Aha! I really appreciate this article. As a former City of London solicitor, now retired, I would have loved to have set up a scheme like this when I was still in practice. Still, I did a similar thing and built a team of agents to help me write, research, and market my self-published historical thriller, The Stars of Samarkand.
Here's a link where I explained the genesis of the team and how they assist me with the project: https://samarkandstars.substack.com/p/build-your-own-ai-team. You'll see that I actually directed people to your site for more detailed info on how to use AI.
So it all comes round in a nice, virtuous circle. Bravo to Ms Rojas, and bravo to you for teaching us all so much about AI. Thanks very much.
Named, role-scoped agents work well until two of them end up touching the same decision, and then you get the exact failure the setup was supposed to prevent: nobody quite owns the call, or two agents hand you conflicting answers built from the same notes.
Curious how you handle that: who gets final say when two of your agents' lanes overlap on one decision.
As yet, I have not had an instance of this, so I cannot say for sure. What I have observed, though, is that each agent sticks to its rules quite rigorously, and sometimes one will simply say, 'That's not really a question for me, that's for Agent X.'
One of the things I did find, though, was that my research agent, codename Chekhov, on at least one occasion simply presented as fact things that I knew to be patently false. That is when it was brought home to me quite viscerally that agents need to use their training data as a guide, but to be explicit about what is known and what is not. In this instance, the Chekhov agent was 'trying to be helpful' and simply made up some facts that it thought would be reassuring. I did not realise until that point that an agent would do that.
So, I halted all of them, investigated it deeply, and rewrote their instruction files to ensure, as far as possible, that this does not happen again. Each agent now has to give a score out of 10 on the reliability of what they have presented, and to highlight where their knowledge stops and where I need to go and do my own independent research.
In whatever field, I think that one needs to approach AI as a useful but perhaps inexperienced research assistant. In my former role as a lawyer in London, I had many assistants working under me, and I would never accept their word as gospel on important issues without checking it myself. I think one needs to treat AI in exactly the same way — apply critical thinking, and don't take the AI response as the last word.
The lawfull4us.com is dedicated to help immigrants in the USA, it have a collection of procedures and regulations based on current laws and legislation. I help pro-bono attorney and perse clients
De longe o teu jeito de ver as coisas com ia são as melhores que encontrei na web, e eu falo de como consegue ser tão eficaz no que procuro vivenciar quando estou criando qualquer coisa.
Uma base muito clara e eficiente de compreensão da ferramenta, tudo muito simples, e no Brasil um compositor diz em uma música sua que "o simples resolve tudo" e você simplifica o que parece complicado, qualidade que confunde com termos mais etéreos como "magia" "milagre" e outros do tipo.
Para fechar meu raciocínio muito menos lógico e eficaz que o seu eu senti uma vontade absurda de refazer cada um dos produtos e soluções que eu ja criei com esse recorte desse post aqui.
Você é incrível querida menina, simplificando tudo❤️
Aha! I really appreciate this article. As a former City of London solicitor, now retired, I would have loved to have set up a scheme like this when I was still in practice. Still, I did a similar thing and built a team of agents to help me write, research, and market my self-published historical thriller, The Stars of Samarkand.
Here's a link where I explained the genesis of the team and how they assist me with the project: https://samarkandstars.substack.com/p/build-your-own-ai-team. You'll see that I actually directed people to your site for more detailed info on how to use AI.
So it all comes round in a nice, virtuous circle. Bravo to Ms Rojas, and bravo to you for teaching us all so much about AI. Thanks very much.
thanks for reading!! very cool re your thriller!
Named, role-scoped agents work well until two of them end up touching the same decision, and then you get the exact failure the setup was supposed to prevent: nobody quite owns the call, or two agents hand you conflicting answers built from the same notes.
Curious how you handle that: who gets final say when two of your agents' lanes overlap on one decision.
As yet, I have not had an instance of this, so I cannot say for sure. What I have observed, though, is that each agent sticks to its rules quite rigorously, and sometimes one will simply say, 'That's not really a question for me, that's for Agent X.'
One of the things I did find, though, was that my research agent, codename Chekhov, on at least one occasion simply presented as fact things that I knew to be patently false. That is when it was brought home to me quite viscerally that agents need to use their training data as a guide, but to be explicit about what is known and what is not. In this instance, the Chekhov agent was 'trying to be helpful' and simply made up some facts that it thought would be reassuring. I did not realise until that point that an agent would do that.
So, I halted all of them, investigated it deeply, and rewrote their instruction files to ensure, as far as possible, that this does not happen again. Each agent now has to give a score out of 10 on the reliability of what they have presented, and to highlight where their knowledge stops and where I need to go and do my own independent research.
In whatever field, I think that one needs to approach AI as a useful but perhaps inexperienced research assistant. In my former role as a lawyer in London, I had many assistants working under me, and I would never accept their word as gospel on important issues without checking it myself. I think one needs to treat AI in exactly the same way — apply critical thinking, and don't take the AI response as the last word.
The lawfull4us.com is dedicated to help immigrants in the USA, it have a collection of procedures and regulations based on current laws and legislation. I help pro-bono attorney and perse clients
De longe o teu jeito de ver as coisas com ia são as melhores que encontrei na web, e eu falo de como consegue ser tão eficaz no que procuro vivenciar quando estou criando qualquer coisa.
Uma base muito clara e eficiente de compreensão da ferramenta, tudo muito simples, e no Brasil um compositor diz em uma música sua que "o simples resolve tudo" e você simplifica o que parece complicado, qualidade que confunde com termos mais etéreos como "magia" "milagre" e outros do tipo.
Para fechar meu raciocínio muito menos lógico e eficaz que o seu eu senti uma vontade absurda de refazer cada um dos produtos e soluções que eu ja criei com esse recorte desse post aqui.
Você é incrível querida menina, simplificando tudo❤️
Do you teach a boot camp or is someone scamming people on Threads?
scam
Such a detailed article!