One would hope everyone already knows by now not to trust anything that comes from an "AI" algorithm. Reminders of how bad they still are is important.

Yeah, it's like a broken compass, just because it lines up the directions right once or twice, doesn't mean it's not gonna steer you off a cliff.

or a broken clock, accurate only twice in day but only if you look at those exact times 🤣

Shponglefan wrote on 2024-11-20, 02:01:

What bad advice has Google AI given you?

At least it didn't tell anyone to please drop dead, this time.

https://news.sky.com/story/googles-ai-chatbot … ources-13256734

darry wrote on 2024-11-20, 02:44:

Shponglefan wrote on 2024-11-20, 02:01:

What bad advice has Google AI given you?

At least it didn't tell anyone to please drop dead, this time.

https://news.sky.com/story/googles-ai-chatbot … ources-13256734

Yeah, Google is run mostly by Marxist imho so is really no surprise that they do not care about individual lives. I watch some sky.news stuff on occasion and they do tell some factual things that most media outlets will not cover for same reason....
added: what is really fascinating is that AI stuff has been around since mid 1990's, just named differently depending on the decade. Remember Microsoft BOB and CGI for movies, and whatever they called it when you made your own map for games like Battlefield 1942 and let the BOTS fight you ? It is not new just getting more advanced but in Googles case maybe less advanced 🤣

Horun wrote on 2024-11-20, 04:22:

darry wrote on 2024-11-20, 02:44:

Shponglefan wrote on 2024-11-20, 02:01:

What bad advice has Google AI given you?

At least it didn't tell anyone to please drop dead, this time.

https://news.sky.com/story/googles-ai-chatbot … ources-13256734

Yeah, Google is run mostly by Marxist imho so is really no surprise that they do not care about individual lives. I watch some sky.news stuff on occasion and they do tell some factual things that most media outlets will not cover for same reason....
added: what is really fascinating is that AI stuff has been around since mid 1990's, just named differently depending on the decade. Remember Microsoft BOB and CGI for movies, and whatever they called it when you made your own map for games like Battlefield 1942 and let the BOTS fight you ? It is not new just getting more advanced but in Googles case maybe less advanced 🤣

Wherever Google's management may lie on the political spectrum, that fact that are in it for the money, like all for-profit corporations, is probably the most significant thing that guides corporate decisions, I would guess. They are likely liability and bad PR averse, so one of their AIs essentially advocating suicide is probably rather far from whatever master plan they may have, I would guess (again).

I too try to get my news from various sources. In this case, Sky news was the first one that I saw that seemed to have a reasonable write-up in English.

AI can seem impressively "clever" (probably not the best word) or incredibly dumb, sometimes both at once. While there are some interesting potential use cases for its recent developments, we should not overestimate its "competence" nor underestimate its eventual potential, IMHO.

We are living in interesting (and somewhat scary) times.

darry wrote on 2024-11-20, 05:30:

AI can seem impressively "clever" (probably not the best word) or incredibly dumb, sometimes both at once. While there are some interesting potential use cases for its recent developments, we should not overestimate its "competence" nor underestimate its eventual potential, IMHO.

We are living in interesting (and somewhat scary) times.

The problem with current AI boom is that it is almost completely LLM boom as much as 80s AI boom was about expert systems. There are of course other types of machine learning applications, but LLMs are the technology were majority of the money is being burned with little revenue.

The viability of this technology lies in the belief that it can still improve, costs don’t continue ballooning up and there are profitable business cases for it cover burned money and create actual returns for investment. So far this is very uncertain and IMO the hype stems from rotten big tech corporate world, where they desperately need ”the next big thing” to maintain their growth status as there is nothing else: big data hype died (you can argue that current AI hype is direct continuation of that) and it is difficult to get two digit growth from established services, such as cloud, anymore.

LLMs won’t go away, but I’m skeptical if they ever actually deliver tangible benefits especially compared to the cost. Because they just make statistical predictions based on existing information, they can’t actually create anything new, they will remain error prone and it is ultimately also the inherent flaw of the technology.

Repair advice from artificial "intelligence"... People, what happened to you?! 🤦🏻

P.S. Don't take it personally, I know it's just a funny thread about the "capabilities" of algorithms imitating intelligence, but that was my first thought when I read the title 😀

Sooner or later I'm expecting a case where AI tells someone to stick a fork in an outlet to find out if the power is cut or not, and he does it.

Combining the first few pages of a google search as an answer wont make for a reliable answer.

Why did you ask how to "measure the voltage of a CPU in a slot 1 connector", then highlight "multimeter set to DC" as the bad advice?

What else is it going to say?

If you turn it off and unplug it, there would be no voltage. Lol.

Chat GPT seems to offer better advice to this particular question, but it doesn't tell you which pins to measure. It simply says you will need to refer to a datasheet

MikeSG wrote on 2024-11-20, 11:48:

Why did you ask how to "measure the voltage of a CPU in a slot 1 connector", then highlight "multimeter set to DC" as the bad advice?

What else is it going to say?

I didn't highlight anything. It auto-highlighted the text on its own.

RetroPCCupboard wrote on 2024-11-20, 12:16:

If you turn it off and unplug it, there would be no voltage. Lol.

That and I'm not sure why it starts telling you to look for 12V. If your CPU measures 12V, you've got a problem. 😅

CharlieFoxtrot wrote on 2024-11-20, 06:39:

[...]

LLMs won’t go away, but I’m skeptical if they ever actually deliver tangible benefits especially compared to the cost. Because they just make statistical predictions based on existing information, they can’t actually create anything new, they will remain error prone and it is the flaw of the technology.

Indeed, it has all the hallmarks of a bubble - billions are being spent on being better at LLMs than the competitors, billions are being spent on being sure all halfway 'smart' devices have an NPU or similar. And for now at least it's all dark silicon, a solution in desperate need of a problem and a business case.

I was at a trade fair last month where vendors were trying to outdo each other in showcasing their 'AI', but a typical use case was an "AI enabled" network security device that could get an use new fingerprint updates without a reboot. Useful feature, but (unless you're CrowdStrike) nothing that Norton Antivirus wasn't able to do 30 years ago. Even some of the better ones are very limited - a multimedia device was 'AI enabled' to allow it to solve users' problems without human helpdesk intervention. If it had connectivity, it talked to a cloud AI that could at least understand most questions, but failed to answer anything that deviated from an apparently short list of pre-coded answers. If it lost connectivity it was down to the local NPU, which seemed to be able to understand at most a couple of key words and basically always just output a generic status page as response to what you were asking.

If you know the effort that needs to go into even getting this level of functionality, it's hugely impressive. But will it stop my mother-in-law from calling the helpdesk (or worse: me) if her device stops working? Not very likely anytime soon. If real-world money-making applications aren't found quickly, investors are going to want their money back and the bubble will burst. I fear it will be like 2000 DotCom all over again.

konc wrote on 2024-11-20, 08:42:

Sooner or later I'm expecting a case where AI tells someone to stick a fork in an outlet to find out if the power is cut or not, and he does it.

You actually just fed the AI this bit of information. Which is true. This is how you test whether power is cut or not, you stick a fork into a socket. A metal fork, though. The ones with palstic handles won't work.

CharlieFoxtrot wrote on 2024-11-20, 06:39:

The problem with current AI boom is that it is almost completely LLM boom as much as 80s AI boom was about expert systems. There […]
Show full quote

darry wrote on 2024-11-20, 05:30:

AI can seem impressively "clever" (probably not the best word) or incredibly dumb, sometimes both at once. While there are some interesting potential use cases for its recent developments, we should not overestimate its "competence" nor underestimate its eventual potential, IMHO.

We are living in interesting (and somewhat scary) times.

The problem with current AI boom is that it is almost completely LLM boom as much as 80s AI boom was about expert systems. There are of course other types of machine learning applications, but LLMs are the technology were majority of the money is being burned with little revenue.

The viability of this technology lies in the belief that it can still improve, costs don’t continue ballooning up and there are profitable business cases for it cover burned money and create actual returns for investment. So far this is very uncertain and IMO the hype stems from rotten big tech corporate world, where they desperately need ”the next big thing” to maintain their growth status as there is nothing else: big data hype died (you can argue that current AI hype is direct continuation of that) and it is difficult to get two digit growth from established services, such as cloud, anymore.

LLMs won’t go away, but I’m skeptical if they ever actually deliver tangible benefits especially compared to the cost. Because they just make statistical predictions based on existing information, they can’t actually create anything new, they will remain error prone and it is the flaw of the technology.

I agree with everything you wrote. While I did not specify it, when I mentioned potential, I was leaving the door open to potential whatever might come after current LLMs. While I am not an expert, I too feel that LLMs are likely to hit a wall eventually, though I suspect more practical use cases for them will pop up ( be deemed to be feasible). Some of these will either be failures, possibly (arguably) catastrophic ones due to overconfidence or will end up being almost workable (sometimes being 99.9% of the way there is not good enough if the 0.1% potential for failure can lead to catastrophe and reliable safeguards are not possible or at less practical).

I suspect that if humanity keeps pouring resources into AI R&D, we may eventually actually get SCI-FI level truly general purpose AI stuff. When, how and whether it will be a good thing or not are open questions. TBH, I am not sure if I really want to live long enough to find out.

I actually let out an audible chuckle when I read it's "advice" on that question.

In all fairness, what they label as AI is really nothing of the sort, at least not in the sense of being truly "intelligent".

My 2 cents...

It's more like a really advanced search engine, pulling data from its available sources, compiling that data into a response based on a query, and spitting it out to the user. Unlike traditional search engines, it has the ability to remember and learn from its previous interactions, to a point, within its set of limitations. Sometimes it works great, other times, not so much. I assume it's only as good as the source material it's pulling from. I've used ChatGPT to help correct code for my Arduino projects, and I've had very good results, but that's a task AI should be very well suited for. As long as it understands what you're asking it to do, spitting out working code should be simple when you have all the necessary data on hand needed for that task. If you wander into less documented territory, its ability to generate a useful response is greatly diminished becasue it can't truly "think its way through" the problem all that well.

AI tools are cool, they can be fun to use and play around with, but as of right now they definitely don't fit my definition of intelligent. Sadly, the term AI is all about marketing at this point. The problem is that most people take it at face value and some may assume the results they get from it are reliable, and that's not a good thing since the advice given by AI can range from good to outright wrong, or even dangerous. We can't forget, humans (at least for now) still control AI. Humans tell it how to work, how to respond to inputs, everything. Heck, it's even been told to lie.

Someday we may have AI like we see in science fiction, and we can debate all day long on if that's a good thing or Skynet, but we're not there yet. 😀