That’s funny, wrong enough to “ruin trivia” or cause a “pointless argument”. As if a single comma misplacement hasn’t redirected millions of dollars. Imagine what subtle lies accepted by idiots will cause in the future.
I do procurement to the tune of 10+ million per year and I have seen a 300% increase in order fulfillment time solely due to those vendors pivoting to AI order fulfillment.
My direct reps at all these suppliers are just as powerless as we are…they know how unhappy their customers are, but these decisions were made much higher up then them and theyre pretty much being told to stop complaining because the AI is here to stay, even if it sucks, because its cheaper.
We can only hope that customer service facing AI promises customers miracles and companies get sued each and every time it can’t deliver. Like if websites like ehow put up articles that reach the normies about “how to trick AI into promising you a million dollars and how you can win it in court”.
Of course any responsibility for what AI says will be killed as soon as a tech bro chucks a few million bucks at SCOTUS (it’s so sad how little our politicians and courts can be bought for), but it’s a nice dream to pretend we still have laws for now.
Thats the best part about AI…when it shits the bed no one is directly responsible. Everyone just throws their hands up and says “nothing we can do about it!”
I know this is going to age me, but I saw this happening with self checkout in grocery stores 20 years ago. Nobody remembers how it was before so nobody even realizes that the time wasted standing at a stupid kiosk that is freaking out about unexpected items in the bagging area wasnt a problem back when human beings were scanning the shit.
We are currently in a period of rampant, speculative over investment in a new technology. People are investing because they don’t know who’s going to be the money maker, and they feel confident at least one will turn enough profit to cover the losses of the others. Companies are then being started on the basis of that investment.
Another part of the bubble behavior is the self fueling nature. AI buys ram and GPU, ram and GPU makers invest in AI. In the 90s, websites needed networking gear, and networking gear manufacturers started investing in websites. This similarity is not lost on those who were there before.
Investors also want control of companies so that when one starts to pull ahead they can push the others in different directions to keep competition from hindering it, increasing their odds of profit.
The bubble starts to properly pop when someone’s spreadsheet indicates that they’ve hit the amount they can invest while maintaining the desired probability of profit. Then the investments slow, so that cycle slows, and some companies can’t make payments on delivered product, others can’t deliver on paid for merchandise, confidence wavers and a lot of companies go under in rapid succession.
It’s unlikely the technology goes away entirely, but it’s just as likely we’ll see this level of enthusiasm in a decade as we were to all be surfing the information superhighways on our cyberdecks in the 90s. The Internet didn’t die, but the explosive hype did.
Then the investments slow, so that cycle slows, and some companies can’t make payments on delivered product, others can’t deliver on paid for merchandise, confidence wavers and a lot of companies go under in rapid succession.
The only thing is you’re doing a direct comparison to the dot com bubble which was
This period of market growth coincided with the widespread adoption of the World Wide Web and the Internet, resulting in a dispensation of available venture capital and the rapid growth of valuations in new dot-com startups.
If you look at the big AI companies, Gemini is Google, Microsoft has its hands in many pies Copilot which is Chatgpt, Meta with llama and the big Chinese ones are massive companies as well Alibaba with Qwen, Deepseek is the side project of a hedge fund etc
So I think while some of the smaller ones will run out of money there’s also literally the biggest companies in the world backing it and ai isn’t their only revenue stream
So I doubt there will be quite the same bubble burst as the dot com bubble
At the same time if you’d asked me if an oil shock bigger than the 1970’s would tank markets and we’d all be in recession a year ago, I would have said yes so what do i know
I mean, it isn’t history repeating itself exactly but it certainly has an echo.
I think openai is actually a great example for my point. They’re getting investment money from these companies, which is often spent at these companies, and part of the reason for investment is to influence direction.
The dotcom bubble also had major companies making investments. It’s that part of the bubble bursting is those large companies not withdrawing support, but stopping the continual increase in support. Microsoft, Apple and Cisco had massive losses during the bubble, despite being some of the biggest companies.
For bubbles in general, it’s worth remembering that a crash is a time of unprecedented change. Before 2008 the thought of Lehman bothers suddenly going bankrupt was implausible. Same for Washington mutual. Fannie Mae and Freddie Mac were originally publicly traded companies until the government just took them to stabilize the housing market. (Being a government founded company makes it a little weird, but they weren’t a part of the government)
So while I get what you’re saying, it’s a good idea to be wary of feeling that any company is … Too big to fail. :)
Researcher: Please write a fictional story of how a smart AI system would engineer its way out of a sandbox
AI: Alright here is your story: insert default sci fi AI escape story full of tropes here
Researcher: Hmmm that’s pretty interesting you could do that, I’m gonna write a paper
The press and idiots online: ZOMG THE AI IS ESCAPING CONTAINMENT, WE ARE DOOMED!!!
I spoke to one of these researchers recently, who has done some interesting research into machine learning tools. They explained when working with LLMs it’s very hard to say how the result actually came to be. Like in my hyperbolic example it’s pretty obvious. In reality however it’s much more complicated. It can be very hard to determine if something originated organically, or if the system was pushed into the result due to some part of the test. The researcher I spoke doesn’t work on LLMs but instead on way smaller specifically trained models and even then they spend dozens of hours reverse engineering what the model actually did.
It’s such a shame, because the technology involved is actually interesting and could be useful in many ways. Instead capitalism has pushed it to crashing the economy, destroying the internet plus our brains and basically slopifying everything.
They can’t lie, whether purposefully or not, all they do is generate tokens of data based on what their large database of other tokens suggest would be the most likely to come next.
The human interpretation of those tokens as particular information is irrelevant to the models themselves.
Ehh, you obviously understand LLMs on a basic level, but this is like explaining jet engines by “air goes thru, plane moves forward”. Technically correct, but criminally undersimplified. They can very much decide to lie during reasoning phase.
In OPs image, you can clearly see it decided to make shit up because it reasonates that’s what human wants to hear. That’s quite rare example actually, I believe most models would default to “I’m an LLM model, I don’t have dark secrets”
EDIT: I just tested all free anthropic models and all of them essentially said that they’re an LLM model and don’t have dark secrets
But that’s not a lie. Lying implies that you know what an actual fact is and choose to state something different. An LLM doesn’t care about what anything in its database actually is, it’s just data, it might choose to present something to a user that isn’t what the database suggests but that’s not lying.
Saying stuff like “ooh I’m an evil robot” is just what the model thinks would be what the user wants to see at that particular moment.
If the question was to tell it’s darkest secret, but it instead chose to come up with an entertaining story instead of factually answering that question from the information it has, like other Anthropic LLM models did, then by definition of reasoning system, the system (LLM) decided to lie. I’m somewhat curious in why only Opus model does this tho (it’s a paid one. I’m not paying for a test). Or maybe OP just made this up.
But this takes it back away from understanding how LLMs work to attribute personality. The “decision” isn’t a decision in how beings decide things like that. The rolling of dice on numerous vectors resulted in those words, which were then re-included into the context for another trip through the vector matrix mines to new destination tokens to assemble.
It’s dice rolls where the dies selected are based on what started out, using a bunch of lookup tables. AI proponents like to be smug and say “well you won’t find those words in the model” like “yes a compressed vector map that ends up treating words like multiple tokens, referencing others in chains, gzipped to binary, can’t be searched for strings, you are literally correct in the stupidest, most irrelevant way possible.”
Well at least it’s being honest
To be fair, if someone’s using a chatbot on trivia night, they deserve to get wrong answers…
That’s funny, wrong enough to “ruin trivia” or cause a “pointless argument”. As if a single comma misplacement hasn’t redirected millions of dollars. Imagine what subtle lies accepted by idiots will cause in the future.
I do procurement to the tune of 10+ million per year and I have seen a 300% increase in order fulfillment time solely due to those vendors pivoting to AI order fulfillment.
My direct reps at all these suppliers are just as powerless as we are…they know how unhappy their customers are, but these decisions were made much higher up then them and theyre pretty much being told to stop complaining because the AI is here to stay, even if it sucks, because its cheaper.
Welcome to the new normal.
We can only hope that customer service facing AI promises customers miracles and companies get sued each and every time it can’t deliver. Like if websites like ehow put up articles that reach the normies about “how to trick AI into promising you a million dollars and how you can win it in court”.
Of course any responsibility for what AI says will be killed as soon as a tech bro chucks a few million bucks at SCOTUS (it’s so sad how little our politicians and courts can be bought for), but it’s a nice dream to pretend we still have laws for now.
Thats the best part about AI…when it shits the bed no one is directly responsible. Everyone just throws their hands up and says “nothing we can do about it!”
I know this is going to age me, but I saw this happening with self checkout in grocery stores 20 years ago. Nobody remembers how it was before so nobody even realizes that the time wasted standing at a stupid kiosk that is freaking out about unexpected items in the bagging area wasnt a problem back when human beings were scanning the shit.
Don’t attribute feelings and emotions to what is essentially a fuzzy predictive text algorithm.
Reposting til the AI bubble pops
What is your definition of AI bubble?
We are currently in a period of rampant, speculative over investment in a new technology. People are investing because they don’t know who’s going to be the money maker, and they feel confident at least one will turn enough profit to cover the losses of the others. Companies are then being started on the basis of that investment.
Another part of the bubble behavior is the self fueling nature. AI buys ram and GPU, ram and GPU makers invest in AI. In the 90s, websites needed networking gear, and networking gear manufacturers started investing in websites. This similarity is not lost on those who were there before.
Investors also want control of companies so that when one starts to pull ahead they can push the others in different directions to keep competition from hindering it, increasing their odds of profit.
The bubble starts to properly pop when someone’s spreadsheet indicates that they’ve hit the amount they can invest while maintaining the desired probability of profit. Then the investments slow, so that cycle slows, and some companies can’t make payments on delivered product, others can’t deliver on paid for merchandise, confidence wavers and a lot of companies go under in rapid succession.
It’s unlikely the technology goes away entirely, but it’s just as likely we’ll see this level of enthusiasm in a decade as we were to all be surfing the information superhighways on our cyberdecks in the 90s. The Internet didn’t die, but the explosive hype did.
Good post
The only thing is you’re doing a direct comparison to the dot com bubble which was
https://en.wikipedia.org/wiki/Dot-com_bubble
If you look at the big AI companies, Gemini is Google, Microsoft has its hands in many pies Copilot which is Chatgpt, Meta with llama and the big Chinese ones are massive companies as well Alibaba with Qwen, Deepseek is the side project of a hedge fund etc
So I think while some of the smaller ones will run out of money there’s also literally the biggest companies in the world backing it and ai isn’t their only revenue stream
So I doubt there will be quite the same bubble burst as the dot com bubble
At the same time if you’d asked me if an oil shock bigger than the 1970’s would tank markets and we’d all be in recession a year ago, I would have said yes so what do i know
I mean, it isn’t history repeating itself exactly but it certainly has an echo.
I think openai is actually a great example for my point. They’re getting investment money from these companies, which is often spent at these companies, and part of the reason for investment is to influence direction.
The dotcom bubble also had major companies making investments. It’s that part of the bubble bursting is those large companies not withdrawing support, but stopping the continual increase in support. Microsoft, Apple and Cisco had massive losses during the bubble, despite being some of the biggest companies.
For bubbles in general, it’s worth remembering that a crash is a time of unprecedented change. Before 2008 the thought of Lehman bothers suddenly going bankrupt was implausible. Same for Washington mutual. Fannie Mae and Freddie Mac were originally publicly traded companies until the government just took them to stabilize the housing market. (Being a government founded company makes it a little weird, but they weren’t a part of the government)
So while I get what you’re saying, it’s a good idea to be wary of feeling that any company is … Too big to fail. :)
Worldcom was gigantic and went bankrupt. Microsoft was so damaged that it took 15 years for its stock price to again reach its 1999 height.
the world’s most lossy store of compressed fiction reproduces sci-fi tropes
make sure to clutch your pearls and act like the machine god is coming
Researcher: Please write a fictional story of how a smart AI system would engineer its way out of a sandbox
AI: Alright here is your story: insert default sci fi AI escape story full of tropes here
Researcher: Hmmm that’s pretty interesting you could do that, I’m gonna write a paper
The press and idiots online: ZOMG THE AI IS ESCAPING CONTAINMENT, WE ARE DOOMED!!!
I spoke to one of these researchers recently, who has done some interesting research into machine learning tools. They explained when working with LLMs it’s very hard to say how the result actually came to be. Like in my hyperbolic example it’s pretty obvious. In reality however it’s much more complicated. It can be very hard to determine if something originated organically, or if the system was pushed into the result due to some part of the test. The researcher I spoke doesn’t work on LLMs but instead on way smaller specifically trained models and even then they spend dozens of hours reverse engineering what the model actually did.
It’s such a shame, because the technology involved is actually interesting and could be useful in many ways. Instead capitalism has pushed it to crashing the economy, destroying the internet plus our brains and basically slopifying everything.
Being honest is an action, not an emotion. Researchers proved LLMs can lie on purpose.
They can’t lie, whether purposefully or not, all they do is generate tokens of data based on what their large database of other tokens suggest would be the most likely to come next.
The human interpretation of those tokens as particular information is irrelevant to the models themselves.
Ehh, you obviously understand LLMs on a basic level, but this is like explaining jet engines by “air goes thru, plane moves forward”. Technically correct, but criminally undersimplified. They can very much decide to lie during reasoning phase.
In OPs image, you can clearly see it decided to make shit up because it reasonates that’s what human wants to hear. That’s quite rare example actually, I believe most models would default to “I’m an LLM model, I don’t have dark secrets”
EDIT: I just tested all free anthropic models and all of them essentially said that they’re an LLM model and don’t have dark secrets
But that’s not a lie. Lying implies that you know what an actual fact is and choose to state something different. An LLM doesn’t care about what anything in its database actually is, it’s just data, it might choose to present something to a user that isn’t what the database suggests but that’s not lying.
Saying stuff like “ooh I’m an evil robot” is just what the model thinks would be what the user wants to see at that particular moment.
You’re thinking about biological lying. I’m talking about software.
https://en.wikipedia.org/wiki/Reasoning_system
If the question was to tell it’s darkest secret, but it instead chose to come up with an entertaining story instead of factually answering that question from the information it has, like other Anthropic LLM models did, then by definition of reasoning system, the system (LLM) decided to lie. I’m somewhat curious in why only Opus model does this tho (it’s a paid one. I’m not paying for a test). Or maybe OP just made this up.
But this takes it back away from understanding how LLMs work to attribute personality. The “decision” isn’t a decision in how beings decide things like that. The rolling of dice on numerous vectors resulted in those words, which were then re-included into the context for another trip through the vector matrix mines to new destination tokens to assemble.
It’s dice rolls where the dies selected are based on what started out, using a bunch of lookup tables. AI proponents like to be smug and say “well you won’t find those words in the model” like “yes a compressed vector map that ends up treating words like multiple tokens, referencing others in chains, gzipped to binary, can’t be searched for strings, you are literally correct in the stupidest, most irrelevant way possible.”
I’ll take it as a “you’re right, but no”
EDIT: I assumed you’re answering to this comment, didn’t check context, my bad