Posts by Fonant 580 publicly visible posts • joined 20 Mar 2014
This is one of the problems of "AI" (GPT / LLM) - it can't be trusted to be accurate or correct, and its trust levels can only ever go down as more and more slop is generated using it.
The other problem is that it's VERY expensive on compute resources.
It used to be said that you could pick any two of:
* Short timescale
* High quality
* Low cost
With LLMs you get to choose one of:
* Short timescale
along with:
* Unknown quality
* High cost
Article:
"Published Thu 09 Jul 2026"
Also article:
"The aerospace agency sent commands to the probe last July, instructing it to commence hibernation on August 7 and then resume activity in July 2026.On July 23, NASA checked to see if New Horizons had obeyed the instruction to wake up and was pleased to find it was online again."
Someone predicting news 24 days in advance? Relativity? Hoax?
The Psion apps were amazingly well written. Highly efficient, no need to "Save", cross-application embedding, but most of all a UI that was both simple and usefully flexible. Embedding program IDs in the headers of data files was clever too. No need for three-character filename extensions to indicate file type with a Psion :)
The 3A was brilliant in that it would run for a whole month on two ordinary AA batteries. I used mine a lot, Agenda, Spreadsheet, Data, and a whole bunch of freeware and shareware.
I live in hope that he starts to believe his own bullshit, and the AI bullshit from the others, and all these obscenely-rich people hop on a Starship to emigrate to Mars.
Then the rest of us can sigh with relief and get on with being nice to each other and cooperating for a better future.
The "AI" mega-corporations can fiddle with their models and "safeguards" as much as they like. LLMs are still just bullshit generation systems. Highly plausible, but not necessarily accurate or correct.
Google said the quiet bit out loud recently, telling German courts that "everyone should know that our AI summaries can't be trusted".
As title.
"AI", as in LLMs, has no intelligence at all. It's fancy statistical models based on hastily-collected and almost entirely un-curated contents of the internet. LLMs can't even be reliably correct, let alone accurate. It's a very-very-expensive scam.
Icon: what the bubble bursting might look like.
The cost is what will save us all, I hope!
A few massive corporations have apparently invested BILLIONS in this LLM thing, and they're going to have HUGE ongoing costs for power and cooling in the massive data centres they're building.
LLMs might be popular while they're still free, or cheap, but that can't last much longer. LLMs will still perhaps be quick to generate stuff, but they'll still be randomly wrong and they'll also be very expensive. How much are people willing to pay for highly plausible bullshit?
Programming is deterministic: for any given set of inputs, the output should be the same. Humans doing programming are taught rules of coding, and learn from experience what types of thing cause bugs. We learn to consider unexpected inputs, and make sure any possible input is handled correctly. Before making code changes, we think about the possible side-effects. We don't just type something that looks plausible because we've seen it lots of other places.
LLMs are statistical, and their models are constantly changing. They will produce output that looks highly plausible (because it uses words that are statistically very likely) and that seems to satisfy the problem as described to it. But it's just language probabilities, there is no "intelligence" involved at all. Even though us humans are so keen to anthropomorphise everything we see that looks even a little bit human-looking.
The test suite re-writing by LLM does seem on first inspection to be a Bit of a Nightmare.
Writing test suites that are useful is harder than writing the code to be tested. You want the test to fail if a bug exists, whereas an LLM agent will tend make all tests pass, giving them plausible names to make it look like it has done a good job.
You want code that works without bugs, the LLM just "wants" you to be happy. Not that an LLM has any intelligence at all. It's just a sophisticated (and VERY expensive) plagiarism machine.
LLM output is bullshit: highly plausible, but if it's correct or accurate that happened entirely by accident.
Google used to be a household name because of their excellent search engine. This allowed them to sell advertising, and their business boomed.
Google will soon be infamous as one of those brands that died because people just don't want "AI" bullshit or subliminal advertising.
Salespeople and Big Business love spending obscene amounts of money on advertising, to sell things to people that they otherwise wouldn't buy. This is what makes Big Business so powerful - more powerful than national governments these days!
Individual humans don't like being tricked into spending money on things they don't need. They don't like bullshit, either.
nine in ten students who use AI in their studies have already encountered problems with it, including factual errors and completely fabricated sources.
That's the problem with "AI" (statistical LLM systems) - they generate highly plausible, but not accurate, outputs. Very useful if you need bullshit quickly, almost entirely useless otherwise.
I was going to say "if you need bullshit quickly and cheaply", but the "cheaply" part is only a temporary thing: "AI" is going to get very expensive soon. Someone has to pay for the many billions invested in it, and it's not going to be the billionaires.
Isn't "banning VPNs", or "requiring VPNs to enforce age restrictions", just completely impossible?
Not that much different from trying to "ban mathematics".
Aha, perhaps that's why governments and mega-corporations are so keen on "AI" - a dumbed-down population with no ability to think for themselves would be much easier to control! (see icon).
So the big AI companies are not only charging you for generating bullshit (highly plausible but not necessarily correct) they're making the payment system obscure so that people don't realise they're being charged the huge real costs of generating this bullshit.
Isn't this just the world's biggest scam?
Seems a lot like "putting restrictions on the victims" instead of "persuading the unpleasant people not to attack".
Too much power is held by multi-national over-valued merchants of advertising (which is simply making people want what they don't need).
Luckily in my children's generation there is a distinct wish to ditch fancy digital stuff and get back to basics for a "real life". Film cameras, record players, outdoor activities, etc. I am sure that society will turn against AI and Advertising, people really don't like bullshit.
Ahem, there has been no such thing as "road tax" since 1 April 1937.
Vehicle Excise Duty is now "vehicle tax" or "car tax", and is a general tax - no longer hypothecated for road maintenance.
https://en.wikipedia.org/wiki/Motoring_taxation_in_the_United_Kingdom
Everyone seems to have forgotten that "AI" (in the current context - Large Language Models and similar multi-dimensional statistical models) is just a very-plausible-bullshit generator. It is excellent at producing bullshit, and very useful where bullshit is needed in large quantities. That may, or may not, be useful in running a business.
Probably because is_adult() requires the subtraction of DoB from the current time, which is a complicated operation including calendar oddities and timezones. The "adult" age might vary with country, and purpose of asking (e.g. "wants to marry", "wants to drive a car", "wants to drink alcohol").
dob_check() presumably just provides a date, which the requesting code can then use as it wants.
The trouble with "age" as a number is that it's only correct for a maximum of one year. After that it needs changing annually, forever, on the anniversary of the person's DoB. For which updating you need to know the person's DoB.
At least DoB is a constant for each person (except, perhaps "born again" Christians?).
I reckon the might be a statistical spike in births on 1 Jan 1970 if this ever becomes commonly used. That'll confuse the historians of the future!
It's all very well UK government departments issuing big fines to foreign companies, but will these companies pay the fines?
4chan told Ofcom to go away. Perhaps Reddit could too?
I can see a future where people who don't want their identities databased by Peter Thiel and Palantir will just use online services located in China, Russia, etc. Or just fire up a VPN or TOR.
Quite agree. An interview should not be trying to catch someone out, and it shouldn't be a rigid set of questions either. Start with a scenario and see where the conversation goes.
A good interview should discover what the candidate's thought processes are for solving some problem they haven't come across before. What experience do they have of similar problems? How deep is their understanding of the issues? Do they enjoy problem-solving of this kind? Are they willing to make mistakes and then learn from them? Can they communicate effectively in the problem domain? Would they fit in with the existing team?
This is all human judgement, something that "AI" cannot do, even if it ever does become "intelligent".
LLMs are not intelligent, they're statistical pattern matching systems that generate bullshit that may or may not be correct or true.
Here's an idea to make LOTS of money: invent a complicated but impressive bullshit generator, make $$$ from building and improving it, then make more $$$ from selling add-ons that make it more accurate, more correct, or less dangerous.
In general, the best solution to a problem is the simplest solution to the problem. The rush to use AI is just making everything more and more complicated and impossible to understand.
By controlling the money.
They already do: the cost of living crisis isn't unsolvable, the ultra-rich could just pay a bit more in taxes. It's just that the ultra-rich benefit from people having to work multiple jobs for minimal pay. They like having more money than they can spend, it makes them feel important.
Until a revolution against the elites happens...
Since generative AI is essentially bullshit[*], why does the government think that providing bullshit as answers to questions is a Good Idea?
[*] https://link.springer.com/article/10.1007/s10676-024-09775-5
Our Kia e-Niro has sounds when it's moving slower than 20 km/h.
The nice thing is that it has a switch to turn the sounds off, really good at night or in slow-moving traffic jams.
The illegal thing is that it has a "beep beep" sound when reversing, as well as the "engine" noise. The "beep beep" is illegal in the UK for a reversing car. But if you turn the sounds off it's illegal because it doesn't have the required "engine" noise.
FWIW the "beep beep" is incredibly annoying, we can hear another Kia with the same noise several houses down the road! So we turn it off as a matter of course when reversing out of our driveway.
