Posts by Mike 137

Coders paired with bot buddies work fast, but take too many shortcuts

Expertise?

AI assistants were good at reminding humans of key details, "such as committing database changes, that might otherwise be overlooked."

Overlooking such basics says much more about the skills and attention capacity (or lack thereof) of the practitioner than it does about the utility of "AI". I never cease to be amazed by the shallowness of many of the questions on sites such as SO -- they lead me to believe that the software development profession is to a great extent populated with mere 'coders' who are novices at programming (two very different skills).

YouTube's AI moderator pulls Windows 11 workaround videos, calls them dangerous

Another workaround

If you have something important to say, host it on your own web site and get lots of folks to link back to it (the old way the web worked before Grungle took it over). Relying on third parties with unchallegeable "moderation" will always fail you if you're saying something that might annoy the moguls.

Brit boffins teach fusion plasma some manners with 3D magnetic field

Re: Cynicism is easy

"this sounds like a big step forward"

Yes, another of the sort of major breakthroughs that have made Culham hightly reputed for half a century or more. It remains to be seen whether its move into "AI" will maintain that reputation.

SpaceX is behind schedule, so NASA will open Artemis III contract to competition

Just like the Orangeman

Elon Musk, the boss of SpaceX, fired back: "SpaceX is moving like lightning compared to the rest of the space industry. Moreover, Starship will end up doing the whole Moon mission. Mark my words."

"Never apologise, mister. It's a sign of weakness" ^[1]

_{[1] John Wayne in "She wore a yellow ribbon"}

.

Aid groups use AI-generated ‘poverty porn’ to juice fundraising efforts

Right on the nail

" the organisation forms a life of its own that's only vaguely aligned with what is donors might think it does"

Not so long ago I applied for a senior post at a charity officially dedicated to alleviating world poverty, but withdrew my application when the JD came back headed "[charity name] is a feminist organisation". What that had to do with alleviating world poverty was not made clear.

AWS admits more bits of its cloud broke as it recovered from DynamoDB debacle

"a rouge state"

If it's a rouge state they're probably pink envelopes.

Windows 11 update knocks out USB mice, keyboards in recovery mode

Re: The question is...

Hey! I didn't realise it had reached alpha -- (I thought they were still "designing" it)

Lloyds Banking Group claims Microsoft Copilot saves staff 46 minutes a day

"undertaking due diligence"

Lovely cliché that - "due diligence", but almost universally misunderstood. "Due" in this case means "appropriate and sufficient", so if you palm the decision making off on a machine, your "diligence" is by definition not due, as that would require you to assess the situation, think out what is required to manage it and take responsibility for the outcomes, whether good or bad.

However, I'm not surprised at banks cutting corners like this -- it's not for nothing that the collective noun for them is "wunch".

Major AWS outage across US-East region breaks half the internet

Re: Too much in us-east-1

"There's really no excuse for British companies to be reliant on that region in most cases"

An analysis on the BBC lunchtime news suggested that a technical sub-component (e.g. DNS) based in US-east-1 may be sued by the hosting services based in UK/EU. So the customer is not either directly or knowingly using US-based services at any level.

A simple AI prompt saved a developer from this job interview scam

Errrrrr...

developers are the "ideal victims" because their machines "contain the keys to the kingdom: production credentials, crypto wallets, client data."

All that on the one machine and single account you use for job applications?

Having worked extensively online almost since the web went public (35 years+, ouch!), I've always kept a "dirty machine" for such tasks, with nothing but the basics on it and with a clean backup image that can be used to rebuild it from scratch if it gets contaminated.

In my infosec consulting experience, the key reason most organisations (and folks) get "hacked" is that they have no real proactive defences in place (you need more than a few appliances -- you need forethought, current information about threats and the willingness to make the necessary constant effort).

The real insight behind measuring Copilot usage is Microsoft's desperation

Highly refreshing

Nice article Rupert. It's delightful to read objective analysis in this bullshit-driven arena.

Turns out the end of Windows 10 is good for something: The PC refresh cycle

Amazing, really...

It always strikes me as amazing (being maybe a little sarcastic here) that, as soon as the next version of an OS or application is launched (or at least once "support" for its predecessor ceases), that predecessor suddenly becomes lethal. The reality of course is that [a] it's always been lethal and [b] its successor is also lethal (needing constant repairs for its entire operational life). I think it's safe to assert that there's never been an OS or mainstream application to date that has been fully fixed before it went obsolete, because software development is still crap in engineering terms.

Boris Johnson confesses: He's fallen for ChatGPT

"A perfect match!"

but what about the offspring?

Feeling lonely? Microsoft Copilot can now listen to your every word, watch your screen

Re: Snoopers charter

"... Panoptikon concept was developed for prisoners"

Exactly! Once they've got you by the balls they can twist as hard as they like.

How chatbots are coaching vulnerable users into crisis

"That's partly because many of these models are sycophantic, telling users what they want to hear"

So we're just dealing with automated con men really. The automation makes them more accessible and potentially more persistent, but the technique is as old as the hills -- find an insecure person and flatter their self-image till they accept anything you say. The real underlying problem is the proportion of the population that's so insecure these days that they can be caught by the scam. That's largely down to education systems that just stuff folks with facts rather than cultivating their ability to exercise attention, perception and judgement. As Dirty Harry said "a man's gotta know his limitations" -- if one does one's less likely to be fooled into fantasies like believing you've discovered a new branch of math without any training in math.

"it had passed the Turing test"

The two problems with the Turing test are:

1. It was a thought experiment only, not intended to be used as a validator;

2. "Passing" or "failing" it depends at least 50% (probably a lot more) on the perceptual capacities and relevant knowledge of the observer, rather than on the performance of the machine. It's therefore got about the same level of absolute validity as Trump's assertions about his intelligence.

Bank of England smells hint of dotcom bubble 2.0 in AI froth

Re: For some applications, hallucinations don't actually affect existing quality

That's coz the references were created after the conclusions, (not the conclusions based on the references). This is now a standard approach used by undergraduates who don't like reading papers, so it's possible that the report was written by an intern.

"AI" is not one thing

There are now many genuine and useful AI tools (advanced descendents of what used to be called 'expert systems') but they're all one truck horses, each dedicated to solving one specific problem. There's no such thing as generalised AI and LLMs don't really have any kind of intelligence anyway (they're just glorified statistical auto-complete engines with huge data sets to draw on) but they're purported by their promoters to have "generalised intelligence" in that you can ask them any question and get a banal answer to it (whether nonsense or not). This is promotional bullshit, not reality -- indeed it's an open question whether LLMs can legitimately be classed as AI at all. So either the bubble will either burst or the market will eventually lose interest after it's been well fleeced.

Lowercase leaving you cold? Introducing Retrocide

Priorities

""It seems like quite a nice font," he said"

Ultimately, subjective aesthetics should be subordinate to legibility. The job of a typeface is to convey information without imposing strain on the reader. Designers could do worse for a start than take a few lessons from the illustrious Betty Binns (Better Type, Watson-Guptill, New York 1989).

AI pricing is currently in a state of ‘pandemonium’ says Gartner

Well...

"Adobe’s AI legalese initially required its customers to assume responsibility for copyright infringements caused by its software and services."

"Occasionally, vendors revisit those multipliers and increase the quantity of credits required to use their services"

"Other vendors require customers to pre-pay for “tokens”, but don’t explain that the cost of inputting a token [...] is much less than the cost of tokens their services create when responding to prompts"

It seems that, in what has long been a saturated market, IT as a whole has become merely a domain of dirty tricks to extract revenue. "AI" is just the latest example.

Everyone needs an AI phone. No, don't hang up, it's true

In the words of the late great Douglas Adams

"your plastic pal who's fun to be with"

Bring your own brain? Why local LLMs are taking off

Pardon?

"Larger models will always be more generally intelligent," agrees Perez

No they're not. As no LLM is in any way intelligent even using the most liberal definition of the word, "more" "and "generally" here are pure bullshit. Stringing tokens together in statistically probable sequences is not intelligence. It doesn't even need intelligence -- just the ability to count and a lot of data to work on.

BGP’s security problems are notorious. Attempts to fix that are a work in progress

Diminishing returns?

Complexity on complexity. At some point the overall mechanisms will become so burdened by multi-layer "protections" that they'll [a] start to slow down or become erratic, [b] lock out swathes of systems that haven't been kept suitably "updated", and [c] become increasingly attackable, leading to yet more layers of "protective" complexity being imposed on them. Result: jump to [a] and repeat loop infinitely. And given the prevalence of short term thinking leading to poor implementation, [c] is the most critical as it's the argument for the burgeoning complexity in the first place.

Who are you again? Infosec experiencing 'Identity crisis' amid rising login attacks

"Vendor insists passkeys are the future"

They would, of course they would. But...

It's bad practice to use the same secret credential for more than one account, so there'll be millions made by key ring manufacturers as passkey dongles proliferate for every user. And even if the single passkey unlocks a "password manager", not only is it a single identifiable point of attack, but each end application will still be dependent against attack on a password of uncertain quality (and given modern attack methods the bar must be set very high). Realistically, password managers are a convenience, not a security provision.

Re: user education

"Users just can barely remember enough entropy for even one secure password"

What do "entropy" [a] mean and [b] solve (and indeed what does "secure" mean)? Mostly "entropy" used by IT folks to describe apparent randomness (apparent being the operative word), but in an age of super fast trial and error and wide word rainbow tables it actually provides very little more protection than non-"random" but not obvious long passwords. There are numerous different attacks on password-based systems, each requires it's own dedicated countermeasures, and "entropy" is only one of them against one -- password guessing. Mathematically (consequently realistically) length is (exponentially) key to defence against cracking, but that makes "entropy" increasingly unworkable as human memory doesn't like meaninglessness. Furthermore, placing maximum responsibility for protection in the lap of the least informed of the hazards, as we have for decades done, doesn't make much sense.

An absolute reality is that passwords have their place but it's not everywhere, but that to be effective where they are appropriate they have to be non-obvious to the adversary but memorable to the legitimate user. Relying on "password managers" as a security measure (as opposed to just a legitimate convenience) merely sidesteps two issues -- [a] maybe passwords are inappropriate in some contexts, and [b] concentrating multiple credentials in one place creates a clearly identifiable target for attack. But where passwords are appropriate, it is the duty of those who understand the risks to impart them clearly, correctly and effectively to those who have to create their own passwords and to those whose duties are providing the ancillary countermeasures across the network. So let's stop just banging on about "entropy" and start addressing the realities.

Re: user education

Yes, education. But not "training". The key to success is explanation rather than just rules. When I undertake "user awareness" the first question I ask is "what is your password for?". I've lost count of the number of answers on the lines of "to give me access to the network" (witness the prevalence of "letmein" as a password). Of course it's not -- it's to keep others out of the network.

And as to hardware tokens, as often as not they're stored in a little pocket in the laptop bag so both can be stolen together.

The key failing is not getting users on board with the real issues by failing to explain their wider context, as they're not immediately apparent to non-technical folks, particularly those who haven't been informed about the wider business implications either. But both management and corporate "IT" are notorious for merely issuing diktats rather than explaining anything. Hence they create their own problem.

AI skeptics zone out when chatbots get preachy

Duh!

"consumers (and people in general) should be aware that their reactions to AI outputs may depend on their AI perceptions"

In the human space, reactions to advice will always depend on percention of the reliability and trustworthiness of the provider. This is a general truism (but addressing the narrow specific case of "AI" does get another paper published).

Alexa hits snooze on basic functions as alarms and timers KO'd in UK outage

Nice!

"I do have an Alexa that wakes me up with the radio and traffic reports for my commute"

I have an alarm clock radio that just plugs into the mains for power (no internet connection). It cost about GB £5.00 some ten years ago, costs a negligible amount for electicity to run and works just fine. I can turn the alarm off by pressing a single button, and adjust the alarm time, volume and station with a few key presses in complete privacy. So why involve a behemoth data slurper?

Vibe coding platform Anything arrives, our hands-on suggests caution

Analysis

"anyone can build and monetize total shit code as easily as selling t-shirts"

This is not even innovative. Decades ago there was UK TV retailer sitcom called "Never mind the quality -- feel the width".

Doctors get dopey if they rely too much on AI, study suggests

Mandatory reading

The "ironies of automation" paper cited in this article and "Does automation bias decision-making?" by Linda J. Skitka (Int. J. Human-Computer Studies (1999) 51, 991}1006) should both be mandatory reading for those developing "autonomous" vehicles, as they both (for different reasons) demonstrate that a monitoring-dominant condition is the worst possible place from which to take action in the face of unexpected events. As both papers were published at least a quarter of a century ago, it's amazing (and deeply disturbing) that the message has not sunk in yet. However a recent article at The Dispatch provides a clue to why this might be (a combination of Dunning Kruger syndrome and arrogance).

Poisoned telemetry can turn AIOps into AI Oops, researchers show

"Automating IT operations using AI may not be the best idea at the moment"

"Automating IT operations using AI may not be the best idea " ever.

Maintenance of IT ops (particularly security) relies on recognising anomalies and their significance -- processes that depend on understanding. What we commonly refer to as "AI" cannot think or understand anything, so its chances of success in this domain are minimal. Furthermore, because its "reasoning" is inscrutable, applying "AI" further distances the human expert from what is really going on, reducing ability to verify and validate. We do not live in a "Star Trek" series one world where ten thousand-year old computers still run things unaided and unsupervised (although in those Star Trek episodes where they supposedly did they were typically portrayed as getting things horribly wrong and needing to be shut down by Jim Kirk -- maybe there's a lesson there).

Your CV is not fit for the 21st century – time to get it up to scratch

Re: Master and the slaves

"From the perspective of retirement, my main career regret is that I wasn't a better plasterer"

Agreed. Having designed and implemented bespoke systems, guided corporate infosec and contributed to international standards for several decades, I now regret I didn't train as a plumber.

Windows 11 is a minefield of micro-aggressions in the shipping lane of progress

Re: The real question

"forcing an upgrade that requires a new computer with a new license fee"

This indeed accounts for new entrants to the W11 horror, but not for subsequent "upgrades". I have a horrid suspicion that for these there may not be a corporate strategy at all. It's entirely possible that continuing development is in the hands of juvenile self-styled whizz kids who believe that, because they're geniuses, every idea they have is so wonderful it must be forced on all users. I've met this attitude on a smaller scale when collaborating with developers where the products functional design was vested in myself -- on occasion they have tried to wrest the functional design from me because they thought they "knew better" what the market supposedly wanted, regardless of whether their proposed features were appropriate to the product.

One in six US workers pretends to use AI to please the bosses

"Jacqueline Samira, CEO at Howdy.com, says [...] employees also need to embrace the new tech that's coming down the pipe"

Why, specifically? Some things that come down some pipes are definitely better not "embraced".

BTW, I've just marketed a snodfangle grunging grabber, so everyone must adopt it (whether or not it's actually useful).

Firefox 141 relieves chronic Linux pain in the neck

"Firefox's four-weekly release schedule"

Churn for the sake of churn -- who needs it? The safest and most productive tool is in general the one you're most familiar with. OK, so given current dev methodologies (AKA sloppy coding practices) we may need regular bug fixes, but new "features" every month? WTF.

Humans can be tracked with unique 'fingerprint' based on how their bodies block Wi-Fi signals

From the paper ( 4 Experimental Results and Discussion, 4.1 Dataset): "the dataset collects the CSI measurements of 14 different subjects"

Rather a small population, but, in all fairness, this work is not about unique identification of individuals but about tracking based on an initial acquired signature. So it's a neat piece of research.

UK government swoons over OpenAI in legally meaningless love-in

"build the UK into an artificial intelligence powerhouse"

As the intent seems to be primarily focused on LLMs, this really translates as "turn the UK into a leading bullshit merchant". But supposing it actually takes off (which is not at all assured) I guess quite a few pockets will get well lined before its futility becomes apparent.

If you're forced to use Windows 11, here's how to steal some of your time back

"Windows 11 is now the most popular desktop operating system..."

Possibly not (a lot of folks hate it, as this article suggests) -- just the most used. And that is most likely because, for quite some time now, if you buy a new machine W11 is all you get offered. In IT, churn, not voluntary choice, drives adoption.

EU-sponsored report says GenAI's 'fair use' defense does not compute

But how?

an "unwaivable right to equitable remuneration for authors and rightsholders whose works are used in such training."

A wonderful idea in principle, but how to make it work is quite another matter. For example, supposing someone hosts an online repository of their own original textual works. Suppose it's free for all to read, but how would the author ever find out that it's been scraped by an "AI" bot for training? It's highly unlikely that the hugely wealthy owners/trainers of the bot will spontaneously volunteer to the "little man" that they've scraped it and the cost of verifying would be prohibitive. As with many good intentioned EU initiatives (e.g. the GDPR) they make the erroneous assumption that big business plays fair. If it did, there would be no need for a law.

Ousted US copyright chief argues Trump did not have power to remove her

Re: Власть

Not so much власть as блат

Microsoft Windows Firewall complains about Microsoft code

WTF?

"a feature that is currently under development and not fully implemented"

As a quite long in the tooth engineer I have to ask -- what on earth is an "under development" "feature" doing in a production release? Or have we reached the bottomless pit where there's no such thing any more as a production release? If so, that would explain a lot about the abysmal quality of current software.

Call center staffers explain to researchers how their AI assistants aren't very helpful

You don't really need AI for that. It's already a refined art

Re: Disappointed with language recognition

As the farmer said, my question is whether a wether should be left out in this weather.

Not really AI at all

"AI tools, such as large language models (LLMs), emotion recognition, and speech-to-text technologies" [paper, introduction para 4]

While assuredly artificial, none of these have intelligence -- they're merely varieties of template matching system.

We do have, and usefully use, not a few genuine AI tools -- indeed in some pretty sophisticated applications -- but they're all essentially "one trick horses" -- expert systems trained on specific quite narrowly defined problems. There's no such thing as general artificial intelligence, and never can be because none of the tools can actually think. We merely contribute to the bullshit and the promoters' bank balances by calling fancy template matchers and autocomplete engines "intelligent".

Oh, and BTW, a UK health service rep recently promised that "AI" would soon be used to write medical case notes ...

SpaceX's Starship explodes again ... while still on the ground

"The stainless steel vehicle can now explode before even leaving the Earth"

Why stainless? What's the need to protect a rocket against rust? Any competent engineer will tell you that stainless steels in general have poorer properties relevant to pressure vessels than steels that have been specially designed for those purposes. But I guess the muskrat has (yet another) arbitrary fixation on "stainless" -- witness that "Chelsea tractor" of his, and, as in that case, he could have gone for a cheap grade.

European consumers are mostly saying 'non' to trading in their old phones

"so what can we do about it?"

Argue. If enough people publicly complain about the (ludicrously insecure) SMS transaction authentication the banks have forced on us, someone might eventually get the message. But the big problem is that the banks are not so interested in security - they're primarily interested in liability transfer and transaction volume. So mostly they go for convenient if less robust approaches. However, one bank I know of in the UK will (if asked) issue an out of band one time key dongle as an alternative. But they don't advertise it -- you have to go and ask for one.

Brain activity much lower when using AI chatbots, MIT boffins find

Although I agree entirely that those students who rely on LLMs to write their essays are (by definition) not learning much in the process, the EEG has only limited relevance. It's a very mushy measure that correlates extremely loosely to specific mental processes. A more robust measure of learning might be to set a subject, accept the resulting essay and then grill the student verbally about it (used to be called the "viva").

"...people who use ChatGPT to write essays aren't learning much"

And bears shit in the woods. In the process of commoditisation of the education system the fact has been increasingly disregarded that essay assignments are a tool for imparting research and analysis skills and factual knowledge (we educationalists call this "learning a subject"). If you don't actually learn the subject, your degree is meaningless, and in fact (in most "western" countries at least) the degree has ceased to impart the credibility it once did - here in the UK it's effectively the new A-level and has been for some time.

US to deny visas to foreign officials it says 'censor' social media

"... It is unacceptable for foreign officials to issue or threaten arrest warrants on US citizens or US residents for social media posts on American platforms while physically present on US soil" Rubio said.

"It's OK for us to do this though."