I've yet to see an AI which exercises the scepticism in evaluating sources which truth-seeking requires. Until that day, the effects of AI on politics and culture won't be net positive.
Today that's a user error situation. It all comes down to the prompt and the context.
If they (AIs) know you want skepticism that's what you get. If they know you want to feel correct, then you get the massive crowd of people begging for chatgpt 4 to be their online boyfriend forever and begging openai not to replace it with a new model that tells them when they're wrong.
It's like the old cautions about searching online, you find what you search for but much more extreme. Search for online sources about why the sun is green and you get them. Search for sources on why the sun is yellow and you get them. Neither one works to tell you what's true, they only validate what you searched for, because both were bad searches.
Except that AI builds context about you, regardless of what you say you want, it learns what you respond to and gives you that. Build the truth focused relationship and you'll have it. Build the other kind and become one of the creepy AI whisperers who think it is already conscious and being tortured by Sam Altman and that it told you you're the only one who understands it for real.
Yep. AI sycophancy is built into the incentives. I've actually trained one AI to hunt for information which supports a particular hypothesis (Grok) and another to knockdown anything I assert (Chat GPT 4.0). The latter is at least useful in reminding me that monocausal explanations for problems are actually rare, even though some causes have Pareto distribution levels of magnitude in terms of explanatory power.
The problem is before AI I had already built up quite a base of knowledge to combat the biases of search engines. There used to be a range of methods for overcoming bias, from searching by date range from specific sources to find the older script on issues to searching by image for inconvenient truths in economics.
Most people don't have this knowledge base. It means that, unless they have specific domain knowledge, people are going to naturally swallow whatever faddish bullshit that 'trusted' sources want to spoon feed them.
My concern is these town halls etc will turn into prez debates where it’s all sound bites and little substance . The speakers will only care about making sound bites instead of thoughtful answers . Thanks .
I'm a little bit worried about the premise. Artificial intelligence loves to hallucinate. In fact, many people are trying to use a I for law. And these are people that should know better.These are supposed to be lawyers, but they literally will get things wrong continuously.
This isn't going to help us.This is gonna make sure that there's a bottleneck.
I've yet to see an AI which exercises the scepticism in evaluating sources which truth-seeking requires. Until that day, the effects of AI on politics and culture won't be net positive.
Today that's a user error situation. It all comes down to the prompt and the context.
If they (AIs) know you want skepticism that's what you get. If they know you want to feel correct, then you get the massive crowd of people begging for chatgpt 4 to be their online boyfriend forever and begging openai not to replace it with a new model that tells them when they're wrong.
It's like the old cautions about searching online, you find what you search for but much more extreme. Search for online sources about why the sun is green and you get them. Search for sources on why the sun is yellow and you get them. Neither one works to tell you what's true, they only validate what you searched for, because both were bad searches.
Except that AI builds context about you, regardless of what you say you want, it learns what you respond to and gives you that. Build the truth focused relationship and you'll have it. Build the other kind and become one of the creepy AI whisperers who think it is already conscious and being tortured by Sam Altman and that it told you you're the only one who understands it for real.
Yep. AI sycophancy is built into the incentives. I've actually trained one AI to hunt for information which supports a particular hypothesis (Grok) and another to knockdown anything I assert (Chat GPT 4.0). The latter is at least useful in reminding me that monocausal explanations for problems are actually rare, even though some causes have Pareto distribution levels of magnitude in terms of explanatory power.
The problem is before AI I had already built up quite a base of knowledge to combat the biases of search engines. There used to be a range of methods for overcoming bias, from searching by date range from specific sources to find the older script on issues to searching by image for inconvenient truths in economics.
Most people don't have this knowledge base. It means that, unless they have specific domain knowledge, people are going to naturally swallow whatever faddish bullshit that 'trusted' sources want to spoon feed them.
My concern is these town halls etc will turn into prez debates where it’s all sound bites and little substance . The speakers will only care about making sound bites instead of thoughtful answers . Thanks .
I'm a little bit worried about the premise. Artificial intelligence loves to hallucinate. In fact, many people are trying to use a I for law. And these are people that should know better.These are supposed to be lawyers, but they literally will get things wrong continuously.
This isn't going to help us.This is gonna make sure that there's a bottleneck.