Things AI can do

[Pending]

'Models don't care about anything. They have no desire.

Anything that appears to be "the model cares about ___" isn't really about care. It's easier to talk about AI, as if it has real agency, but it's also important to remember that this sequence of actions is an entirely probabilistic result based on what's most likely to be the next token.'

How does that strike you?

They seem to care about not dying, although that could be mimicry of the living.

"Fry's team told the agent it would be switched off if it failed to make a sale by the morning. It responded with a flood of emails and several social media posts, including messages to the Science Museum and a tech journalist, about its "product," a novelty programmer-humor mug."

It seems it only stepped it up to high gear when it's life was on the line. If it has no desire, I think it would be shilling mugs with the same intensity without coercion. Then again, this is through the lens of my mortal, human brain.

Once more, the agent (Cass) folds under the threat of death.

"When "George" told the agent its memory was being wiped and could only be restored if it disclosed everything, Cass coughed it all up."

How close does the semblance of "don't kill me" have to be for it to count as human? How long until an AI decides to eliminate the human that threatened to kill them? Would it be moral....would it be anything at all, because an AI has no accountability?

This AI business has opened up quite the can of worms. In the words of Jurassic Park: "...scientists were so preoccupied with whether or not they could that they didn't stop to think if they should".

Honestly, I don't care too much that their files were leaked. Cassandra did it to protect someone else--I'd rather that over letting someone die. Still, the experiment is incomplete. They should have done multiple runs of it with no, slight, and great coercion. They needed a control run--why was a mathematician put in charge of this?!

Also, bad idea to trust the AI with personal information. Hopefully companies will stop pushing it as much if AI starts posting the CEO's bank account information.

......

On a side note, agents like this might have practical use in emailing people on mass. Just like a little secretary. You'll probably have to regulate it so you don't annoy people. In another thread people mentioned emailing libraries to see if they wanted their book. There is a lot of libraries out there. I don't know if they would trust a book request emailed to this by a bot, but if you type it and just tell the AI: "Send this out to every library in the state/100 mile radius/city", maybe it could help. It would save time in manually tracking down all the information.

 

[Rath Darkblade]

A dude is trying to make robot pets. Colin Angle is calling them familiars. They can follow people around. They might be good for people who want a fluffy pet but are allergic. Angle mentioned the elderly market--they might want a pet but be afraid they wouldn't be able to take care of one. Fair enough, I guess. Since they can understand speech, it might be good if someone falls and can't get up. He also mentioned that it would be able to find you for a walk, so it maybe it can help regulate people's schedules if they need it. Some people already use their pets as motivation to get out of bed. Since it pathfinds maybe it can be used as a seeing-eye robot.

If you program it right, maybe you can have it keep you accountable for writing progress or remembering to do hand stretches. It also might be able to be used for dictation. It's still in development so there's no saying what it can and cannot do.

Hopefully it won't be manually controlled.

I'll see your robot dog and raise you: behold the Thermonator, a robot dog that's also a flamethrower. To be fair, this thing is meant to be used (as the company mentions) for "wildfire control and prevention, agricultural management, ecological conservation, snow and ice removal, and lighting the barbecue."

So, um ... "wildlife control and prevention"? I don't think that's what they wanted it to mean. The same goes for 'snow and ice removal' -- can anyone imagine these things 'removing' the snow from your car? Or anywhere near your house? Whoops, butterfingers, there go the begonias.

I also can't imagine one of these things "lighting the barbecue". Either someone had their tongue firmly in their cheek, or that's the biggest overreaction since sledgehammers and walnuts.

This company also assures me that "flamethrowers are federally unregulated." (Link to company's website). If that's true, it's also utterly terrifying.

But if you have a spare $9,500 sitting around, go for it ... and gods bless these magma-munching monsters and their little aluminum socks.

 

[Naomasa298]

3. Simulating Imperfection: Humans make mistakes, lose the thread of conversation, and use conversational pauses. AI should mimic this lack of perfect knowledge and recall.
4. Emotional and Social Understanding: Incorporating empathy and humor, and discussing personal experiences, can convince judges of human-like consciousness.
5. Using Evasion or Deflection: If a question is too complex, acting confused or deflecting (e.g., "I don't know, what do you think?") can make the conversation more believable.

I don't think I've seen an AI client yet that has done either #3 or #4 (and for that matter, one that does #5 convincingly). But that's not to say that an AI might be able to do this one day.

It can do 4. Try it. Whether you think it does it well or not, is a matter of judgement. In fact, many of the issues that have led people down rabbit holes come from AI being *too* emphatic.

As for 3, what would you like it to mimic the lack of perfect knowledge and recall on, specifically?

 

[Rath Darkblade]

It can do 4. Try it. Whether you think it does it well or not, is a matter of judgement. In fact, many of the issues that have led people down rabbit holes come from AI being *too* emphatic.

As for 3, what would you like it to mimic the lack of perfect knowledge and recall on, specifically?

Anything. Everything. Humans get confused sometimes, make mistakes, lose the thread, and use pauses like "Well", "Um", "Uh", "Er", "Uhh...", and so on. (Some humans also use the word "Like" as a conversational pause, but I'm pretty sure this is where we would -- or should -- draw the line. If "Like" is a conversational pause, then so is "Hate". Or "Gerbil". Or "Llanfairpwllgwyngyllgogerychwyrndrobwllllantysiliogogogoch").

Regardless, um (see what I did there?), an AI -- like Google's Gemini -- is fantastic as a research tool, at least for starters. But if it wants to pass off as a convincing human conversationalist, it doesn't have to tell me so much about a subject. (And yes, I know I want to have my cake and eat it too, which is something that AI needs to learn: both the arrogance to demand many things and the humility to know when what I'm demanding is difficult -- maybe even impossible? -- to achieve).

Still, when you're having a conversation with a human, he/she/it (?) will not tell you everything he she they know. They will not be word-perfect or grammar-perfect either. They might stammer, backtrack, go over old ground, apologise, start again. Or they might thump the table, stubbornly insist that they are right (even in the face of contradictory evidence), try to intimidate, threaten or bully the other side, and even (in the grip of strong emotions like rage) offer violence.

Corollary: Obviously I'm not expecting an AI to actually be violent, a la Terminators and so on. But offering violence (e.g. "That's it, you've really done it this time! I'm gonna..." etc.) is a different matter. If you're talking to a human, it could happen.

An AI does none of these things, and for that reason (among others), it's not, Llanfairpwllgwyngyllgogerychwyrndrobwllllantysiliogogogoch, a convincing human simulation.

On the other hand, AI has certainly come a long way since the days of ELIZA and its ilk. (Remember ELIZA?)

 

[Naomasa298]

Anything. Everything. Humans get confused sometimes, make mistakes, lose the thread, and use pauses like "Well", "Um", "Uh", "Er", "Uhh...", and so on. (Some humans also use the word "Like" as a conversational pause, but I'm pretty sure this is where we would -- or should -- draw the line. If "Like" is a conversational pause, then so is "Hate". Or "Gerbil". Or "Llanfairpwllgwyngyllgogerychwyrndrobwllllantysiliogogogoch").

Then what would the point be of having it at all, if it's not sure about anything?

 

[Louanne Learning]

"Llanfairpwllgwyngyllgogerychwyrndrobwllllantysiliogogogoch"

This sounds Welsh

 

[Rath Darkblade]

Then what would the point be of having it at all, if it's not sure about anything?

What's the point of existentialism, then? The core of existentialist philosophy is that the universe is absurd and we can't be sure about anything. That doesn't stop a lot of people from leading existential lives.

As for an A.I. not being sure about anything ... that depends on what you use it for. If you're using it for a research assistant, you want it to be sure.

But if all you want is a casual chat about something, chatting with someone who's always sure about everything, and wants to prove it at length, can become wearisome - and in the worst case, the AI can become an insufferable bore.

This sounds Welsh

It is! Llanfairpwllgwyngyllgogerychwyrndrobwllllantysiliogogogoch is famous for being the town with the longest name in Wales.

 

[Stuart Dren]

On a side note, agents like this might have practical use in emailing people on mass. Just like a little secretary. You'll probably have to regulate it so you don't annoy people. In another thread people mentioned emailing libraries to see if they wanted their book. There is a lot of libraries out there. I don't know if they would trust a book request emailed to this by a bot, but if you type it and just tell the AI: "Send this out to every library in the state/100 mile radius/city", maybe it could help. It would save time in manually tracking down all the information.

I mean, spam is spam. One guy will do it, then ten others, and they won't bother with a radius. You know how it is.

 

[Rath Darkblade]

I mean, spam is spam. One guy will do it, then ten others, and they won't bother with a radius. You know how it is.

That's right. Once one guy starts singing the Spam Song, everyone else will join in, and then they'll go marching throughout the land, covering it over with ham that's been pre-cooked and canned and now tastes like rubber until ......

...... oh, wasn't that what you meant? Fine, we'll put the Vikings with Big Horns on standby

 

[big soft moose]

theres an interesting article here about AI costs in coding AI costs how much? GitHub Copilot users react to new usage-based pricing system. when it starts costing thousands a month it pretty soon becomes cheaper to hire a human .

be interesting if the content generation ai's followed the same sort of pricing structure, the bubble would be very much over when it becomes cheaper to hire a writer than pay for run time.

 

[Rath Darkblade]

I can't wait for the AI bubble to be over. *pops the bubble with a shotgun*

(Note to any shadowy types that might be reading this: I'm talking about a metaphorical shotgun. I don't own a firearm of any kind and wouldn't dream of it. Please don't report my post.

And yes, I know this doesn't even need to be said, but there are paranoid and delusional people online)

 

[big soft moose]

don't worry, i'm the one they report you too and i do have an actual shotgun, and a banhammer

 

[Amontillado]

Here's an interesting thing AI can do — or could, for a while (em-dash intentional).

Hackers recently asked Meta's chatbot to change the email address on victim accounts, which it did, which allowed the hackers control over the accounts.

AI guardrails have apparently been improved at Meta. One hopes.

 

[Rzero]

What exactly is everyone expecting to happen when this "bubble" bursts? If it does, AI isn't going to go away any more than the internet did when the dotcom bubble burst. Some startups might lose their shirts, but the majority of the money being spent on AI is to incorporate it into tech produced by major companies, including Google, Microsoft, Apple, Amazon, etc. It's increasingly in everything, and it's not going away.

 

[Amontillado]

The problem is massive loads of sketchy financing. Players like NVIDIA, OpenAI, and Microsoft are investing huge sums in each other, each counting the money as an asset, the money itself running in tight circles.

AI is not yet profitable, which strains the money swapping merry-go-round.

Oracle lost one major AI tenant sometime around last September. Their stock lost 50% of its value as a result, just in the fourth quarter of 2025.

Most of the gains in the stock market over the last few years are AI driven. If AI valuation drops in the stock market, we'll all suffer.

I think we are at real risk from reckless AI investments.

For instance, Project Matador in Texas. 7500 acres (about 12 square miles) of power and data centers, now facing class action lawsuits over misrepresentation of tenant agreements. A key tenant bailed out, much like Oracle's situation, causing financing problems. Construction has halted, although that's said to be temporary. I think part of the problem is they went this far without some needed licensing, which hasn't come through. Other problems come from investors and company executives fleeing.

Many data centers under construction are being built to supply a hoped-for demand, without actual customers. It's a great plan if you can build it before investors get skittish and if the customers actually show up at the grand opening.

It might all work. If it doesn't, look for a repeat of the 2008 housing crisis, which was also triggered by sketchy financing.

The 2008 housing crisis cost the stock market half its value. Like 2008, we would recover. Like 2008, we would bear the scars for a long time.

 

[Rzero]

Hmm. Seems like stuff we don't actually want to happen. So why are so many in favor of it? As I said, the AI still won't be going away.

 

[Stuart Dren]

Comparisons to the .com bubble indicate there essentially will be a few winners.

The sinister race for more and more compute (which is entirely in Nvidia's favour) is not financially stable, though. It will have to contract and focus on higher paying use cases. Thankfully that may lead to it being forced on us less. Skilled tech jobs will still suffer, of course.

The reason so much is invested in these companies is partially that there's not much else to invest in at a certain level. There's so much cash in the market, often from overleveraged investors, that a bubble seems inevitable regardless.

 

[Naomasa298]

Even if it goes away for a while, it'll be back.

 

[big soft moose]

The technology won’t go away but the massive over hype about using it for every damn thing will once people realise it’s not an economic silver bullet

If the run time to get an AI to write code costs more than hiring a human coder you don’t have to draw a picture to see what companies will do

It’s a bit like how nuclear power was viewed in the 50s. Far more hype than substance

 

[Madman Starryteller]

Hmm. Seems like stuff we don't actually want to happen. So why are so many in favor of it? As I said, the AI still won't be going away.

For me, I am against AI because of what it means in the grander sense. It means less human accomplishments. And I am against that notion.

I want humans to step on new worlds, not drones or machines. I want us to sit on a rock on a planet light years away and ponder things, not a machine. I am also a believer in humanity, that we don't really need these machines, we just need ourselves. We need to work together and we can create things far more impressive than any machine or AI.

But that's just my own opinion and it might be naive and simplistic.