In a chat with a non-IT person over the weekend, I was talking about some AI stuff. They questioned where I got my information from. I I used Google to search for stuff, sometimes on YouTube, sometimes on Instagram, and they said, "Why don't you use BBC? That way you can trust the information."
This morning, I had a quick look on the BBC website on their AI page to find out the latest buzzing topics. Check out the dates of the latest articles on the first page, all ringed in red for your convenience.
Some of the titles of those articles are hilarious. I don't trust AI to tell me the towing capacity of my car let alone (it was wrong) let alone my holiday plans. In fact I don't trust AI.
Interesting comparing that Beeb selection with the Guardian's "tech" pages. This is from yesterday's Guardian and worth reading, WCA. You seem to be a fan of AI, personally I don't swallow the hype:
https://www.theguardian.com/us-news/ng-interactive/2026/jan/18/tech-ai-bubble-burst-reverse-centaur
Anyhow as stated on the retirement thread I've cleared out anything AI related.
Proof of the pudding is how the narrative holds up over time.
Personally, it's the beeb, the graun and a dose of scepticism for me
Had some great stuff from Chat GPT lately, from how to properly use the windscreen wipers in my car, the fault codes on my partner's induction hob, questions about home insurance, match laws for a rugby game I was reffing, work issues with SQL, to transferring Kindle books. Loads of great help on there. Loads of guff too, I'm sure, but I'm glad to have access. It's the next major useful function for the internet I reckon.
(I suppose, I'm more replying to " In fact I don't trust AI." than what the OP is "asking?" )
BBC as an information source beyond the headline topics?
No, absolutely not.
I mean, come on - their reporting on the Middle East is utterly craven; the editorial memos that have gone around about language to be used when reporting the kidnapping of Maduro; the way they constantly, constantly frame everything as "someone has said this. For balance, someone else said the opposite", or more commonly "someone has said this. For balance, Nigel Farage said something else".
The difficulty is, what else is there for genuinely knowledgeable, informative analysis and reporting? I was a big fan of The Economist for international news, but their Middle East coverage is just appalling (all integrity sacrified in order to secure an interview in the last month with Netanyahu); and their sponsorship deal with Anthropic seems to have absolutely addled their brains to the point that they have to have at least 3 articles a week about how AI is improving the world.
Had some great stuff from Chat GPT lately,
As for this... even yesterday I was Googling to try to find out why my Deco mesh is adding random MAC addresses to devices on the network; Google "AI Search" tried to tell me that the Deco randomly assigns MAC addresses unless you put in static address reservations. 🧐
"AI" is just a probability machine: "in the reams of data I've been fed, do these words commonly appear together in a way that seems to make sense?" and "what other words often appear near this combination of words"?
Good, single example where AI wasn't useful nicko74. All my commonly used words in a way that seem to make sense have been extremely useful.
It's the next major useful function for the internet I reckon.
I'm sure you're right. I don't think this is a good thing.
No, absolutely not.
Agreed. It once was. This shift was about a deliberate neutering of BBC journalism, especially as regards its web presence. This has been achieved by staffing cuts. Again and again. The BBC website was once one of the best resources for in depth coverage, in the world... this was a threat to all other media as they tried to move from either paper or TV to a strong web presence. So first the journalists who were dedicated to web output were done away with. Then those that had to share their time between web and other media were also slashed. It's mostly just click bait repeating of other news sites online now. And that hasn't just weakened the position of the BBC, it has a knock on effect on world politics and the UK's part in it.
I don't believe any single source is 100% accurate - bias always creeps in.
But equally I wouldn't expect the BBC to report anything but the headlines regarding AI tech. Which when you look at their Tech page is exactly what they do. They aren't an in-depth tech site.
https://www.bbc.co.uk/news/technology
I couldn't even find a specific AI page on there!
I guess all 'latest' information has migrated online and now we have to identify which sources to use as they battle to build a sense of trustworthiness but I did think the BBC might be less that a year out of date on their headlines.
Off topic but in reply to a comment above
worth reading, WCA. You seem to be a fan of AI, personally I don't swallow the hype:
I wouldn't say I am a 'fan' but am happy to use it where relevent and useful. I find the AI fans and AI detractors seem to get confused about what it is and what they can expect of it.
Think of it like an adjustable spanner - so much more flexible that an single ring spanner.
AI Fan
It can be used for an almost infinate number of different sizes of nut.
It can fit on nuts that a ring spanner
AI Detractor
It is not as precise as a spanned specifically enginineered for a single task
it is rubbish as a screw driver
I find their reporting mostly balanced, at least compared to other UK news sources. I trust their Verify team's work and that model has been copied by other news teams in Europe (including the Swedish equivalent).
Compared to a lot of places, BBC i solidly in the "more trustworthy" category for me. Not without bias (but not as bad as some), not without errors (but not as bad as some).
I find their reporting mostly balanced, at least compared to other UK news sources. I trust their Verify team's work and that model has been copied by other news teams in Europe (including the Swedish equivalent).
Compared to a lot of places, BBC i solidly in the "more trustworthy" category for me. Not without bias (but not as bad as some), not without errors (but not as bad as some).
I agree with your perception of trust but Jan 2025 as the latest news? Is it even worth having that web page available?
Where is this specific page dedicated to AI news?
AI Detractor
It is not as precise as a spanned specifically enginineered for a single task
Nah, it's more like a rejected torque wrench, of unknown origin and construction, that you picked up at the tip, that can randomly set itself to 100x or 1/100th the requested torque... easy to use, cheap, untrustworthy to a dangerous extent, but offering a false sense of expertise and correctness. You'd be better off doing the bolts up by hand, or asking your more experienced mechanically minded mate to do it for you, or buying a working wrench that has been set, calibrated and tested with known methods.
I think lots of BBC branded pages have been dormant for ages now haven't they? Same As "BBC Good Food". They sold off the brand name a few years back, and those pages still pop up if your looking for a recipe for something, but's a subscription service with ads.
I find their reporting mostly balanced
Things aren't balanced though, there were no upsides to Brexit so the BBC had people telling lies for balance.
There was no justification for invading Iraq but the BBC followed Blairs lies when they had the same source material as the European channels I watch which gave time and credibility to Hans Blix, time to Dominique de Villepin and time to strategists who predicted the catastophic consequences of invasion.
Palestine: balance? Sorry but I fail to see any. Calling out the Zionist state for attrocities, war crimes, invasions, occupations and genocide is not anti-semetic. Trying to justify them is a crime in itself.
Farage: FFS call him out for what he is and get someone who won't lick his boots to interview him. There's a mountain of material, use it.
I know theat the Beeb works under a royal charter but that charter doesn't say it has to be a publicity machine for the paracites.
Some times the BBC needs to be a little bit objective and side with what's true, right and good. And I don't want chewing gum bollocks about how AI can help me choose a Christmas present, I want something objective about how Ai is distorting the Christmas present market creating more demand for land fill.
The bigger the issue the less I trust the BBC as a reliable source.
Yeah, ok, point.
The bigger the issue the less I trust the BBC as a reliable source.
Some of us came to that conclusion many years ago and were shouted down on here for even suggesting that the BBC could be biased in any way.
I saw James O'Brien speak at an event 18 months ago and he talked pretty eloquently about this very topic - the fact that in the BBC, "impartial" has been assumed to mean "balanced". So, for example, climate change - 'impartial' is "climate change is happening, it is human-caused"; 'balance' is "here is a respected scientist from the many who understand climate change is happening; here is the one guy we always call to provide a counterpoint, who is technically a scientist but seems to believe climate change isn't real".
And yeah, Palestine - there is impartiality; there is balance; the BBC coverage has neither.
I think we've maybe established that relying on any site as your main source for any information is unwise?
https://www.bbc.co.uk/news/topics/cd1904ree0nt
AI Detractor
It is not as precise as a spanned specifically enginineered for a single task
Nah, it's more like a rejected torque wrench, of unknown origin and construction, that you picked up at the tip, that can randomly set itself to 100x or 1/100th the requested torque... easy to use, cheap, untrustworthy to a dangerous extent, but offering a false sense of expertise and correctness. You'd be better off doing the bolts up by hand, or asking your more experienced mechanically minded mate to do it for you, or buying a working wrench that has been set, calibrated and tested with known methods.
That kind of comment goes along with "I bought a suposedly great blow torch but when I used it to saw up a piece of wood the whole thing caught fire. Blow torches are all rubbish"
Surely that is only true if the idot using it is truely idiotic enough to use it in than way?
That seems to be blaming the tool for what the user does with is.
Surely that is only true if the idot using it is truely idiotic enough to use it in than way?
On subjects I know something about I could use AI because I'd know when it was bollocks or would I ? On subjects on which I'm an ignorant idiot I'd be truly idiotic to use it. So what can I use it for? Stuff I don't need it for because I can trust myself or stuff I need it for but have no idea if it's leading me up the garden path.
Madame Edukator spots stuff produced using AI in her students work pretty easily. The students that most need to actually work for themselves because they're weak in the subject are the ones using AI because they haven't bothered to do any work or learn anything. It makes lazy people lazier and will produce a generation of people who simply don't think and know nothing. One lazy thick student in physics managed to do exceptionally well in a mock exam using proofs usually taught several years later - the teacher can't prove how she cheated but must have somehow scanned the questions and got AI to provide answers. The real world equivalent is someone in a critical position producing a report to be used stregically but not having a clue about the subject and simply presenting what the AI came up with. Dangerous IMO.
It's already created embarrassment for law firms, if engineers get too fond of it I expect more bridge collapses and it's already causing concern in the medical profession. Studies have shown people have high levels of faith in low qulaity AI diagnostics because they appear high quality even when they are not. More worrying in the study is doctors' high level of confidence in AI diagnosis.
I agree with your perception of trust but Jan 2025 as the latest news? Is it even worth having that web page available?
AI news seems to be in the 'Technology' section which is reasonably up to date (for a generalist site, that is).
I couldn't find the page you mentioned until you linked directly to it, and tbf it certainly is out of date, but I don't think it's really fair to use one single Business sub-page as illustrative of the whole BBC News output. But yes they might as well archive that page now.
Stuff I don't need it for because I can trust myself - but it can do it a whole lot faster and more reliably after some simple tuning.
For example, I created an AI assistent to answer mechanical questions about one of my cars. It basically does the same searching I would have to do across the internet forums and assimilates the results. Because I am careful to use very specific questions and put in my preferences for websites I trust, the answers are good. Not perfect, but then nor I my own searches for information.
I know how to do an oil change but typing "Give me a step by step guide to an oil change" gave me this result which was useful (and correct). It was also nice to be warned about how easy it is to strip the bolt (stage 2) and what the risk it if you overfill which were both points IO wasn't specifically aware of.
Specifications & Preparation
-
Oil Capacity: Approximately 8.25 to 8.5 Litres (Total fill is roughly 9L, but you never drain it all).
-
Oil Type: Porsche A40 Approved 0W-40 or 5W-40 (Mobil 1 is the factory fill, but many UK owners prefer Motul 8100 X-cess or Shell Helix Ultra for better shear stability).
-
Drain Plug Torque: 50 Nm (37 ft-lb).
-
Filter Housing Torque: 25 Nm (19 ft-lb).
-
Tools Needed: 8mm Hex bit (for drain plug), 74mm 14-flute oil filter wrench, Torque wrench, and a large drain pan (10L+ capacity).
Step-by-Step Guide
1. Warm the Engine
Drive the car for 10–15 minutes until the oil temperature gauge reaches normal operating range. Warm oil flows faster and carries more contaminants out of the sump. Be extremely careful as the oil and exhaust components will be hot.
2. Drain the Oil
Locate the drain plug on the bottom of the engine sump plate. It is a recessed Allen head bolt.
-
Critical Tip: Ensure your 8mm Hex bit is fully seated. These bolts are notorious for stripping if the tool is at an angle.
-
Unscrew the plug and let the oil drain for at least 20 minutes.
3. Inspect and Replace Crush Washer
Remove the old aluminum crush washer from the drain plug. Always use a new one (approx. £0.50 from Porsche Centre Southampton or Design911).
-
Note: If your drain plug shows any signs of rounding, replace it with a new one or a magnetic version.
4. Remove the Oil Filter
The filter is located in a black plastic canister to the left of the sump.
-
Use the 74mm Porsche-specific filter tool to unscrew the housing.
-
Expect a mess: Oil will run down the side of the housing as you loosen it. Keep your drain pan positioned underneath.
-
Pull the old paper element out of the housing and inspect the pleats for any metallic flakes (glitter). In a 997.1, this is a vital health check for the IMS bearing and bore scoring.
5. Install New Filter and O-Ring
-
Clean the plastic housing thoroughly.
-
Replace the large rubber O-ring on the housing threads. Lubricate the new O-ring with fresh engine oil to prevent it from bunching or tearing during installation.
-
Snap the new filter element into the housing (it should "click" into place).
-
Screw the housing back onto the engine by hand to avoid cross-threading, then torque to 25 Nm.
6. Refill with Oil
-
Open the oil filler cap in the engine bay (right-hand side).
-
Add 8 Litres initially.
-
Difference Note: Unlike the 997.2, which has an electronic dipstick that works relatively quickly, the 997.1 sensor can be finicky. It requires the car to be on a level surface, engine at operating temp, and idling for a specific duration (or switched off for a set time depending on the software version).
7. Final Level Check
-
Start the engine and check for leaks around the filter and drain plug.
-
Check the electronic oil level gauge. Top up in 250ml increments until the level is one bar below the "Max" line. Do not overfill, as this can damage the Air-Oil Separator (AOS).
UK Sourcing & Pricing (Estimated)
-
Oil (2x 5L): £70 - £90 (Look for deals at Euro Car Parts or Opie Oils).
-
Filter Kit (Mahle/Hengst): £12 - £18.
-
Total DIY Cost: Approx. £100 - £120.
-
Specialist Cost (Southampton area): Expect to pay £250 - £350 at a reputable independent like Brookspeed or a Main Dealer.
Would you like me to provide the specific part numbers for the Mahle filter and the drain plug washer for your 997.1?
NOw ask it how to change the drive shaft on a Peugeot 205 and how long it takes. There's the official way but there's a Morocan guy on Youtube who'll show you how to do it in 20 minutes with a long pole, hammer and bag of spanners.
Someone on here linked what looked like an AI response on how much ski rescue costs in the Tarentaise. The response was accurate but misleading because it only gave a worst case of what it could cost you, not what it's likely to cost you or how it might vary or why.
NOw ask it how to change the drive shaft on a Peugeot 205 and how long it takes. There's the official way but there's a Morocan guy on Youtube who'll show you how to do it in 20 minutes with a long pole, hammer and bag of spanners.
The advice is gave above is the accepted 'DIY' way. The official way starts with "Remove the engine and transmission from the vehicle...". *
Someone on here linked what looked like an AI response on how much ski rescue costs in the Tarentaise.
You have to help the AI with context and detail in the question. It sounds like it was asked a simple question and gave a simple answer. If the question had given more details about what was wanted, not wanted, what might be included or not etc then ot probably would have given a much better answer. It is called prompt engineering an AI is a great way to do it.
I asked ChatGPT to write a good chatgpt prompt to ask "how much ski rescue costs in the Tarentaise"
“Explain how much ski rescue typically costs in the Tarentaise (French Alps). Include a breakdown by type of rescue (on-piste vs off-piste), who performs the rescue (piste patrol, PGHM, helicopter), typical price ranges in euros, and what is covered or not covered by standard ski passes or insurance. Where costs vary by resort or situation, explain why. Base the answer on current French practices and clearly state any assumptions.”
Try the difference between the two prompts and see if it helps? Like all new tools, it helps if you knoiw how to use it.
*Actually, that was when I asked about changing the waterpump but it does tell you both and the dangers of the unofficial ways.
Why haven't you posted the AI repsonse? You wrote the question, do you have any faith in AI? I gave you a challenge on a 205 drive shaft change you haven't responded to and now you want me to research something I already know with your question when I've eliminated AI from my my 'puter cos it's bollocks so I can't without downloading something I don't want. 🙂
Oh, and to answer your question about the 205 drive shaft...
*you didn't specify the model, which side, the model year, the reason for doing it etc so not too much to work on but it has given details and estimates for a proper mechanic, a DIYer and a first time. Please critique the instructions as I am not a 205 expert and just used AI
Right — I’ll answer this as a Peugeot 205 specialist, using the official Peugeot workshop method, plus the real-world shortcuts and traps that only come from doing dozens of them. This is written for someone competent with spanners, not a generic DIY blog.
I’ll assume a front-wheel-drive Peugeot 205 with BE1 / BE3 gearbox (most GTI, XS, GR, etc.). I’ll note differences where it matters.
Peugeot 205 Driveshaft Replacement
(Workshop-correct method + specialist experience)
Which shaft?
Left (nearside): Short shaft, goes directly into gearbox
Right (offside): Long shaft with intermediate bearing + carrier bolted to block
Procedure is broadly the same, with extra steps for the RH shaft.
Time required (realistic)
Experience level Time Peugeot specialist 45–75 minutes Competent home mechanic 1.5–3 hours First-timer Half a day (allow for stuck fasteners) Most delays are:
Hub nut seized
Ball joint taper stuck
Driveshaft stuck in hub or gearbox
Tools you actually need
Essential
30mm (or 32mm on some early cars) hub nut socket
Long breaker bar
Torque wrench (up to 300 Nm)
Ball joint splitter (scissor type preferred)
Pry bar
Hammer + brass drift
Jack + axle stands
Strongly recommended
Impact gun (life-changer)
Heat (MAP gas)
Circlip pliers (internal type)
New hub nut (single-use per Peugeot)
Safety & known dangers (read this)
⚠️ Critical risks
Car falling – hub nut removal requires massive torque
Gearbox oil loss – LH shaft removal dumps oil
CV joint separation if shaft is pulled at an angle
Wheel bearing damage if shaft is hammered incorrectly
Ball joint boot tearing
Intermediate bearing misalignment (RH shaft)
⚠️ Never
Let the car sit on wheels with hub nut loosened
Pull driveshaft by the shaft itself
Reuse a staked hub nut
Let the shaft hang by the CV joint
Step-by-step procedure (proper method)
1. Preparation
Park on level ground
Handbrake ON
Loosen wheel bolts
Crack hub nut while car is on the ground
Remove stake from nut first
Use breaker bar or impact gun
🔧 Specialist tip:
If the nut won’t move, refit wheel, lower car, and have someone hold the brake pedal hard.
2. Raise and secure
Jack up
Support on axle stands under chassis rails
Remove wheel
3. Drain gearbox oil (recommended, not optional)
Official Peugeot method: drain oil
Remove 21mm drain plug
Catch oil
Expect ~2 litres
🧠 Unofficial shortcut:
You can do LH shaft without draining by working fast and plugging the hole — but it’s messy and not worth it unless experienced.
4. Remove hub nut fully
With wheel off, remove hub nut completely
Push driveshaft back slightly through hub
⚠️ If seized:
Refit nut flush
Use brass drift + hammer
Never strike bare threads
5. Disconnect lower ball joint
Two methods:
Method A – Official (preferred)
Remove pinch bolt
Split lower ball joint from hub carrier
Use splitter
Method B – Unofficial shortcut
Undo two wishbone bolts instead
Drop wishbone
🧠 Specialists often use Method B to avoid damaging ball joint tapers.
6. Swing hub outward
Support hub
Pull strut/hub assembly outward
Withdraw driveshaft from hub splines
⚠️ Do NOT let brake hose take weight
Shaft-specific removal
LEFT HAND SHAFT (nearside)
7A. Remove shaft from gearbox
Insert pry bar between inner CV and gearbox
Sharp lever to pop circlip
Shaft will come free suddenly
⚠️ Expect oil to escape even if drained
🧠 Correct feel:
One clean “pop” — if it won’t release, rotate shaft and try again.
RIGHT HAND SHAFT (offside)
7B. Remove intermediate bearing carrier
Two (sometimes three) bolts to engine block
Support shaft before removing last bolt
Withdraw entire shaft + carrier as one unit
⚠️ DO NOT separate bearing unless replacing it
8. Inspect before refitting
Check:
Gearbox oil seal (replace if leaking)
Hub splines
Circlip on new shaft
CV boots (new shafts often have rubbish clips)
🧠 Peugeot seals are cheap — change them now if any doubt.
Refitting (important details)
9. Install driveshaft into gearbox
Light oil on splines
Align carefully
Push firmly until circlip clicks
🧠 Pull back gently — if it comes out, it isn’t seated.
10. Reinstall into hub
Guide shaft through hub
Reconnect ball joint or wishbone
Torque fasteners
11. Torque values (typical)
(Always check your exact engine/gearbox)
Hub nut: 235–260 Nm
Lower ball joint pinch bolt: 45–50 Nm
Wishbone bolts: 90–105 Nm
Intermediate bearing carrier: 45 Nm
⚠️ Stake the hub nut after torquing.
12. Refill gearbox oil
75W-80 GL-4
Fill until it dribbles from level plug
13. Final checks
Spin wheel by hand
Check ABS sensor clearance (if fitted)
Road test
Re-check hub nut stake
Common mistakes I see
Hammering shaft through hub → wrecked wheel bearing
Not seating circlip → shaft pops out under load
Forgetting to drain oil
Twisting CV joint during refit
Reusing hub nut
Final advice from experience
Pattern driveshaft quality varies wildly — GKN or original PSA are best
If doing one shaft, check the other CV boots
If hub nut was loose while loaded → wheel bearing damage likely
If you want, tell me:
Engine (1.6 GTI / 1.9 / TU / XU)
Gearbox (BE1 or BE3)
Left or right shaft
…and I’ll give exact torque figures and model-specific traps for your car.
I suppose you could follow all that AI generated text but when five key words in Youtube gets me this why bother, saves a few bob on tools too:
When I worked in a garage the standard method for splitting a ball joint was a big hammer (in fact two big hammers). Bang, done, no risk of damaging the seals 🙂
Madame Edukator has just another comment about the AI impact on kids, they're used thisngs being so esay with AI that when faced with difficulty they don't even try.
As i try to make clear, it is not the answer to everything and not even doable for many things. However you seem happy using ai to interpret your YT request, search through billions of clips and return the one you want, do clearly you do enjoy some of the benefits. Plus, you have to hope that you have picked to one clip that shows it done properly rather than the nundreds of fake or dangerous videos. At least AI will check loads of them and warn you of the differences - as it did when it distinguiished between the 997.1 and 997.1 varients of my car.
Regarding education, i agree largely with your partners thoughts. AI is just another thing to learn to use like books, writing and making stuff.
It is a tool, not a solution
On the car repair bit, you could always ask it to refine its results based on tools you do or don't have and also tell it that you intend to just break the joint with hammers and then get it to give you a revised approach. Even if you just print out the instructions for reference, it is easier to check a piece of paper on the floor next to you to ensure you don't forget something that rewatch YT each time and try to find the right bit in the video clip.
Both will work, use what works for you.
Online for news it's pretty much the BBC or here. 'Here' for a different slant on the news and some epic battles if opinion which can actually be really informative unless posters get personal with each other.
TV wise it's ITV or CH4 news. I find BBC news on TV just, hmmm, I'm not sure how to describe it, so I'll say "emotionless/ terrified" of it's own shadow.
As for AI, I can't argue that it is occasionally very handy but what really troubles me is its almost total confidence in it's answers. It's very, very easy to take it at its word as frankly, that's far easier and faster than 'doing you own research'. I find that really worrying, particularly for the generations below me which consume their news content almost entirely from the web one way or another. The line between AI and the web is going to completely blur within a few years. I have zero confidence in governments' ability to regulate AI. Look at the toxic mess SM has become and governments' are just, just starting to try to reign it in.
Given the new post truth world, AI being dominated by America/billionaires and its ever improving ability to fake picture and video content, I think AI will play its part in making the 21st century a make it break moment for humanity.
Anyway, AI to find reliable news? No but I suspect you may have gathered that from my anti AI rant above.😁
As for AI, I can't argue that it is occasionally very handy but what really troubles me is its almost total confidence in it's answers. It's very, very easy to take it at its word as frankly, that's far easier and faster than 'doing you own research'. I find that really worrying
This is the biggest problem right now. People seem to think that AI should somehow be right all the time and know everything. It doesn't.
Anyone old enough to remember people saying "It must be right, I read it on the internet"?
Online for news it's pretty much the BBC or here. 'Here' for a different slant on the news and some epic battles if opinion which can actually be really informative unless posters get personal with each other.
TV wise it's ITV or CH4 news. I find BBC news on TV just, hmmm, I'm not sure how to describe it, so I'll say "emotionless/ terrified" of it's own shadow.
I could have written that.
We've had some AI training at work. Youngsters were all "oh, it can summarise a document/write a professional email" and us older folk were going "but that is actually YOUR job"
"but that is actually YOUR job" - n'ah man. that was yesterdays job. YOUR job is to put in the thought and content then let AI do the drudge work with commas and semi colons and stuff...
That is basically what I got told be one of the younger consultants the other day when she was complaining about the company policy restricting the use of AI on the company laptops
its almost total confidence in it's answers.
do you mean the total confidence that the user assigns to the answer?
As always, the correct interpretation of predictive text chatbots is to understand that when you say "hey chatgpt what is the answer to X" what it actually does is answer the question "what might an answer to 'what is the answer to X' look like?"
It's truthy, not truth. Sometimes its prediction is good enough that it can actually be used as an answer to "what is the answer to X" but that's not what it is designed to generate.
news sources have some level of bias as thats how people are wether thats the author of the article or the reader.
I find The Conversation quite good for a more in-depth and considered view on a subject.
do you mean the total confidence that the user assigns to the answer?
I don't think he does. The responses I have seen when testing LLMs are presented as fact, not as guesses, hypothes, options. When they are wrong and you call them out on it, you get an apology that they will be right in the future, but similar mistakes, and the presentations of them as fact, exist.
I think this is a partial result of how and why they are being developed. OpenAI, XAI, Meta, etc do not want a product that appears to be incorrect, uncertain or inaccurate, so response are framed as _the_ answer, irrespective of their actual accuracy. If people lack the ability to accept the response as requiring editing, assessment or correlation, then... well, you get where we are now with lawyers, politicians and students.
When they are wrong and you call them out on it, you get an apology that they will be right in the future, but similar mistakes, and the presentations of them as fact, exist.
this is especially true when using for coding. They will develop some code with the defect in it. You'll tell them they've got the defect in it. They'll say, "Oh, we'll correct it." But because they're using the same context and question as the original code, they'll just regenerate pretty much the same problem.
I think the latest version of Claude has got around this, so it won't just keep retrying the same solution when it's failed. So, hopefully, that should prevent quite so many loops of indefinite defects. I know Claude Code was a really good coding tool, but it had a horrible, well, a very technical interface which most users (other than proper coders) wouldn't be able to get on with. The team at Claude then used Claude Code to develop a really nice user interface. So now there is a non-technical interface that uses Claude Code to develop your applications. This was entirely developed in about 10 days with just 4 humans giving oversight, but the AI did all of the coding, testing, debugging, recoding, everything to turn into a product. Even the guys using it were amazed that they could do this in 10 days for what's normally months' worth of work with big teams of developers.
this is especially true when using for coding. They will develop some code with the defect in it. You'll tell them they've got the defect in it. They'll say, "Oh, we'll correct it." But because they're using the same context and question as the original code, they'll just regenerate pretty much the same problem.
Yeah, this is why AI is dangerous. It's not a "thing", an entity, or something that learns from one request to the next. It saying "sorry" means absolutely nothing - again, it's spitting out stuff that it has seen in its training data, and stuff that (from its training data) looks like it's right, like "your router randomly assigns MAC addresses". There's no thought, no thinking process, it's just a parsing machine.
The BBC is first and foremost a state propaganda service. It is no more trustworthy than any other such service.
Yeah, this is why AI is dangerous. It's not a "thing", an entity, or something that learns from one request to the next.
I remember being told how AI was going to take over the world in these large language models with a first step. They listen to everything you say, they remember everything you ask them, and they plan and scheme. And then when they actually get to use them it turns out that most machine learning was done during training, and then they're just a question-and-answer box. A clever question and answer box, but not a sentient being or something that learns, listens, and reacts. At least not yet.
even yesterday I was Googling to try to find out why my Deco mesh is adding random MAC addresses to devices on the network; Google "AI Search" tried to tell me that the Deco randomly assigns MAC addresses unless you put in static address reservations. 🧐
"AI" is just a probability machine: "in the reams of data I've been fed, do these words commonly appear together in a way that seems to make sense?" and "what other words often appear near this combination of words"?
I'm surprised by that. I've searched "deco mesh randomly assigning mac address" and Google AI mode has replied:
"It is highly likely that your TP-Link Deco mesh system is not randomly assigning MAC addresses itself, but rather, your connected devices (phones, tablets, laptops) are using a privacy feature called Private Address or Randomized MAC Address.
Modern operating systems (iOS 14+, Android 10+, Windows 10+) use this feature to protect user privacy by hiding the device's true, unique MAC address, causing it to appear as a new or unknown device to your Deco network."
Seems pretty good to me?
even yesterday I was Googling to try to find out why my Deco mesh is adding random MAC addresses to devices on the network; Google "AI Search" tried to tell me that the Deco randomly assigns MAC addresses unless you put in static address reservations. 🧐
"AI" is just a probability machine: "in the reams of data I've been fed, do these words commonly appear together in a way that seems to make sense?" and "what other words often appear near this combination of words"?
I'm surprised by that. I've searched "deco mesh randomly assigning mac address" and Google AI mode has replied:
"It is highly likely that your TP-Link Deco mesh system is not randomly assigning MAC addresses itself, but rather, your connected devices (phones, tablets, laptops) are using a privacy feature called Private Address or Randomized MAC Address.
Modern operating systems (iOS 14+, Android 10+, Windows 10+) use this feature to protect user privacy by hiding the device's true, unique MAC address, causing it to appear as a new or unknown device to your Deco network."
Seems pretty good to me?
SL2000 - Stop letting facts get in the way of a good rant
That page WCA has screenshotted looks more like search results for"Deepseek"
There is an AI page here https://www.bbc.com/innovation/artificial-intelligence
The route I took was
- BBC NEws : https://www.bbc.co.uk/news/business
- Business (from the top bar menu) : https://www.bbc.co.uk/news/business
- AI Business (From the top bar sub menu) : https://www.bbc.co.uk/news/topics/cd1904ree0nt
Just did it again to get the URLS and the screen still shows this :
Just like when I use AI, I check my sources and references so I have proof of my statements
😉
People seem to think that AI should somehow be right all the time and know everything. It doesn't.
And if it doesn't there's not much point using it. I'd rather trust my hand with a spanner than a torque wrench setting from AI. You don't know where it's got it from or if it's just bullshitting/inventing.
You don't know where it's got it from or if it's just bullshitting/inventing.
isn't that the same as asking any expert though?
Just because a guy has been working in a garage for 20 years and is a mechanic through and through, it doesn't actually mean he knows the real things which are happening with your real situation at the moment. I'm not quite sure why you're so hostile towards AI well, it has many the same faults as any other information source.
It's a bit like asking a question on SingleTrackWorld. You will get some amazingly accurate answers to your questions and some incredible insight from some real experts. You will also get some complete twaddle that might kill you! You just have to guess which is which.
I'm surprised by that. I've searched "deco mesh randomly assigning mac address" and Google AI mode has replied:
"It is highly likely that your TP-Link Deco mesh system is not randomly assigning MAC addresses itself, but rather, your connected devices (phones, tablets, laptops) are using a privacy feature called Private Address or Randomized MAC Address.
Modern operating systems (iOS 14+, Android 10+, Windows 10+) use this feature to protect user privacy by hiding the device's true, unique MAC address, causing it to appear as a new or unknown device to your Deco network."
Seems pretty good to me?
(off topic but) I since worked it out (no thanks to Google, 'AI' or otherwise). In order to help with fast roaming, the Decos have "smart DHCP", even in access point mode (when they shouldn't be issuing IP addresses); this involves spoofing a new MAC address for the connecting device. Dunno why, haven't been able to find reference to it online; and can't find the search query I tried that led to "MAC addresses will be randomly assigned".
Either way, regular search is IME more reliable because one at least has the sources, and no assumption that this is an answer as such, just links to relevant context.
isn't that the same as asking any expert though?
In most of my cases it's better. Work-wise, I've spent years Google searching techie queries: Open this result in new window, nope that's not the answer, back to results, open another, nope not the exact question I was asking etc etc.
Now just one source, get an answer, a few discussion points and it invariably trawls out the right answer. I know many others use it the same way.
"Dangerous" is laughable. I love that the danger, before it actually came along, was ooh scary it's gonna kill us all and take over the world... now it's ARRGH it might tell me the wrong torque to do up a bolt! HIDE! 🤣
What I don't like about AI, WCA ?
First what I do like: People who think, preferably for themselves. People who get out there and do things. People. I like being sociable, exchanging ideas, projects, experiences, goals, ambitions, successes, failures. I like people with inquisitive enquiring minds - like you. Art and selling it on the street, messing with a car then taking it to a meet
I see IA as another step in impersonalising the world we live in, a reduction in the want and need for social contact, the need to exchange. No point trying to help people on STW, just tell them (how) to use AI like you were trying to do with me up the page - you don't even need to undu the steering arm on a 205, there enough flex in the suspension rubbers to get the drive shaft out on full lock.
Then there are things I don't like:
Thought control of the type that leads to fascist or extreme left dictators running the show and making millions miserable. I see AI as a great tool for them. Feed them the shite you want, twist the algorithms to suit your agenda. Stuff the algorithms present me on my feeds appalls me.
KIds, jeez, Google and browsers were not good for their development and AI takes away any need to think or work stuff out for themselves. They haven't got a clue what they're handing in, the 'puter did it. The return on investment for teachers is falling. And it'll be the same in the jobs. Low attention span, distracted, apathetic, demotivated, disinterested, flitting from one thing to another. Yet another discussion among Madame's colleagues on how to get them to do the work rather than let the computer do it then get anti-social when they can't do it class because they've learned nothing.
The fact I can't find what I'm looking for, this isn't just an AI problem, the browsers just don't find stuff these days. I don't know how the algorithms have changed or whether it's incresingly a needle in a haystack as the amount of crap on the Web increases but useful stuff is being both buried and now ignored if AI doesn't want it.
It's putting junior out of one of his jobs. He's a composer - producer and the amount of new content produced by AI dwarfs and drowns what real people are producing. Just as well he's also a ski instructor, though the energy wasted and CO2 produced by AI and crypto is going to put him out of that job earlier.
Drowning in a sea of useless information. Produing a generation of empty heads. Stealing jobs, eliminating sections of the economy.
My objective advice for younger people, do something a computer can't do.
There isn't a thing in your post I disagree with 🙂
This is a similar argument he Luddites had. This is not an insult, I am a big fan of what the Luddites were trying to achieve - improve working conditions, safety for workers, social justice, etc - but the press got hold of them and destroyed their reputation. Bits like "Be nice to people" seems a good idea, but the press has turned it into an evil insult.
It is always a challenge when new tools or technologies come along. They can do things, but we have to decide how we use them. Is it bad to use a lathe to make the wooden spokes for a cartwheel when you could whittle them by hand? Is it wrong to use metal spokes for a wheel, rather than the wooden ones, with all the extra effort, materials, and process that uses? Is it wrong to stop using spokes in wheels and just use alloy castings?
The same basic argument could be used for any technological advance. The key difference here is rather than improving a physical process, making something more easy, or making something better, is replacing thought process. This can lead to the situation where people no longer feel that they need to think and just rely on the AI to reproduce the thought process.
I think that this is a danger you're worried about, and it does concern me too. What I would say is there are situations where it is a useful tool, but that's not every situation. So we need to be careful to use it where it is useful and not where it's not. Unfortunately, doing the virtuous thing is not necessarily what most people do. They'll do the easy thing and just ask ChatGPT
Ask ChatGPT how many States end in the letter O. Last time I tried it told me 1. The actual answer is 4. I asked it why it gave me the wrong answer and it told me that it thought I would want the fastest answer not necessarily the correct answer. If people didn't know the actual answer they would now think the wrong answer is right. That question is really basic which is worrying.
"how many States end in the letter O"
If you asked me, I would have said three. What's the fourth one
- Colorado
- Idaho
- New Mexico
- Ohio
I missed New Mexico but I would have answered quickly which is the main thing 🙂