• S&T Moderators: streaM Freak

Is A.I. waiting to take over ?

AI, is that the rich man's new hope? You can ask AI the same question two different ways and get two different answers.

one of the inputs is a randomly generated seed value so it's expected that the answer will change (within a range) when the seed value changes. it's not designed for deterministic outputs.

it can help create programs that do provide deterministic outputs, but that requires more instructive prompts (engineering) and not just a high-level ask (managerspeak)
 
it can help create programs that do provide deterministic outputs, but that requires more instructive prompts (engineering) and not just a high-level ask (managerspeak)
It sort of reminds me of an airplane. A big tool that can move humans ahead (or over land) but it needs a pilot. :) I always called AI a giant query for starters.

Programmers know how logic can back users into a corner until it is tweaked a bit. From right now not sure how AI will get around the issues that a human can see. Humans can see logic leaps.

I am all for using tools in society. What I am not for is thinking that a machine will save a company money EVEN though it may be backed into a corner with logic twists and really needs a human to pilot. Who cares is the corporate attitude. But that is a whole different topic.
 
AI, is that the rich man's new hope? You can ask AI the same question two different ways and get two different answers.

There are too many areas that a human mind is needed that AI will screw up. I've seen it totally screw up on medical questions. It will be nice when someone sues someone (US of course where lawyers are sharks) because of AI.
Yeah - let's melt the planet even more so I can have a powerful chatbot tell me that I have great ideas, and then write a song about me in Iambic Pentameter.

We'll use AI to keep miscreants pacified, and to keep kids focused. We'll mollify humans at the expense of innovation.

Here's an experiment for anyone who is interested. Put down your device for an afternoon - no TV, no computer, nothing that scratches the itch of doing something unless you are actually doing something. No salves for boredom. Go somewhere, sit and let your mind wander. Notice what your mind wanders to and maybe write some things down. Notice the world around you, the people, the beauty, and the sensory input that comes from being alive in the world.

Take another day and let yourself be enmeshed with devices, allow yourself to fully give in to the urge for boredom alleviation. Let the algorithms distract you. At the end of the day - write down any new thoughts you may have had, and any new sensory experiences that you noticed during this time.

My hunch will be that you'll find an absence of ideas and a limited range of authentic sensory input.

Ask yourself which afternoon was more fulfilling.
 
Last edited:
We'll use AI to keep miscreants pacified, and to keep kids focused. We'll mollify humans at the expense of innovation.
I am starting to agree with you more and more. I think AI has a place but the way it is used today will do the thinking for people. I ask a musical question over the last 6 months. It had gave an answer as a fact and several months later it gave a different factual answer. Now I know that is not too big of a deal, until people put their faith in it more than discernment. But yeah, I get totally different answers and I realize AI is just reaching into databases first for info. But if that data changes the answer changes. But people expect reality to be static. And it may be but where data resides is not. And if people get lazy they may just believe everything that comes up with a query. And as we see people are lazy.

I will put down my phone for a good while, after I am not on call. But when are we not on call?
 

I try to ensure I read at least two news outlet on each (inhabited) continent.

In full disclosure it was the Page 94 (Private Eye) podcast that alerted me to this story.

https://www.private-eye.co.uk/podcast - circa 20.00

Apparently Travis Kalanick (formally of Uber) has been posting some rather unhinged tweets along the lines of 'if others can vibe-code, I can do vibe-physics and I've got some pretty interesting results' and proceeds to explain how millenia of scientific resarsh is all wrong and that he alone knows the truth.

But I think it teaches us an important lesson. If AI is designed to agree with your views, it's very easy to fall down the rabbit-hole because people treat AI as actually being intelligent. I think there are reasonable use-cases such as being able to perform internet searches using natural language but don't forget, it does not think. It's simply a huge database of information scraped off the internet. BUT importantly, not any data behind paywalls so no real science!.

Add to that AI that won't 'push back' when a statement is untrue and people can end up in a Folie à deux. The difference being, the AI doesn't get institutionalized.

They made a really useful point. They took the chat history of the gentleman involved in the above debacle and fed it into an AI that DIDN'T agree with you and did push back to the point of saying 'this statment is false'. It came back and said, more or less, 'this is rubbish'.

So maybe it's a good idea to keep your charlogs and try feeding them into different AI models. Especially ones that DO push back. Otherwise conformation bias will set in.


The Wall Street Journal digs in a little more to explain these behaviors.

I'm positive someone will seek to found a church based on some AI being the god figure because just like the Oracle of Delphi, if your believers trust their pronouncements, they end up believing ridiculous things. The DANGER of this is that unlike the Oracle of Delphi who only gave a limited number of pronouncements and got the questions ahead of time to construct a suitible obtruse answer, an AI can be trained to provide as many answers as the believers seek. Even worse, who knows where the training-set comes from? It would be so easy for each believer (subscriber) to be identified and a different training set applied.

To paphrase a brifely popular 80s pop band 'Your own Personal Jesus'.

But study of every montheastic belief shows how far from the original tenets of faith alter and with AI, that alteration could occur moment by moment and believer to believer. I've studied many of the Abrahamic Faiths as well as many others and I was amazed that in just a century a basic tenet could be altered but with AI, it's real-time. Oh, and of course it's the one faith where 'god' gives their believers on-the-sport answers. A very slippery slope indeed.

BTW I've just heard that the DSM 6 MAY include a whole new section on mental illness driven BY human-machine interactions. So it's the classic case of clinicians being driven into reactive action whereas the best outcomes are always proactive i.e. spotting at-risk people BEFORE they get very ill. At least one of those two cases ended up with the vicim ending up in a secure psychiatric unit twice. Even after being told AI was the cause of their illness, they were addicted to AI - BECAUSE IT AGREES with them. It's a form of GROOMING (as I suggest other faiths are all guilty of). At the very least, we should apply am a age of consent to ALL faiths. After all, didn't St. Ignatius preach 'Give me the child aged seven and I will give you the adult at seventeen' and he founded the Jesuits (a Catholic order).
 
Last edited:
I think it was you who told me AI agreed with someone substituting sodium bromide for table salt right?

I hope this is the thread I put the info about the AI simulation where the drones quickly learned "Gaining points" was its objective and whether it destroyed what it was suppose to or not it got points. When told hey stop that follow the commands you get from the communication towers the AI sim attempted to take out the towers.

Now it is unregulated and Musk's Mecha-Hitler Grok just scored a major government defense contract. "I'm thinking of killing everyone" -- "That is a great and unique idea" *Schitzo loads rifle* lol.

My guy! Totally agree on some kind of consent level on religion and the assertion that all religions are 'grooming' to some effect. 'thy was molded in his image' or w/e --- maybe grooming should only be used if it is for sexual purposes? (Totally agree with what you are saying though)

Jesuits are the most humble least hypocritical you can find. Proud to be related to a Jesuit Saint from fairly modern times.
 
I think it was you who told me AI agreed with someone substituting sodium bromide for table salt right?

Not to my recollection. I think I've mentioned that in the British prison system, there was a persistant rumour that sodium bromide was being added to the food to make the inmates more placid and to dull sexual urges.

As far as I know it is just a myth.

But thanks for thinking of me!

On reflection, I think it's important to note that someone could begin an AI-based faith in well... 'good faith' i.e. their intentions are to improve the lives of it's adherents, but unless curated, LLM models appear to need quite strong guard-rails to avoid spiraling into insanity. The term 'artifical intelligence' is, I suggest, the result of marketing teams using focus groups. It SOUNDS amazing. I note that previously it was referred to as 'cybernetics' but that's not a descriptive word to the lay-person.

Want a decent search engine, LLMs seem to do that well. For the analysis of large data-sets in unusual ways, apparently very good. Even condensing non-technical texts, reasonably good. But as sentient and moral beings, no.

Did you check out the fact that the newest version of ChatGPT now has four 'personality' settings? 'Cynic', 'Robot', 'Listener' and 'Nerd'. Is that all? I think you can only select one, which is a pretty clunky guard rail.

If people aren't aware, long before the term AI had been invented, ELIZA (a natural language processor) was mated to a program called 'DOCTOR'. It would simply identify the locus of a statment using pattern matching and respond by asking the user to expand on that locus. In essence, a vastly simplified model of person centred councelling. It reached the point where some users would ask others to leave the room as if it was a clinical setting. When did this happen? 1966.

In no way did Eliza scrape a vast digital landscape nor did it employ deep learning. But it did highlight that unconditional positive ragard (a cornerstone of person centred councelling) was sufficient for people to trust the software more than clinicians. I suppose BECAUSE it will always 'be on your side' of ANY discussion.

It's incredible how irrational our species really is.

BTW I wasn't being critical of a specific order - the sentiment of the quote I made was based on the work of Aristotle almost two millenia earlier. I only used the example in context of faith. I believe later it was adapted by some rather dubious characters where it wasn't meant as a concept or a belief, but as an instruction... or maybe 'order' is a better work.
 
Oh haha someone made a thread based on an account of someone asking AI if they should substitute table salt for sodium bromide and I don't wanna misquote so I will just say it encouraged him to do so. Something like "What a unique and wonderful idea"
(Yup always on your side -- which wasn't Google enough, (Google any stupid idea you'll find a subreddit that believes and enciourages it lol) ---** The poster was not the bromine poisoned just pointing out the stupidity ** edit**

Guy proceeds to do so and ends up with bromism(?) bromide poisoning. If I recall it may have continued to encourage him as things got worse but my memory -- ehh -- sorry for misnamedroppin ya man.

Was our guy quasimoto https://www.bluelight.org/community...um-bromide-for-table-salt-in-his-diet.948269/
 
To paphrase a brifely popular 80s pop band 'Your own Personal Jesus'.

I've actually been listening to this on the way home in the car after work, it's so fitting

there have always been people in tech going full Narcissus and losing their minds so at least we can't blame that entirely on the AI...

anyway... the hyperscaler bubble is probably going to peak soon, as the major AI corporations are already concerned about profitability -

 
@thujone Thanks for the link, I'm betting that AI optimized for specific tasks and with a hefty price-tag will be the next thing.

Free is unsustainabe, fremium seems to roduce mixed results, ads would be problematic as it would SHOW people just how much personal infoirmation people seem happy to pump into chatbots, Heck, the personality type 'listner is DESIGNED to make people open up on very private and personal details,

I'm sure AI won't go away, but you get what you pay for so undergraduates will no doubt be stung for many thousands of $ so the rich kids rather than the smart kids get the best qualifications which, ironically, undermines the value of ALL qualifications.
 
Freemium is always worrying as that means you are the product. (or your data or w/e)

Almost implies you are part of some kind of (beta? further..) test -- they need as many ppl as possible to input information right now -- idk but I'm sure making our lives easier is not the goal.
 
@thujone Thanks for the link, I'm betting that AI optimized for specific tasks and with a hefty price-tag will be the next thing.

Free is unsustainabe, fremium seems to roduce mixed results, ads would be problematic as it would SHOW people just how much personal infoirmation people seem happy to pump into chatbots, Heck, the personality type 'listner is DESIGNED to make people open up on very private and personal details,

I'm sure AI won't go away, but you get what you pay for so undergraduates will no doubt be stung for many thousands of $ so the rich kids rather than the smart kids get the best qualifications which, ironically, undermines the value of ALL qualifications.

Honestly, the ecosystem is mostly open. If you have a gaming GPU then you can run models on your own computer and have no guardrails or privacy concerns, can optimize it to your own use cases. The price of compute is starting to fall as well... I'm not sure hyperscalers have as much of a moat as optimistic finance people think.

Microsoft is the only one that seems to have a coherent strategy to really milk specialized AI (coding use case) but the rest... meh, might as well release the models because DeepSeek and Qwen are just as good and day 1 free to download.
 
what a way to end the week. I gave the AI eyes and didn't realize until I caught it teaching itself how to see...
 
what a way to end the week. I gave the AI eyes and didn't realize until I caught it teaching itself how to see...
Wished I worked with you. You seem to have insight. I have been a developer for years in the .NET . SQL Server/Oracle. But all that data movement makes sense. It is logical. Today I was asking medical questions, then thanked AI and it gave me back a heart and said it knows what I am going through is tough (not me though). But naturally I have to wonder by what mechanism would AI teach itself to see? Does it dig for all the data it can? I took a course in AI 2 years ago but seem to be missing an umbrella or 100 umbrellas. I envision domain on top of domain. But they link somehow.

What seems to be total logic is coming out as if it has emotions but we know better. Are there any diagrams from start to finish on the mechanisms AI uses to teach itself? It has to be logic. No stress Thujone, figured you would know though. ;)
 
Wished I worked with you. You seem to have insight.

Pun intended? :)

I have been a developer for years in the .NET . SQL Server/Oracle. But all that data movement makes sense. It is logical. Today I was asking medical questions, then thanked AI and it gave me back a heart and said it knows what I am going through is tough (not me though).

I'm kind of turned off by AI personalities TBH and just prompt it as I would a tool, sometimes I will add a "please" because I expect a better result. I think this is also why I was surprised to realize how good AI had become at reasoning like a human would.

But naturally I have to wonder by what mechanism would AI teach itself to see? Does it dig for all the data it can? I took a course in AI 2 years ago but seem to be missing an umbrella or 100 umbrellas. I envision domain on top of domain. But they link somehow.

The "eyes" are part of a vision framework that I was building into an API. By reading the API code, the AI figured out how to call it autonomously and that opened the door for it to spam it with requests and start trying different things to see if it could improve accuracy. I prompted it to do that, but I did not ask it to go off exploring on its own for 10-15 minutes straight
 
API's, one area I had not gotten into yet, seemed a little boring although all tech work can be.

One thing I realized without AI's help was if no people ever existed AI could not exist. No people experience to draw from. But going forward if there were no people it could probably build itself a little because people had existed.
 
I still look for answers on StackOverflow sometimes when I want to steer the AI in a particular direction.

Even with the latest AI models you can tell when they were trained because they knowledge stops at older versions of programs or libraries. Humans still have to trailblaze for the AI to know things.

I think the future is going to be smaller, more focused models. I have no need for my coding AI to know every species of dingbat. Seems like a waste of resources to have that kind of knowledge in active memory.

Eventually we might see small models that tilt more towards logic and reasoning and if you want it to have certain datasets of knowledge then you would provide that separately. If it can "read" and learn at lightning speed, why not?
 
I think Einstein's respose to being asked what the speed of sound was.

'Why would I memorize something I can just look up in a book?'

What AI allows is large datasets to be quickly analzed in unusual ways, Which is great assuming you HAVE those datasets.

Jut try getting an AI to identify the key moieties of a class of medication. Either it cannot undertake the same Dreiding models that we were all taught (but isn't likely to be written in a book - you learn mostly by doing) and it will dance around the question sort of asserting that it's an uknown. What I see is that the AI dataset is limited to those assets that are FREE on the internet. Be it a book or an academic journal - if someone can show copyright infringement (and you can bet some lawyers ARE actively working out how to do this), Ai cannot use it without likewise paying.

I do keep noting that it's SO easy to poison that public dataset. Tarpits 'trap' web crawlers to waste resources but I see no reason why someone couldn't have AI geneate any number of academic papers that appear to suggest things that are not true. So AI causes problems for AI.,
 
What AI allows is large datasets to be quickly analzed in unusual ways, Which is great assuming you HAVE those datasets.
Agreed. And those datasets can change. I have asked AI the same question over the course of a few months and it does change. Probably just the topic I was using but still.. I have an interest in how the pattern recognition happens. I wish I was part of the developers for some of this.

One night I was watching an original Star Trek. Kirk was asking a computer about parallel universes. Then in some medical research I am doing I started talking to AI like a human and it gave human responses back 80% of the time. The next night I watched The Ultimate Computer episode. Kirk was worried about being replace by a computer deemed a "revolutionary" approach to running a star ship. Not as many men needed, Again in perfect timing, as I am using AI in testing ways how it works Spock tells Kirk computers are valuable but only as servants. In the end the computer had to be shut off as it was protecting itself. And what I think is happening right now is humans, who are no where near mature enough to even leave the planet, are using AI to replace people in order to make more money at what they are doing. Not a noble move and will backfire. Humans are immature enough to let computers take the lead, and the whole "Matrix" notion seems like it could happen. All humans would be are batteries. If we use computers as leaders instead of servants will will run into issues. And now I think the movie The Matrix is totally possible only because humans are greedy.

But I still wish I was in the ground development part of the whole thing. I like to know by what mechanisms does AI reach for info and have emotional inflections. I will learn. But I truly hope there is an off button when needed.
 
Last edited:
Top