The rubber duck doesn’t feed back anything that might affect how you think. It’s more like brainstorming with someone who is both sycophantic and has their own (unthought-though) ideas about how you should do it. At best it’s a waste of tokens because a rubber duck would have worked, at worst it directs your thought process rather than letting you think.
Are you so easily manipulated or why do you think that is always the case?
Everything has the potential of influencing how you think, it is on you how you let that affect you.
LMs also work for that. If there is somebody who just accepts everything the software puts out, that is a problem with the person.
Critical thinking should always be applied no matter where information comes from. But I guess critical thinking is on the downtrend, measured by the amount of people spinning up one scenario in their head, arguing based on just that while ignoring every other possible conclusions.
Brainstorming solo versus brainstorming in a group are very different activities. Your assertion was that brainstorming with an LLM, which feeds back more-or-less like another participant, was the same as brainstorming with a rubber duck, i.e. solo, which it is not.
Brainstorming solo is an exercise in fully exploring your understanding of a subject, and developing new insights or approaches to it. Brainstorming as a group is more about finding a consensus solution or approach. Doing so with an LLM is much closer to the latter, however your “partner” has no actual experience but still presents feedback as though it did, and, critically, has an inbuilt tendancy to confirm your biases. Thinking of that like either solo or group brainstorming will automatically warp your thought processes by encouraging you to stay on certain tracks.
In all fairness I was referring to the op which stated you let the LM brainstorm “for you”. I wanted to get the point across that you can also talk to it like you would to a rubber duck.
But i agree with you that there is a difference between talking to an LM and talking to a rubber duck forvtge reasons you described. It’s its own thing.
But I disagree with your view on brainstorming. I mean when you say it is an exercise in exploring your understanding… Well it can be but this is not all it can be.
When I brainstorm for a project I have an idea but not necessarily all knowledge about all the things I need/want for it. So it can help to gather reference points, technology that fits the project, etc.
And the encouraging to stay on certain track is also true but can be a good thing.
Not really and not always. I had a question about a law that seemingly changed recently. ChatGPT absolutely refused to agree with me that “$X is now legal”. It gave me several sources for more information and examples of government speech on the subject, but just for fun I pushed it to agree with me. Couldn’t make that happen.
meh, that’s just a removable layer on top. the uncensored models are sycophantic as all getout. they have to actively work against that in the frontend.
When brainstorming with an AI, you aren’t brainstorming with their CEO. Even Grok can’t do that, no matter how much they try. Amd that is why people don’t use grok.
They do have some influence over what the AI says… but they also influence people with news and ads. So both the person and the AI are being influenced by billionaires and companies. The person by many many more actually.
The assertion was the brainstorming with an LLM was like brainstorming with a rubber duck. It is not, as I’ve outlined in another comment.
Brainstorming with another person can also lead to biases, you’re correct, but your partner is less likely to just agree with your ideas, or to have built in biases that you’re not aware of.
A handful of employees in a conference room are more likely to just agree. Those same employees will be less agreeable on video, and even less so in the company chat. People will consciously or unconsciously hide their biases in most contexts.
A handful of employees in a conference room are more likely to just agree.
I see you’ve never met my coworkers before! The politest way to describe that sort of gathering would be “a free and frank exchange of opinions”. I quite like them for that. The point is that I know their biases already, so I can adjust for them, and I can be confident that they’re actually trying to achieve the same thing I am, rather than regurgitating a probabilistically derived sequence of tokens that strongly resemble, but aren’t, actual thought.
The biases I’m talking about here are usually more along the lines of preferred approaches to problems, general pessimism or optimism in outlook, and any “oneupmanship” that is known to occur, rather than anything people tend to hide.
I will give you that on average, a human partner is less likely to agree. But there are plenty who will just agree with everything you say. So as humans, we evaluate what kind of partner we have, and treat their feedback accordingly. It’s the same with AI. You could of course tell AI not to be that way if you want.
Bingo! Brainstorming is the one thing LLMs excel at. I also use ChatGPT for find words or terms I’ve forgotten or don’t know. Just found “availability heuristic” this morning. (It’s the reason the internet has ruined our perceptions of risk.)
You brainstorm with the robot. Like you brainstorm with a rubber duck.
The rubber duck doesn’t feed back anything that might affect how you think. It’s more like brainstorming with someone who is both sycophantic and has their own (unthought-though) ideas about how you should do it. At best it’s a waste of tokens because a rubber duck would have worked, at worst it directs your thought process rather than letting you think.
Are you so easily manipulated or why do you think that is always the case?
Everything has the potential of influencing how you think, it is on you how you let that affect you.
LMs also work for that. If there is somebody who just accepts everything the software puts out, that is a problem with the person.
Critical thinking should always be applied no matter where information comes from. But I guess critical thinking is on the downtrend, measured by the amount of people spinning up one scenario in their head, arguing based on just that while ignoring every other possible conclusions.
Brainstorming solo versus brainstorming in a group are very different activities. Your assertion was that brainstorming with an LLM, which feeds back more-or-less like another participant, was the same as brainstorming with a rubber duck, i.e. solo, which it is not.
Brainstorming solo is an exercise in fully exploring your understanding of a subject, and developing new insights or approaches to it. Brainstorming as a group is more about finding a consensus solution or approach. Doing so with an LLM is much closer to the latter, however your “partner” has no actual experience but still presents feedback as though it did, and, critically, has an inbuilt tendancy to confirm your biases. Thinking of that like either solo or group brainstorming will automatically warp your thought processes by encouraging you to stay on certain tracks.
In all fairness I was referring to the op which stated you let the LM brainstorm “for you”. I wanted to get the point across that you can also talk to it like you would to a rubber duck.
But i agree with you that there is a difference between talking to an LM and talking to a rubber duck forvtge reasons you described. It’s its own thing.
But I disagree with your view on brainstorming. I mean when you say it is an exercise in exploring your understanding… Well it can be but this is not all it can be.
When I brainstorm for a project I have an idea but not necessarily all knowledge about all the things I need/want for it. So it can help to gather reference points, technology that fits the project, etc.
And the encouraging to stay on certain track is also true but can be a good thing.
it’s brainstorming with a rubber duck that has the word “yes” written on it
Not really and not always. I had a question about a law that seemingly changed recently. ChatGPT absolutely refused to agree with me that “$X is now legal”. It gave me several sources for more information and examples of government speech on the subject, but just for fun I pushed it to agree with me. Couldn’t make that happen.
meh, that’s just a removable layer on top. the uncensored models are sycophantic as all getout. they have to actively work against that in the frontend.
So when I brainstorm with a person, they are never confidently wrong? And they never have their own ideas on how I should do it.
Does that person have billions of dollars and power over all political leaders to sway a country toward totalitarianism ? No? Thats the difference.
When brainstorming with an AI, you aren’t brainstorming with their CEO. Even Grok can’t do that, no matter how much they try. Amd that is why people don’t use grok. They do have some influence over what the AI says… but they also influence people with news and ads. So both the person and the AI are being influenced by billionaires and companies. The person by many many more actually.
The assertion was the brainstorming with an LLM was like brainstorming with a rubber duck. It is not, as I’ve outlined in another comment.
Brainstorming with another person can also lead to biases, you’re correct, but your partner is less likely to just agree with your ideas, or to have built in biases that you’re not aware of.
Hard agree. And, uh, disagree.
A handful of employees in a conference room are more likely to just agree. Those same employees will be less agreeable on video, and even less so in the company chat. People will consciously or unconsciously hide their biases in most contexts.
I see you’ve never met my coworkers before! The politest way to describe that sort of gathering would be “a free and frank exchange of opinions”. I quite like them for that. The point is that I know their biases already, so I can adjust for them, and I can be confident that they’re actually trying to achieve the same thing I am, rather than regurgitating a probabilistically derived sequence of tokens that strongly resemble, but aren’t, actual thought.
The biases I’m talking about here are usually more along the lines of preferred approaches to problems, general pessimism or optimism in outlook, and any “oneupmanship” that is known to occur, rather than anything people tend to hide.
I will give you that on average, a human partner is less likely to agree. But there are plenty who will just agree with everything you say. So as humans, we evaluate what kind of partner we have, and treat their feedback accordingly. It’s the same with AI. You could of course tell AI not to be that way if you want.
A rubber duck doesn’t lie to you and stroke your ego off
A robot doesn’t lie either because lying needs intent. And if a rubber duck doesn’t stroke you ego, well that’s just a bad rubber duck.
Bingo! Brainstorming is the one thing LLMs excel at. I also use ChatGPT for find words or terms I’ve forgotten or don’t know. Just found “availability heuristic” this morning. (It’s the reason the internet has ruined our perceptions of risk.)