Wild to read this in light of what Gary Marcus says about LLMs (alone) not getting us close to AGI, and that the other (neurosymbolic hybrid) approaches that *could* are too funding-starved. One says “hurricane by Tuesday,” the other says “you don’t even have the goods for hurricanes." Wondering where you come down on these two different takes, Kamil?
To be perfectly honest with you, I don’t know. I do believe, though that even with current technology, we could already automate 70% of knowledge jobs if not more. However, this is the same scenario as with nuclear energy. We’ve had the safest cleanest technology for the last 80 years or so and we’re barely making use of it that we wait for other innovation to come and we keep changing the goal post.
My point is that only because we have access to certain technology it doesn’t matter if we don’t apply it.
Cason point why do we need realtors? Don’t get me wrong. I love realtors. I think they’re real hustlers and very disciplined at least the good ones but technically we wouldn’t need them anymore yet they’re still a huge part of our economy.
Yep, total sense. Realtors = travel agents (remember those)? And a number of occupations can already go this way.
Still, I was fighting Chatty-G for a full hour yesterday trying to get it to actually READ what I uploaded so we could pattern-spot together. I kept pointing out mistakes and it was like “you’re absolutely right to flag this. It’s fixed now” and it still goofed. I told it it’s not going to replace *anybody* with that kind of attitude. 😬
That’s not necessarily cleaned up with the mental models Marcus wants to see, but the error rate is still pretty high and not necessarily closing with simple scaling.
Yeah, I recently published an article on how to use advanced voice mode as a intruder for any subject, including a very lengthy prompt, and to be honest with you. I’m very disappointed because it’s totally hidden and miss and the underlying model seems to keep interpreting my prompt in a different way.
Currently human plus AI is still the best solution, but I don’t see why that would still be the case two years from now.
+1 to this. I have been using Claude and Grok to help me write a story. It’s my new form of entertainment instead of watching TV plus it’s helping me understand what current “AI” can and can’t do.
Anyway, sometimes I tell it to edit something and it says it did but it didn’t. It also gets instructions wrong and sometimes forgets things which is interesting. I will say the sequence of events is ABCD but when it writes, it will do ABDC.
This is a perfect example of how to spend your time wisely while focusing on a narrow AI topic in order to get a greater understanding of how these models work in practice. Bravo.
Wow. A lot to unpack here. Thanks for the article!
Assuming your scenario is true (and as the other commenter mentioned, it might not be), as a business should you even bother because with 99% unemployment, you will have no customers because nobody will have any money.
Alternatively, I wonder if we’ll have a “perma-depression” where we end up with high unemployment which then starves out capital investment in further automation due to lower sales and higher taxes in order to pay for all the social welfare programs. And anytime things get better economically they get worse since companies will be desperate to lower costs which just continues the feedback loop.
Nobody really knows where things will end up going, but I can assure you that this overly speaking knowledge is power. So if you spent some time throughout your day educating yourself on the topics that I cover in this newsletter and in others, then you can be certain to a degree that you will be more adaptable to handle any future scenario better than those who don’t take their time to learn.
Wild to read this in light of what Gary Marcus says about LLMs (alone) not getting us close to AGI, and that the other (neurosymbolic hybrid) approaches that *could* are too funding-starved. One says “hurricane by Tuesday,” the other says “you don’t even have the goods for hurricanes." Wondering where you come down on these two different takes, Kamil?
To be perfectly honest with you, I don’t know. I do believe, though that even with current technology, we could already automate 70% of knowledge jobs if not more. However, this is the same scenario as with nuclear energy. We’ve had the safest cleanest technology for the last 80 years or so and we’re barely making use of it that we wait for other innovation to come and we keep changing the goal post.
My point is that only because we have access to certain technology it doesn’t matter if we don’t apply it.
Cason point why do we need realtors? Don’t get me wrong. I love realtors. I think they’re real hustlers and very disciplined at least the good ones but technically we wouldn’t need them anymore yet they’re still a huge part of our economy.
Does that make sense?
Yep, total sense. Realtors = travel agents (remember those)? And a number of occupations can already go this way.
Still, I was fighting Chatty-G for a full hour yesterday trying to get it to actually READ what I uploaded so we could pattern-spot together. I kept pointing out mistakes and it was like “you’re absolutely right to flag this. It’s fixed now” and it still goofed. I told it it’s not going to replace *anybody* with that kind of attitude. 😬
That’s not necessarily cleaned up with the mental models Marcus wants to see, but the error rate is still pretty high and not necessarily closing with simple scaling.
Yeah, I recently published an article on how to use advanced voice mode as a intruder for any subject, including a very lengthy prompt, and to be honest with you. I’m very disappointed because it’s totally hidden and miss and the underlying model seems to keep interpreting my prompt in a different way.
Currently human plus AI is still the best solution, but I don’t see why that would still be the case two years from now.
+1 to this. I have been using Claude and Grok to help me write a story. It’s my new form of entertainment instead of watching TV plus it’s helping me understand what current “AI” can and can’t do.
Anyway, sometimes I tell it to edit something and it says it did but it didn’t. It also gets instructions wrong and sometimes forgets things which is interesting. I will say the sequence of events is ABCD but when it writes, it will do ABDC.
This is a perfect example of how to spend your time wisely while focusing on a narrow AI topic in order to get a greater understanding of how these models work in practice. Bravo.
Wow. A lot to unpack here. Thanks for the article!
Assuming your scenario is true (and as the other commenter mentioned, it might not be), as a business should you even bother because with 99% unemployment, you will have no customers because nobody will have any money.
Alternatively, I wonder if we’ll have a “perma-depression” where we end up with high unemployment which then starves out capital investment in further automation due to lower sales and higher taxes in order to pay for all the social welfare programs. And anytime things get better economically they get worse since companies will be desperate to lower costs which just continues the feedback loop.
Nobody really knows where things will end up going, but I can assure you that this overly speaking knowledge is power. So if you spent some time throughout your day educating yourself on the topics that I cover in this newsletter and in others, then you can be certain to a degree that you will be more adaptable to handle any future scenario better than those who don’t take their time to learn.