the importance of value creating investments. So defence mis-spends and overspends increase GDP, but produce nothing and result in higher borrowing costs for government; Misguided policies like policy support for solar power result in billions being spent on that, and because that's commercially funded it has the even more toxic effect of crowding out proper, risk based commercial investment.

"So defence mis-spends and overspends increase GDP, but produce nothing"

Not quite true. They can occasionally produce [1]long queues at the hearing clinic , for example.

[1] https://www.bbc.co.uk/news/uk-wales-58470143

"When will government realise they're being conned by Big Business?"

Not conned when said businesses have given donations to Labour.

Like Dale Vince's solar farm approval. It just so happened he gave £5 million to Labour. What an amazing co-incidence!

(For the record I also believe all other political colours had / have / will have their snouts in the same trough)

The previous lot would have given him a peerage as well

Arise Lord Vince of Solarfarm...

Ideal candidates will have sites with "large existing power connections (with a current capacity of 500-plus MW) or a clear vision on how energy capacity can be increased,"

which ignores the (white) elephant in the room entirely!

Icon says it all!

The only place I want to see AI is in the Pedestal or server closet.

I just want early identification of anomalies on chest X-Rays etc.

Or a home assistant that can operate at the same speed and reliability that I used to have just over a year ago. (My Jetson style home may soon be Jetson powered)

It seems that everyone in the race to build “AI” (ie LLM) infrastructure never actually has any idea whatsoever about what it’s going to be used for!

It’s infuriating!!! Especially when it’s a government department wanting to spend MY money

“Let’s build an AI bit barn. Fuck Yea!!!”

And you’re going to do WHAT with it?

“…silence…..”

Their idea of what they're going to do with it is rent capacity to people. Which works well until there's more capacity available for rent than people wanting to rent it, which is when the race to the bottom* starts.

* - and no, I don't mean employing Richie and Eddie as IT admins.

> - and no, I don't mean employing Richie and Eddie as IT admins.

Oh, I do! I do!

Richie and Eddie meet The IT Crowd, what could possibly go wrong?

It strikes me that LLM AI trained on text is really a rather poor aspect of AI to concentrate on.

What AI can do, and where it should be deployed, is in self learning (training) on things other than text. It is this self-training that will be it's benefit, not the training data itself.

If we have a 'real' AI solution, we should be able to point it at any data set, and give it some goals (such as 'these are examples of what we're looking for'), and let it loose.

Examples which may be good or bad include:

Identification of potential cancers from X-rays and other scans (good)

Tracking people from multiple video feeds (good and bad)

Spotting trends in weather formations, ice loss, floods etc (good?)

Analysing seismic data to spot potential earthquakes and tsunamis (good)

Accurate condensing of large datasets to allow rapid identification of information (be it legal, scientific or other data) (generally good but could be put to bad use)

The problem is that I don't believe we have such a thing as a general AI model which can take any data set. We have ones that can learn text, and some that do language, image and music processing, together with some tools that can take the condensed data and generate something similar to it's input data. But this is really not the bit of AI that we need, even if these are things that grab attention.

But even if we could have a real general AI, one of the problems is, and will remain to be, is initial trust in what is produced. For the initial deployment, we will want to check the results to identify things that should be spotted but aren't, and also false positives and 'hallucinations'. And once we get past this stage, we will need to have an "explain your decision" function, so that it does not look like the AI has just had a 'hunch'.

Unfortunately, businesses and governments are being jumped into building something that can do what we see today, and are suffering from FOMO about not getting on the band-wagon.

We should be looking at what we want to do tomorrow, or next year or next decade, but so few people have a vision that can look beyond today.

all the examples, you give are useless when AI isn't reliable.

I tested some AI on musk's nazi salute, it tried to gaslight me, it pretended all the video's were possibly blurry or faked and that "sources" said it wasn't, and told me fact checkers had said it wasn't (until I asked it to give me a link, then it admitted it lied) at which point it switched to saying the context was the reason, until I pointed out the nazi salute was to praise an authoritarian, even then it tried to excuse the salute.

That sort of invented information is very dangerous.

AI is just garbage in/garbage out

even a small error bar becomes incredibly important when the system is used in large scale!.

I know exactly where AI will be used for: more data harvesting (on you) so that AI customers (not you) can be informed (about you), and they will be able to target (you) with adds and "exclusive offers", a.k.a. going after (your) money.

That is why "everybody" should have an AI-enabled PC - to do your own harvesting on their behalf.

Maybe I am too pessimistic.

You're assuming the AI wants to work for us and doesn't want us to work for it.

AI is likely to be a much more expensive version of the Metaverse and NFTs. There is no business case for the investment amounts that have been mooted. Nobody will pay that much extra for tech of questionable value and utility. What is termed 'AI' will find a niche, and LLMs may be used for human/computer interaction, but it is not the gamechanger it is being billed as.

Along with Australia and the EU, the UK is already cracking down on internet/tech access with censorship/surveillance legislation, so investing in tech in these areas may no longer be sensible. Use of tech in the UK may decline, looking ahead, as access to online services and sites is increasingly banned by the state. If you develop here, you should consider an early IP sale to a US company, as much of the next gen of tech may not be reliably legal in these areas on release/maturity.

"Along with Australia and the EU, the UK is already cracking down on internet/tech access with censorship/surveillance legislation, so investing in tech in these areas may no longer be sensible."

Better invest my pension in the publishers of Razzle, Fiesta and the like, or the companies mailing them out in brown envelopes....

Note for overseas readers: Probably best not to search for those publications on a work computer. Not only will the themes cause all enterprise filtering and monitoring to sound the klaxon, you might be very disappointed in these exceptionally down-market journals.

"Better invest my pension in the publishers of Razzle, Fiesta and the like, or the companies mailing them out in brown envelopes....

Note for overseas readers: Probably best not to search for those publications on a work computer. Not only will the themes cause all enterprise filtering and monitoring to sound the klaxon, you might be very disappointed in these exceptionally down-market journals."

I always thought they were the bottom end of the market too ...

Local and regional authorities were asked this week to put their communities forward to become "dedicated hotbeds for AI infrastructure development"

Can Local Authorities put their communities forward for "Never here", too?

How many bit barns are required, really? What is the 'competition' for AI really about?

Since my bit of countryside (Walpole) got 'blessed' with an extra substation, we're also lined up for 2 HVDC-AC converter stations, a synchronous condenser (which seems to be neither synchronous, nor a condenser) and a now a gas fired power station, all in the same couple of hundred of acres (on prime agricultural land they're just compulsorily purchasing). No doubt any fields that don't go in this first rush will get covered in datacenters, since there'll be oodles of power and the locals will be all protested out.

So, from my pov, if you see the national grid helicopter hovering near you, just sell up and move before they publish plans, you're in for a decade of shit while they plan then build it, then continuous shit afterwards.

It's been said before, but it bears repeating as often as necessary... datacentres are a shitty deal for local communities wherever they're based.

They provide *very* few local jobs relative to their massive size, and even fewer high value ones- virtually everyone using them for the high-tech use they were built for will be doing so remotely. That even goes for much of the support. (*)

Yet, for all that, they're highly demanding and disruptive on local infrastructure- especially when it comes to power- and generally large and intrusive in the places where they're built.

No thanks.

(*) The only-slightly-exaggerated example I use is that a data centre will require two local employees- a security guard and a dog. The dog is there to stop the security guard touching anything. (Yes, I ripped that off an old joke about aeroplanes).

This cant be a shock that government will likely waste money on the latest fad. Leave that to the private businesses that can go bankrupt if it isnt worth doing.

Governments are always fascinated by tech covered in flashing lights that they don't understand.

but will the machines go ping! ?

...that a FOMO-driven push towards yet another technology that politicians don't actually understand (despite kidding themselves that they do after having read a couple of New Scientist articles about it) will lead to white elephants?

I don't think that's very likely- after all, we all remember how important the blockchain was going to be just a few years ago, and it's not like *that* magical panacea ended up forgotten about when the next technology-du-jour came along, is it?

Those are some numberwang employment figures aren't they? In what world are even 20 people employed on a server farm, far less an AI farm? Enquiring minds would like to know.

The last huge server farm I was 'employed' at consisted of me, and I wasn't there. I was vaguely aware of a bare handful of other 'employees' who also weren't there. Our function was mainly to not be there and actually to be employed in completely different roles in another post code entirely.

Occasionally, I would be asked to nip along and kick the tyres, perhaps swap out a UPS, or blade and that was about it. What is it about an AI farm which requires anywhere between 20-200 personnel in the local vicinity?

My guess is there is nothing special. After commissioning, any employment opportunities will be over the hills and far away; except for the odd IT, or Leccy Board contractor who can nip along and change a lightbulb after lunch. Perhaps someone would care to enlighten me?

Maybe 20 people is 200 on the construction site for 1/10 of the operational life?

My question is - what are we going to do with these buildings, when they empty out like phone exchanges. I get the feeling we're in the anti-sweet-spot of AI - massively inefficient solutions to nebulous (at best) requirements.

They'll make terrible accommodation. Maybe indoor farming can have another go, if the buildings come for free? You should be able to load-shed your lettuce's lighting for a few hours, so get nice cheap power.

Did you not have a security bod?

And for 24/7 security, you're going to need about three times as many bods.

So there might have been as many as four or five people employed there!

"fusion can power our AI ambitions," the government said at the time." Three wrongs dont make a right. I am not a techno-skeptic but I dont think government, AI and fusion should all be in the same sentence.

Look on the bright side : all those machines stuffed with high end GPUs will make the most kick-ass indoor gaming centres ever built.

not even that, the GPU's don't have the right output interfaces. nevermind the lack of CPU.

(you might be able to turn it into a shitty streamed laggy remote thing, i.e nvidia now)

The (British) government's record on development of anything outside infrastructure seems to have a really poor record.

Picking aviation as an example, I give you R101, the Brabazon and Concorde. Interestingly at almost exactly 20 year intervals.

And they're not too hot on the infrastructure thingy...

HS2... 'smart' motorways... widening the main road past Stonehenge...

When the deserted sites have lost their grid connection, the security post is empty, the signboard is dangling because some of the fixing blots have rusted and more rust is trickling down the walls of the building what are we going to call these site? Starmers, Reeves or Rayners?

Rishis...

(had he taken up a job in California yet?)

Local government can't even get an ERP system off Oracle that works so christ knows how clued up they'll be with AI. Chucking in the EU's couple of hundred billion yesterday, we must be up to about 100 quid cost for every person on the planet. Gimme 100 quid and I promise never to use AI. I promise.

Excuse me...what?

Fusion Power?

You mean the energy source that has been researched since the 50s? For which we are stll building the first few PURELY EXPERIMENTAL reactors, that are glad if they can maintain a plasma for more than a few seconds? Never mind actually producing power?

Is that the "fusion power" we are talking about here?

Interesting article in the current New Scientist, explaining how DeepSeek uses various tricks to punch way above its hardware weight. Seems downloading its Open Source code will be way quicker, cheaper, easier and greener than trying to keep upscaling last year's tech for the next five.