This entirely depends on whether you ask the AI to build code that it has seen before.
Ask it to build an app using a rarely used or novel framework, and it still fumbles spectacularly. So yeah, if you just want it to build a generic app for a website, you may no longer need people. It may do fine by itself. But ask it to try and read a datasheet for a new electronic component and write a device driver for an OS, and watch the AI go into serious meltdown.
And it is unlikely to get any better. Not because the models aren't improving, but because the current approach towards AI (transformers and attention) are actually pretty awful at solving problems they haven't been trained on. It needs several examples to do anything reasonable, and those simply don't exist for novel parts, libraries and mechanisms. Don't be fooled. The "reasoning" models aren't doing anything of the sort. At least not in the way humans do.
So you'll forgive me if I have exactly the opposite impression as this writer. He is talking from a very narrow perspective.
Thank you! I agree with you. Its frightening how susceptible people are to fear mongering.
Ive tried one of your exact examples: writing a device driver for windows. It quite literally did not know where to begin. This was just a few weeks ago.
It would get stuck in endless feedback loops where it would change one line, which broke another, so then it would fix that and break the other one again, and this would go on endlessly. I would hold its hand through that and then it would happen again in a different place in the code
Then you get people like Reggie who just counter anything contrary to ai with "I don't think you understand how LLMs work". The people who say that can never seem to say how it works, while countering people giving insight into exactly how it works
An LLM is a large language model. It is not capable of AGI. Not even fucking close lol
Respectfully, you have NO IDEA of my background, or you wouldn't make such a ridiculous statement. I have a degree in Computational Neural Systems, and have been designing neural networks since the late 80's.
“If” knowledge exists in a quantum realm and “”if”” conscience beings do or are capable of tapping into that for information / knowledge (think Akashic Records) then is it reasonable to think it’s possible a quantum computer might be able to do the same?
Quantum computer, yes. Today's AI's use digital electronics that specifically engineer out the quantum interface and replace it with a stochastic process. It is not possible for an LLM as we have them today to be conscious.
This entirely depends on whether you ask the AI to build code that it has seen before.
Ask it to build an app using a rarely used or novel framework, and it still fumbles spectacularly. So yeah, if you just want it to build a generic app for a website, you may no longer need people. It may do fine by itself. But ask it to try and read a datasheet for a new electronic component and write a device driver for an OS, and watch the AI go into serious meltdown.
And it is unlikely to get any better. Not because the models aren't improving, but because the current approach towards AI (transformers and attention) are actually pretty awful at solving problems they haven't been trained on. It needs several examples to do anything reasonable, and those simply don't exist for novel parts, libraries and mechanisms. Don't be fooled. The "reasoning" models aren't doing anything of the sort. At least not in the way humans do.
So you'll forgive me if I have exactly the opposite impression as this writer. He is talking from a very narrow perspective.
Thank you! I agree with you. Its frightening how susceptible people are to fear mongering.
Ive tried one of your exact examples: writing a device driver for windows. It quite literally did not know where to begin. This was just a few weeks ago.
It would get stuck in endless feedback loops where it would change one line, which broke another, so then it would fix that and break the other one again, and this would go on endlessly. I would hold its hand through that and then it would happen again in a different place in the code
Then you get people like Reggie who just counter anything contrary to ai with "I don't think you understand how LLMs work". The people who say that can never seem to say how it works, while countering people giving insight into exactly how it works
An LLM is a large language model. It is not capable of AGI. Not even fucking close lol
Respectfully, I don't think you understand how LLMs work. So many don't.
let me guess… you do?
Respectfully, you have NO IDEA of my background, or you wouldn't make such a ridiculous statement. I have a degree in Computational Neural Systems, and have been designing neural networks since the late 80's.
“If” knowledge exists in a quantum realm and “”if”” conscience beings do or are capable of tapping into that for information / knowledge (think Akashic Records) then is it reasonable to think it’s possible a quantum computer might be able to do the same?
It seems reasonable.
Quantum computer, yes. Today's AI's use digital electronics that specifically engineer out the quantum interface and replace it with a stochastic process. It is not possible for an LLM as we have them today to be conscious.