Posts by JacobZ

Excel's Copilot function is headed for the Recycle Bin

So how is it going with the strategy...

...of shoving AI into every nook, cranny, niche, and orifice of every piece of software?

It is fair to say that COPILOT() filled a much-needed gap.

Smart glasses are only smart if we train them to be good. Then they'll be fantastic

My fond hope for pervglasses...

...is that Meta decides that displaying ads in the glasses is the way to make the business model work, and that turns off even the kinds of creeps who would buy these things.

I hope that it is only a matter of time, since to the Zuck we are never the customer, always the product.

Re: "Essential"?

This. A guide dog is never going to narc on you to law enforcement, nor does it retain an accessible record of what it sees, nor is it a tool to target you for even more advertising.

Just no.

There is no possible way to redeem a device that non-consensually and secretively points a camera at people. Remove the camera and maybe we'll talk.

It is not at all surprising that Zuckerberg does not understand this given that the original premise of Facebook was to non-consensually "rate" the hotness of women who would not go out with him or his dorky friends, and that everything that has happened in Facebook since then has demonstrated an equal disregard for privacy or respect for the individual.

It's also worth bearing in mind that Zuckerberg would not be where he is today had there not been enough people perfectly comfortable with that original model of Facebook -- provided of course it wasn't pointed at them.

Salesforce reels in customer support AI specialist Fin for $3.6B

ChatFAQ

If 75% of a company's customer support questions can be answered by searching the FAQ, maybe they should make the FAQ more visible.

AI code accelerates production failures and spending, study finds

Obvious solution

Have they considered... you know... not pushing code to production before it's been validated?

And yes, that will lead to a backlog of crappy, production-unready AI-generated slop, but (a) that's better than breaking in production and (2) solved by investing in more validation resources.

Google accused of pushing 'free for life' G Suite users onto paid plans

Conspiracy theory

IIRC Google Workspace commercial accounts come with non-optional AI "services", a change Google made - along with a price hike - to juice the numbers it claims for use of its AI. Obviously I can't say for certain that pushing people onto paid Workspace accounts is part of that effort, but it is certainly convenient for Google's numbers.

On a related note, if you subtract Workspace customers who didn't ask for it and Search users who didn't ask for it and GMail customers who didn't ask for it, how many people are actually using Google's AI by their own deliberate choice?

Wanted: Digital chief for England's schools. Must enjoy data, AI, and concrete problems

Concrete problems?

Because of the number of schools that are falling down from years of under-investment?

Backup script ingested an accidental asterisk and deleted everything

Intelligence

You know what would really impress me as an example of AI? Capability built into the OS that could spot something like this and go "I suppose it's possible you intend to erase everything, but it seems unusual. If that is really what you meant, [do some elaborate procedure much more complicated than simply clicking "Yes" in a dialog box".

Of course, this would probably require a _real_ AI that understood something about systems and ops, not a stochastic parrot that had reduced the entirety of the internet to pink slime.

Re: Restored, which is good.

I once made that mistake on a disk drive with an integrated cartridge drive which had a great feature that allowed you to clone the disk to the tape with a single button push, or vice versa. Which is great right up until the point you accidentally do "vice versa" with a blank tape...

Trump-branded datacenter project fails to make itself great, again

More red flags than a Moscow Square parade

For legal reasons I'm not saying that this AI infrastructure project is a scam, but if somebody were to attempt an AI infrastructure scam it would look an awful lot like this project.

Anthropic reveals $30bn run rate and plans to use 3.5GW of new Google AI chips

Lies, damned lies, and annualized run rates.

Anthropic says lots of things in unaccountable press releases. When its CFO Krishna ‌Rao made a court filing under oath, he stated that revenue "has exceeded $5 billion to date.”*

So we are supposed to believe that a company with $5B lifetime revenue had an ARR of $9B in 2025, $12B mid-February, and now $30B.

Also, I'll believe that Anthropic is serious about buying those chips when Google files a quarterly financial report that shows a corresponding bump in their revenue forecast.

----

*This was in their lawsuit with the Pentagon. Filing can be found at https://storage.courtlistener.com/recap/gov.uscourts.cand.465515/gov.uscourts.cand.465515.6.5.pdf

'Uncle Larry’s biggest fan' cut by email in early morning Oracle layoff spree

Revenge is the best revenge

Many years ago I "separated" from Oracle. I got paid, including for vacation; got my vested options; got my CORBA.

And then about two weeks later, I got the HR letter asking me to sign that I would never say anything disparaging about Oracle. I sent the letter back with a little note saying "Why would I ever sign this?"

And then I got a job as a tech analyst [not Gartner] and said lots of disparaging things about Oracle.

OpenAI gets $122B to 'just build things' as the world blows them up

Re: Strange times

...and the rest will be just wasted

(Nod to George Best)

Usage pricing leaving software vendors guessing what lands on the invoice

That's an ad

That's an for m3ter via a "report" they commissioned from PwC that astonishingly concludes that people should be using m3ter's products.

GitHub backs down, kills Copilot pull-request ads after backlash

Nominee for Weasel Word of the Year

Using "tip" for "ad" has to be an early leader in the contest for Weasel Word of the Year.

Here's a tip: Pick a lie and stick to it.

So was this a conscious but tone-deaf decision to insert ads that they then realized was "icky", as stated by GitHub VP of developer relations Martin Woodward?

Or was it an unintentional "programming logic issue with a GitHub Copilot coding agent tip that surfaced in the wrong context", as stated by Martin Woodward, VP of Developer Relations, GitHub?

Here's a "tip" for you, Martin: pick a lie and stick to it.

Staff too scared of the AI axe to pick it up, Forrester finds

Re: mandated AI

Rewarding people based on how good their work is presumes that managers know what their people are doing and understand it well enough to evaluate it. How often is that true?

Mind you, I did have a boss once who told me "I don't really understand what you do, but people keep telling me how good you are and asking for you on their projects, so keep doing that." At least he was honest about it.

SoftBank to build massive AI datacenter on former US nuclear weapons site

Softbank

You know the end is near when Softbank is committing billions.

Re: 10GW

Serious answer: because the available power is the limiting factor in the capability of a datacenter. It's defined initially and baked into the construction, and increasing it is a major construction exercise.

By contrast, the amount of processing power is something you might target at the beginning of construction, but by the time the facility is online, you really don't know what processors will be capable of for a given power consumption. And even if you did, that number is going to change repeatedly over the lifespan of the building.

Similarly, the footprint of the building is not telling you much about the capabilities.

Payment biz pulls plug on open source charity after KYC spat

...or worse

Im fact, if they wanted to do that check, the LAST thing they should do is to ask FSFE for test credentials.

If FSFE were up to any funny business, they could rig their portal so that the test credentials would be recognized and take the user through acceptable cancellation flows, while continuing to do whatever shenanigans Nexi is worried about for everybody else.

Far better to, as you suggest, to anonymously set up an account like a real contributor without announcing their intentions to FSFE.

Everything about this smells.

IBM stock dives after Anthropic points out AI can rewrite COBOL fast

One of the problems with COBOL (and other old languages) is not just that the code is old and the programming expertise is scarce; it's that the code often IMPLICITLY implements business requirements that are not EXPLICITLY documented anywhere. The Requirements docs or design docs or functional specs are probably long gone. There are likely to be critical chunks of code where the last person who knew why it did a particular thing - or did it a particular way - retired decades ago.

That is a level of understanding - and a level of caution - that an LLM is never going to achieve. It's a big part of what other posters have referred to as the problem of ensuring that the replacement code is functionally equivalent to the code it is replacing.

And I'm guessing the people proposing this have as little understanding of what it was like to develop in the Olden Days as do the people selling off IBM over this latest scare.

Ring kills Flock partnership amid surveillance scrutiny

Big Brother

Imagine installing a video camera on your front door that you don't control, and that the vendor can turn on or off on a whim.

I can't imagine any way that might be abused.

Amazon can't build AI capacity fast enough, throws another $200B at the problem

Re: Is the return worth it?

"So how do we equate energy consumption to processing power?"

<Sensible>

There is actually a sensible answer to this.

During the lifetime of a datacenter (say, 15 years for a regular non-AI datacenter?), the processors are going to be replaced several times, (usually) increasing the processor power. So while processing power is a good measure of the initial capacity of a datacenter, it is also a constantly-changing number, and doesn't tell you anything useful about the capacity, say, five years from now.

The maximum power capacity, however, is a much more stable number, as it's a major engineering exercise to increase it. It's also often the limiting factor, especially in the current madness where the demand for power for datacenters far outstrips the ability to build out more (which is why you have, for example, Microsoft bringing Three Mile Island.) And third, it's hard to measure the processing power of modern datacenters. If you have a mix of CPUs (some ARM, some x86), GPUs, TPUs, etc., how do you add all those up to a sensible number that is comparable across different datacenters?

</Sensible>

We now return you to your regular El Reg comment thread in progress

Oracle expects investors to pump $50 billion into its cloud this year alone

Fantasy money

That $455B that Oracle claims it has booked is entirely imaginary. $300B of it is from OpenAI, which does not have the money. It doesn't even have $30B. It is burning cash so fast that it will have to borrow more than any company in history has ever borrowed to pay any of what it has promised.

But it gets worse. A lot of the remaining book is companies like NVidia and others buying capacity *on behalf of OpenAI*, which they will then lease to OpenAI -- which can't pay them either.

And worse again: Oracle does not get paid on (most of) these contracts - if it gets paid at all - until the datacenters are up and running, at which point it will be hundreds of billions in the hole.

The best thing that can happen to Oracle is that OpenAI fails quickly, before Oracle spends a lot more on datacenters it is never getting paid for.

Big Tech is delusional, Oracle most of all

Big Tech doesn't want to believe that anything but massive cloud-based LLMs can get the job done, regardless of reality. The promise of Big LLMs is that only a handful of companies have the money and scale to deliver them, and that those companies will make massive profits. And that is so seductive to tech CEOs it has completely clouded their judgement.

What's that Upton Sinclair quote? "It is difficult to get a man to understand something, when his salary depends upon his not understanding it." Now substitute "image of himself as a visionary leader transforming the world and also becoming as rich as Croesus" for "salary" and you get some idea of why tech CEOs collectively have lost their minds.

Yes, you can build an AI agent – here's how, using LangFlow

This again?

For the past 40 years I have been reading press releases that promised that non-programmers could code just by connecting boxes.

For about 20 years, they have promised that non-programmers could configure workflows the same way.

And now they promise that non-programmers can configure something something agent something.

I expect it to end in tears.

Banker claims Oracle may slash up to 30,000 jobs, sell health unit to pay for AI build-out

AI job losses

Finally, a real example of AI adoption causing job losses

'Ralph Wiggum' loop prompts Claude to vibe-clone commercial software for $10 an hour

What about original work?

So, after humans have done all the heavy lifting of soliciting requirements, developing code, testing, fixing, scaling, integrating, making secure, iterating to implement additional requirements because people are really bad at identifying requirements until they can see the thing and say "no, that's not what I meant!", thoroughly documenting the whole pile and then productizing the whole thing...

...an AI can reproduce that for $10 an hour (it doesn't say how many hours).

A photocopier can plagiarize a book for a lot less than $10 an hour, and that doesn't impress me either.

AI adoption at work flatlined in Q4, says Gallup

When somebody tells me...

...that AI helped them with their work, it doesn't raise my opinion of AI. It lowers my opinion of their work.

Microsoft CEO: AI sovereignty isn't where it runs, it's who controls it

Re: Datacenter location is the least of an enterprise's worries

"We are spending HOW much now?! How can we get out of this without losing face to the investors?"

The best thing that these numbnuts can hope for at this point is a major recession so that they can dial back their wild overspending and blame it on the economy.

Meanwhile the politicians will blame the recession on wild overspending on AI.

Everybody wins.

Well, except us, obviously.

corporate pirates the lot of them

"SatNad argued that corporate AI sovereignty hinges on control over models trained on proprietary knowledge"

He neglected to mention that their models are trained on other people's proprietary knowledge

Hyperscalers, vendors funding trillion dollar AI spree, but users will have to pay up long term

Re: Who is getting rich?

Yep. Mostly it's Nvidia, sometimes with systems builders like Dell or Supermicro.

Some DC builders may do OK, however some are in a risky position where they keep ownership of the building and lease it back to the company that ordered it, such as Google, based on the "promise" of massive AI usage that will definitely come any day now. And by "builder" here I mean the prime contractor. I hope that the subs pouring concrete and laying cable are getting paid up front...

Workers should control the means of agentic production, suggests WorkBeaver boss

All over by Christmas

If the AI vendors focused on the demand side, they would pack their bags and be home in time for Christmas

Trump Media jumps aboard the speculative nuclear fusion bandwagon

"There is a moonshot opportunity here, a version of the Apollo program."

Such a stupid comparison.

The Apollo program was a massively expensive taxpayer-funded science and engineering project. At the end of it, we did not have rocket ships for the masses, nor was that the goal.

A better comparison might be something like the companies that first gambled on bringing widespread access to electricity.

Breaking ground in 2026

If there's one thing Trump is really good at, it's skimming off a construction project.

Snowflake update caused a blizzard of failures worldwide

Whatever happened to...

...progressive rollouts? Update one region. Let it marinate for a while. Then do some more.

Or even testing before deployment?

"Breaking the database schema" seems like such a fundamental mistake there surely should be a test case for backwards compatibility of the schema.

Jassy taps 27-year Amazon veteran to run AGI org, which is now definitely a thing that exists

Irony?

It seems weird that AGI (which doesn't exist yet, and will not come from GenAI no matter how much money and how many GPUs they throw at it) is under GenAI (which doesn't work).

However, it seems highly appropriate that all of this sits under a Marketing guy.

Hyperscalers fuel $112B server spending spree in Q3

Bubblenomics

Phase 1: Everybody is afraid to be the last one in. Goldrush!

Phase 2: Everybody is afraid to be first one out. Surely the last one standing will make a fortune?

Phase 3: Everybody is afraid to be last one out. Game over.

We entered phase 2 somewhere around mid-year. It's in the nature of bubbles that nobody can predict when we tip into phase 3 - personally I thought it would come as early as Novemner - and when we do, it will happen extremely rapidly.

Bezos-backed Unconventional AI aims to make datacenter power problems go away

Re: Twaddle

One pedantic point: LLMs are not trying to emulate the analog processes of brains in order to reason like brains. They are merely trying to emulate the outputs of brains in the belief that if you scale that large enough, reasoning - or intelligence - will somehow magically appear. Or as linguists might say, if you imitate surface structure well enough, deep structure will magically emerge.

In other words, it's Cargo Cult Computing.

Or if you prefer a cruder analogy, it's like shoving sh*t up a cow's arse and expecting to get grass out of its mouth.

Cloudflare suffers second outage in as many months during routine maintenance

Re: I couldn't give a.....tinker's damn!

El Reg comment threads are antisocial media.

Campbell's CISO canned after lawsuit alleges hour-long rant against staff and customers

Bio-engineered meat insanity

Just for the record, bio-engineered meat costs around $60/kg. Real chicken costs less than $2/kg wholesale, probably even less in the industrial quantities Campbell's buys it in. It would be absolute insanity for them to use it in soup.

Anthropic is at the heart of the latest billion-dollar circular AI investment bonanza

Re: Hot air (and water)

The limiting characteristic of a datacenter, the one that is reasonably stable over time, is how much power it is supplied with and can distribute to its racks for compute and for cooling. The amount of compute, however you choose to measure it, is going to vary over the life of the datacenter, for example if you upgrade the GPUs (megaflops) or change the model (tokens).

So power capacity is both the only thing that is stable and the only thing that is reasonably predictable from the outset.

Stranded assets

The technical term is "stranded assets". And the people holding the bag are the private capital firms who accepted the GPUs in the bit barns and/or the Big Tech leases that "promised" to use those bit barns as collateral for their loans. And possibly the traditional investment firms who lent money to the private capital firms.

The whole thing is the absolute classic definition of a bubble. I suspect the people investing in it already know this, much like the 2008 housing bubble, and the game now is simply to not be the one holding the bag when the music stops.

Commodity memory prices set to double as fabs pivot to AI market

Just in time...

...for the AI bubble to pop, and the chipmakers to have to switch back again.

OpenAI’s viability called into question by reported inference spending with Microsoft

Microsoft knows best

Microsoft knows exactly what OpenAI's revenues are, because it gets a cut.

And it knows what OpenAI's losses are, because it has to report its share in its own quarterly reports.

And of course it knows how much (little) it is paying OpenAI for reselling its models on Azure (bear in mind that under 2% of Microsoft 365 users are licensing Copilot).

And knowing all that... Microsoft is partnering with Anthropic (https://www.cnbc.com/2025/09/24/microsoft-adds-anthropic-model-to-microsoft-365-copilot.html)

Big Tech's control freak era is breaking itself apart

Re: As irrational as it sounds...

"Experience is that thing you get right after you could have used it."

Agents of misfortune: The world isn't ready for autonomous software

Finally...

...a meaningful comparison between AI and Uber:

"Don't waste time asking for permission. Establish market dominance first, then ask forgiveness."

By the way, is it a requirement to be an AI leader that you throw out blatant lies to promote your product, or is it just a coincidence that OpenAI, Anthropic, and Perplexity all do that?

Famed software engineer DJB tries Fil-C… and likes what he sees

Re: Type checking and compatibility

FWIW, and that's probably not much, I prefer "bool flag := <bool expression>". It is a useful redundancy that ensures that the (potentially hard to read) expression is of the type that the author and later reader think it should be. While this is not a big deal for bool, it can be a big deal for numeric types or for references (how many C bugs occur because something is a pointer to a pointer, rather than just a pointer?)

Also FWIW my own language project [actually a pre-processor to Go since they won't add it to the language] goes even further with compile-time checking. It allows the programmer to define named types, much like any language, and then also "compound types", or as I call them, dimensions. For example, in C-ish pseudo-code it might look something like...

type Meters double;

type Seconds double;

type Velocity (Meters/Seconds);

And then if you have m, a variable of type Meters, s of type seconds, and v or type velocity, you can write:

v = m/s;

or

m = v*s;

but not

v = m*s; //wrong dimensions, failed by dimension checker even though underlying types are all double

The dimensions will be checked at compile time, leaving runtime code as efficient as if the dimension checker never existed.

Of course, this is a relatively trivial example. The value becomes more apparent the more complex an expression becomes, and when calculations are chained together.

Re: Not quite ...

Is the hammer responsible for bending nails?

Tony Hoare certainly thought so, and he knew a thing or two about writing languages: "The author of a language is responsible for errors commonly made by its programmers" (wording may not be exact, intention is.)

When I was learning Rust and I got to the sentence that began "A common mistake made by Rust programmers..." I threw the book against the wall. (Not my only issue with Rust).

Look at this way: if you had two hammers, and in the hands of the same skilled carpenter, one of them drove straight every time, and the other one bent one nail in ten, would you say that the hammer was responsible? Because I sure would.