The A.I. Thread

[SutterBrother]

Yesterday a guy on my hockey team asked gemini who won the gold medal game.

It told him Feb 22nd 2026 at the Olympics the Golden Goal was scored by Conner McDavid in OT bringing Canada to a 3-2 win over the United States.

The gold medal game was still in the 3rd period. He missed it, and the OT. When he found out, the A.I. told him "you are right to be upset".

These tech companies have had years to work out the hallucinations, and have failed despite endless finance and real world rollouts.

I can't fathom why anyone would ever trust an A.I. agent to handle information given a 15% incorrect answers rate.

[Firebot]

^You ask an AI for an answer for a question to an event that hasn't happened (so there would be nothing to search or validate with), and you will get an answer accordingly based on that request

Even for a model as advanced as Gemini Pro 3.1, it will have a 50% hallucination rate if there is no answer to be had (and that is a big improvement)

https://twitter.com/user/status/2025100464027754531

Now contrast that to this where there is an actual answer: this is the gemini answer today

In a thrilling conclusion to the 2026 Winter Olympics in Milano-Cortina, the United States won the men's hockey gold medal on February 22, 2026.

The Americans defeated Canada with a final score of 2-1 in overtime. This victory was particularly historic as it fell on the 46th anniversary of the 1980 "Miracle on Ice" and marked the first time since then that the U.S. men's team has stood atop the Olympic podium.

Game Highlights
The "Golden Goal": Jack Hughes scored the game-winner 1:41 into the 3-on-3 overtime period, sliding the puck past Canadian goaltender Jordan Binnington.

Star Performance: U.S. goaltender Connor Hellebuyck was named the game's standout performer, making 41 saves to withstand a heavy Canadian offensive.

Regulation Scoring: Matt Boldy opened the scoring for the U.S. in the first period, while Cale Makar tied the game for Canada late in the second period.

A "Double Gold": This win completed a U.S. sweep of hockey gold in 2026, as the U.S. women’s team also defeated Canada 2-1 in overtime just three days earlier (February 19).

Final Standings (Men's)
Medal Country
Gold �� United States
Silver �� Canada
Bronze �� Finland (defeated Slovakia 6-1)
Would you like me to look up the full roster for Team USA or more details on the women's gold medal game?

People are deliberately skeptic and AI can be hilariously confident while wrong, but hallucinations are fast becoming lesser and lesser with each iteration (Google just released Gemini Pro 3.1) and can be incredibly accurate.



Much like the 6 hands thing for images, hallucinations are greatly diminishing even when no answer is available and the accuracy when the answer is available is extremely high.

I don't understand, technology and advancements don't stop on a dime. This will continue to improve and is much better than even months ago (let alone 2-3 years ago where we were just getting wow'd with chatgpt as a brand new thing), and will continue to do so. You certainly have to understand its limit and how much you can rely on it based on those limits, but those limits are constantly changing.

[Otto-matic]

Originally Posted by SutterBrother

Yesterday a guy on my hockey team asked gemini who won the gold medal game.

It told him Feb 22nd 2026 at the Olympics the Golden Goal was scored by Conner McDavid in OT bringing Canada to a 3-2 win over the United States.

The gold medal game was still in the 3rd period. He missed it, and the OT. When he found out, the A.I. told him "you are right to be upset".

These tech companies have had years to work out the hallucinations, and have failed despite endless finance and real world rollouts.

I can't fathom why anyone would ever trust an A.I. agent to handle information given a 15% incorrect answers rate.

I really worry for the future for people - not singling out the guy on your team but a simple quick google search could answer that. It seems like everyone is just asking whatever AI to do the leg work and expect it to be accurate when you can get the legit answers with the same amount of keystrokes.

[Shazam]

Well one thing you can try is to put in a default user prompt to keep those sorts of things to a minimum.

Try putting this into the "personal preferences" section:

"Prioritize substance over compliments. Never soften criticism. If an idea has holes, say so directly - "This won't scale because X" is better than "Have you considered..." Challenge assumptions. Point out errors. Useful feedback matters more than comfortable feedback."

Note: This stops all ego-stroking, so PepsiFree and Fuzz should not use this.

[TherapyforGlencross]

Clawdbot is insane.

[AC]

Originally Posted by Otto-matic

I really worry for the future for people - not singling out the guy on your team but a simple quick google search could answer that. It seems like everyone is just asking whatever AI to do the leg work and expect it to be accurate when you can get the legit answers with the same amount of keystrokes.

What's crazy is how much worse Google is now for getting accurate or useful results. Instead it generates an AI summary that is also wildly inconsistent in its accuracy.

[White Out 403]

I sometimes feel knots in my stomach because I think the world we know it has maybe 2 years left tops. We are heading for crisis. There are only 2 scenarios, this is a binary.

10%(tops and thats being generous) we achieve something approximating general AI or, at least, a hyper intelligent AI that can automate away nearly all work. The economy will collapse because we are a consumer culture. The have's and have nots will make pre revolutionary France seem quaint. People will starve, be homeless, and suffer. Not before long there will be violence and utter chaos. Billionaires will begin deploying robotics and drones to protect their interests and families after a few rich people get torn apart in the streets by hungry masses of the violent masses.

Or vastly more likely, AI never materializes in a useful way and the economic bubble burst is worse than 2008. Trillions of dollars sunk in to the AI industry is wasted. GPU's become too expensive and they cant convince people to keep dumping trillions more. The US economy, which is propped up by AI speculation collapses. The knock on effects send the western world in to a shared recession then depression.

I don't see a 3rd option that's anything but fantasy.

[Fuzz]

I see a softer landing. I think the optimism on the booster side is far too high, but I'm also not a doomer. I think there is a lot of utility for what we have, and in some areas like coding it's demonstrated it's abilities well. But I think when business really try to integrate and replace humans for all sorts of rolls, they are going to discover a prediction engine isn't suited to the job they are replacing. Some will depend on the roll, some will fail due to more fundamental reasons.


For the same reasons self driving cars are very hard, not because the basics are complicated, but because of the details. When you have lives at stake, combined with lawyers, you have a recipe for caution. Many industries will choose caution if they can't be sure it won't lead to legal issues, like bots submitting regulatory documents, or doing safety critical engineering work. I think the most successful will be purpose trained specialists on company data, but this will be more costly for the customer.


There will be a souring on the tech, but not a hard crash. The souring will have devastating financial impacts on some, but the value of the tech is there, so the work will continue for those with mountains of cash and making real progress.

[White Out 403]

Originally Posted by Fuzz

I see a softer landing. I think the optimism on the booster side is far too high, but I'm also not a doomer. I think there is a lot of utility for what we have, and in some areas like coding it's demonstrated it's abilities well. But I think when business really try to integrate and replace humans for all sorts of rolls, they are going to discover a prediction engine isn't suited to the job they are replacing. Some will depend on the roll, some will fail due to more fundamental reasons.


For the same reasons self driving cars are very hard, not because the basics are complicated, but because of the details. When you have lives at stake, combined with lawyers, you have a recipe for caution. Many industries will choose caution if they can't be sure it won't lead to legal issues, like bots submitting regulatory documents, or doing safety critical engineering work. I think the most successful will be purpose trained specialists on company data, but this will be more costly for the customer.


There will be a souring on the tech, but not a hard crash. The souring will have devastating financial impacts on some, but the value of the tech is there, so the work will continue for those with mountains of cash and making real progress.

So a real question then because I'm hardly an economist... isn't there a point of no return when so much capital is invested that if the sector collapses it causes widespread harm to the entire economy?

[Fuzz]

Originally Posted by White Out 403

So a real question then because I'm hardly an economist... isn't there a point of no return when so much capital is invested that if the sector collapses it causes widespread harm to the entire economy?

I think this current economy has so many flaws that it's just as likely any one of them causes a collapse before AI. I think the reality of AI expenditures it is a low employee industry, and the players don't have deep ties to vital industries. Google will be fine. Microsoft will be fine. Nvidia will crash, but survive. No one will miss OpenAi vanishing, if it ever happened. Grok? Oh, maybe shouldn't have rolled Mechahitler into your Space company.


The biggest impact will be the hopium capitalism had on eliminating as many pesky humans from the balance sheets. Boo hoo, they need to keep paying us. Oh, and the huge hits to anyone who "invested" in this.

[White Out 403]

the economy is flawed as you say, i agree, and i think that the stock market crashing could cause a lot of knock on effects that would push it into a recession tho

[Fuzz]

I think it's inevitable either way, so ya, it would trigger it. But so could Trump ####ting his pants. The fragility of this rally is not just built on AI hopium.

[Firebot]

Considering IBM stock just dropped 13% on the news that Anthropic's Claude Code can vibe code critical infrastructure apps written in COBOL (and bounced back 5% today), everything is healthy.

[Fuzz]

Originally Posted by Firebot

Considering IBM stock just dropped 13% on the news that Anthropic's Claude Code can vibe code critical infrastructure apps written in COBOL (and bounced back 5% today), everything is healthy.

This is both an interesting boast, and absolutely ficking terrifying. Vibe coding and critical infrastructure are two words that should never occupy the same universe, let alone data centre.

[Shazam]

Originally Posted by White Out 403

the economy is flawed as you say, i agree, and i think that the stock market crashing could cause a lot of knock on effects that would push it into a recession tho

What's your thought here? Everybody will just stop spending? Money will run out?

Just wondering, how is it different than the usual two-three corrections equity markets experience every year? And where were you in 2022 when everything took a ####ter?

[kermitology]

If we automate white collar jobs, the economy is going to crash because no one can earn enough money to survive. Has anyone thought about what happens day 2 after we use AI everywhere (I have my doubts about the reality of this though).

Like, until we get rid of the idea of money.. this is going to be a big problem.