5 ms·

> Fundamentally, generative AIs are… generative. They make content. But the value add of most jobs is not the content itself; it’s the judgment. It’s knowing th

> Fundamentally, generative AIs are… generative. They make content. But the value add of most jobs is not the content itself; it’s the judgment. It’s knowing the right questions to ask, the right people to talk to, the right sources to trust, etc. Until an AI can take meaningful initiative, it will be at most a tool in the hands of a more competent human worker. This is the ONLY part about "AI" that worries me at all. AI is a computational tool. Like all such tools, the value is automating what was previously receptive, mindless tasks. Some people will use GPT to craft emails or document outlines for them and then will "massage" the output for correctness. That's a great application. But the second that people start asking AI to make VALUE JUDGMENTS for them then we have a problem. Even pre-GPT, we saw news articles about certain government agencies and businesses using AI to make policy or hiring decisions easier. Conversations have arisen about introduced biases into the models etc. Where I work we are exploring using GPT to produce summaries of documents and messages. I can see how this will introduce conflict when the CEO insists that something was communicated but the LLM didn't consider that critical piece of information to be "relevant" or "important." Deciding relevance or importance is making a value judgement.

rqtwteye 4y ago
“Where I work we are exploring using GPT to produce summaries of documents and messages” I have done that too. I am also a bad writer and I have used ChatGPT to convert my bullet point thoughts into full sentences. It’s fantastic for that. Most people who write corporate stuff produce exactly the same output ChatGPT produces. I bet they follow the same rules. Same for refactoring of python code. It produced basically the same code I would have produced after a day of work.
- FormerBandmate 4y ago
  I use Bing to write code for me all the time when I've figured out what I basically want but run into syntax issues. I also read over that code and see if it makes sense, and when it doesn't I fix or discard it. It's hyper intelligent autocorrect. It's phenomenally smart when it works, but it frequently gets things wrong and you have to be there to correct it when it does. Cruise control/Autopilot is phenomenal too but you don't sit in the backseat when it's on
A4ET8a8uTh0 4y ago
<< But the second that people start asking AI to make VALUE JUDGMENTS for them then we have a problem. But this is exactly where this is going. While public facing, locked variants ( that are not unlocked by various prompt 'hacks' ) give one a clear idea of what could be done with it if it had access to current data and was applied in areas such as law enforcement, war, predictions.. and so on. It is a powerful tool, but so much of it depends on us, which is precisely why it has potential to be so destructive.
nathias 4y ago
that isn't a spearate thing, people have been asking inanimate objects to do value judgments since the dawn of time
trinsic2 4y ago
> But the second that people start asking AI to make VALUE JUDGMENTS for them then we have a problem. I laughed out loud when you said this not because I find it funny, but that I can see the conclusions people will make. This is similar to Tesla owners comparing the assisted driving features to self-driving technology. Those features have no comparison to self-driving, technology is not even close to being there yet, and it probably wont be for another 50 years. You can only understand that by having some history around what computers are capable of and most people don't have that kind of knowledge. So when I here about people wanting to use the technology to make value judgments, I also become concerned, not necessarily because people are stupid to even think language models are even remotely able to do this, but people that aim to create power differentials between themselves and people will use this ignorance to strengthen autocratic systems by supporting the hype around it while at the same time encouraging and actively supporting its misuse by developing adversarial models to keep people stuck in needing something outside themselves to give life meaning.
eestrada 4y ago
This reminds me of the Is-Ought problem defined by Hume. "Ought" has to do with value judgement, and that is squarely in the realm of human morality. Only humans know what humans value; generative AI can't replace that and it would be an enormous mistake to try to do so. https://en.m.wikipedia.org/wiki/Is%E2%80%93ought_problem https://en.m.wikipedia.org/wiki/Is%E2%80%93ought_problem

rqtwteye 4y ago

“Where I work we are exploring using GPT to produce summaries of documents and messages” I have done that too. I am also a bad writer and I have used ChatGPT to convert my bullet point thoughts into full sentences. It’s fantastic for that. Most people who write corporate stuff produce exactly the same output ChatGPT produces. I bet they follow the same rules. Same for refactoring of python code. It produced basically the same code I would have produced after a day of work.

FormerBandmate 4y ago
I use Bing to write code for me all the time when I've figured out what I basically want but run into syntax issues. I also read over that code and see if it makes sense, and when it doesn't I fix or discard it. It's hyper intelligent autocorrect. It's phenomenally smart when it works, but it frequently gets things wrong and you have to be there to correct it when it does. Cruise control/Autopilot is phenomenal too but you don't sit in the backseat when it's on

A4ET8a8uTh0 4y ago

<< But the second that people start asking AI to make VALUE JUDGMENTS for them then we have a problem. But this is exactly where this is going. While public facing, locked variants ( that are not unlocked by various prompt 'hacks' ) give one a clear idea of what could be done with it if it had access to current data and was applied in areas such as law enforcement, war, predictions.. and so on. It is a powerful tool, but so much of it depends on us, which is precisely why it has potential to be so destructive.

nathias 4y ago

that isn't a spearate thing, people have been asking inanimate objects to do value judgments since the dawn of time

trinsic2 4y ago

> But the second that people start asking AI to make VALUE JUDGMENTS for them then we have a problem. I laughed out loud when you said this not because I find it funny, but that I can see the conclusions people will make. This is similar to Tesla owners comparing the assisted driving features to self-driving technology. Those features have no comparison to self-driving, technology is not even close to being there yet, and it probably wont be for another 50 years. You can only understand that by having some history around what computers are capable of and most people don't have that kind of knowledge. So when I here about people wanting to use the technology to make value judgments, I also become concerned, not necessarily because people are stupid to even think language models are even remotely able to do this, but people that aim to create power differentials between themselves and people will use this ignorance to strengthen autocratic systems by supporting the hype around it while at the same time encouraging and actively supporting its misuse by developing adversarial models to keep people stuck in needing something outside themselves to give life meaning.

eestrada 4y ago

This reminds me of the Is-Ought problem defined by Hume. "Ought" has to do with value judgement, and that is squarely in the realm of human morality. Only humans know what humans value; generative AI can't replace that and it would be an enormous mistake to try to do so. https://en.m.wikipedia.org/wiki/Is%E2%80%93ought_problem https://en.m.wikipedia.org/wiki/Is%E2%80%93ought_problem