Thank you can you please create a video detailing how to update the GPTs system instructions you plan to use to account for the new @mentions feature for seamless teamwork and collaboration between your team of GPTs
Interesting! Funny: when you said Tech Lead I thought you were talking about the millionaire UA-camr 😂 quick question: is this a Minecraft plugin you are working on? Or a client? Good video 👍
Thanks! It's a Mineflayer client. My idea is basically to create hard-coded bots to collect data that can then be used to train ML models for more sophisticated bots.
@@ai-lucas fascinating - please keep us posted. Just looking up mineflayer rn 😱 Had no idea this existed. At one point I was using something called “baritone” to make mc an idle game. Mineflayer looks like just the ticket. Going to dig in, thanks for the tip.
I hope I get to use AI to generate code for at least a few years before the AI is doing the whole job 😂. AGI in April would be a bit disappointing in that respect.
Thanks for sharing. I don't want to be negative, but are we sure that if you just had custom instructions reflecting the needs of the different GPTs then it wouldn't have done as good a job?
Great question. You certainly could try providing one GPT multiple sets of Role/Audience/Objective/Method/Context and then telling it which to respond with for each message. It's worth trying if you don't have access to GPT mentions yet. That said, projects like AutoGen and David Shapiro's agent swarms are based on the theory that you can generate better code with multiple agents, each with their own role. AutoGen is a big project and might have data on this somewhere.
@@ai-lucas Yes it's all very interesting. I just think that in the case of chatgpt there's not enough context to benefit much from this sort of thing. But I could be wrong. Definitely will be fun experimenting!
This paper on the REACT prompting method is a good one for showing the power of multiple roles (plan, act, observe): arxiv.org/pdf/2210.03629.pdf I only just realized upon going back to this that I had reproduced this structure with my GPTs. Tech lead plans, coder acts, and reviewer observes. Per the paper, this seems to be an effective structure for arbitrary problem solving. I want to try using it to come-up with the software design itself, which I had to create myself for this video.
Thank you can you please create a video detailing how to update the GPTs system instructions you plan to use to account for the new @mentions feature for seamless teamwork and collaboration between your team of GPTs
Interesting! Funny: when you said Tech Lead I thought you were talking about the millionaire UA-camr 😂 quick question: is this a Minecraft plugin you are working on? Or a client? Good video 👍
Thanks! It's a Mineflayer client. My idea is basically to create hard-coded bots to collect data that can then be used to train ML models for more sophisticated bots.
@@ai-lucas fascinating - please keep us posted. Just looking up mineflayer rn 😱 Had no idea this existed. At one point I was using something called “baritone” to make mc an idle game. Mineflayer looks like just the ticket. Going to dig in, thanks for the tip.
When are we getting an AI generated model of you generating AI models?
I hope I get to use AI to generate code for at least a few years before the AI is doing the whole job 😂. AGI in April would be a bit disappointing in that respect.
Thanks for sharing. I don't want to be negative, but are we sure that if you just had custom instructions reflecting the needs of the different GPTs then it wouldn't have done as good a job?
Great question. You certainly could try providing one GPT multiple sets of Role/Audience/Objective/Method/Context and then telling it which to respond with for each message. It's worth trying if you don't have access to GPT mentions yet.
That said, projects like AutoGen and David Shapiro's agent swarms are based on the theory that you can generate better code with multiple agents, each with their own role. AutoGen is a big project and might have data on this somewhere.
@@ai-lucas Yes it's all very interesting. I just think that in the case of chatgpt there's not enough context to benefit much from this sort of thing. But I could be wrong. Definitely will be fun experimenting!
This paper on the REACT prompting method is a good one for showing the power of multiple roles (plan, act, observe): arxiv.org/pdf/2210.03629.pdf
I only just realized upon going back to this that I had reproduced this structure with my GPTs. Tech lead plans, coder acts, and reviewer observes. Per the paper, this seems to be an effective structure for arbitrary problem solving. I want to try using it to come-up with the software design itself, which I had to create myself for this video.