DEV Community: Victor Bueno

AI in the Wild: A ChatGPT Simulated Town 🤖🔥

Victor Bueno — Mon, 08 Jan 2024 12:32:38 +0000

Have you heard about a village where all its citizens are controlled by artificial intelligence? It sounds like a great movie plot, but it's not fiction—it's a study done a few months ago that I just discovered and felt compelled to share in this article.

The study

Named "Generative Agents: Interactive Simulacra of Human Behavior", this study, led by Google and Stanford researchers, involved 25 ChatGPT agents playing a game where each one assumed the role of a character in a simulated town. The agents were provided with simple descriptions, including their occupation and relationships with other agents, essentially receiving a snapshot of their past memories and motivations.

Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day.

A memory system also was implemented, so all the experiences "lived" by the agents were stored in their own memory. For instance, when an agent named Eddy was telling his father, John, about his homework, John stored that information so later when the kid was off to school and his wife asked about their son, John knew that Eddy was excited with the music that he was composing as a homework and was able to share that.

What happened

Some interesting things happened during the study, such as:

Each agent created their own routine without specific instructions given, just based on their histories, including daily activities as brushing teeth, taking shower and catching up with other agents.
One of the agents was given the intent, in their history, to throw a Valentine's Day party. After proposing the event to other agents, they autonomously spread invitations to the party and even the ones that weren't invited by the host got to know about the event trough others, mirroring real-life information flow in a city;
At the Valentine's Day party, guests and the host coordinated to schedule their arrival, showcasing a level of collaboration akin to real-world social dynamics.
Another example of information diffusion was the fact that a mayoral election was happening and agents that weren't aware of that got information from other agents and were even capable of discussing about the election later.
The agents were reflecting about themselves and what matters to them. One for instance, from his own observations, was able to "understand" how dedicated he is to research activities.

Valentine's Day Party Information Diffusion

AIs more human than... Humans?

A controlled evaluation tested whether ChatGPT agents could exhibit believable human behavior based on their environment and experiences.
As the agents responds to natural language question they were basically "interviewed" to assess five human abilities: maintaining self-knowledge, retrieving memory, generating plans, reacting, and reflecting.

A group of 25 humans watched full replays of the simulation, each following one agent while inspecting the corresponding agent memory data. These individuals were then interviewed, roleplaying as the selected agent. In the final evaluation, other human assessors were invited to rank the scenarios by believability. The result was striking—ChatGPT proved more convincing at roleplaying a human than an actual human, perhaps not surprising at this point.

Scenarios removing agents capabilities of Plan, Reflection and Observation were also ranked

Glitch in the Matrix

ChatGPT is amazing, we already know that, but the agents presented some behaviors that weren't expected by the researchers:

Some agents failed to absorb information adequately, occasionally retrieving incomplete memory fragments during interviews;
The stores at the simulated town closes at 5 p.m. but a few agents enter the stores after closing hours, misunderstanding the shop's operating hours.
Even do the Bar was supposed to be a place to go later on the day, some agents during the simulation preferred to have lunch there instead of the café.

My thoughts

This kind of study are really interesting to me, I don't work directly with AI or machine learning but it's a field that makes me reflect a lot.
And honestly, this is impressive, looking how human an artificial intelligence can get. Even the presented "errors" from the agents sounds a lot like a human to me... Who doesn't forget information that they should know? Who doesn't go to a store just to find out it's closed? And who has never grab a beer earlier than they should?

This makes me think, how far artificial intelligence will go? Perhaps in the near future, video games will feature NPCs (non-playable characters) entirely controlled by AIs, free from predetermined scripts, offering an immersive experience beyond our current imagination.

I would love to hear your thoughts on the matter, what implications do you think this study could lead to?

If you are interested in delving deeper into the research paper, click here to access the official GitHub, where you can find the complete study and instructions to replicate the simulation.

Should I Write Articles in English or in My Native Language?

Victor Bueno — Tue, 26 Dec 2023 18:23:32 +0000

I find myself in a bit of a dilemma and would appreciate your thoughts on the matter.

The Dilemma

As the title of the article suggests, I'm not sure if I should contribute to the community writing articles in English or in my native language. And here is why:

My Expirence

This marks my second article, and the first one was written in Portuguese, my native language. That was a choice that I made trying to contribute with the Brazilians seeking more content on a particular subject but may not be comfortable enough with English to read an entire article. And I was thinking about keeping that way, but one of the first comments that I got was something like: "good article, next time you should do it in english". Since then, I've been grappling with this decision.

Why write in English?

Before writing this, I read some articles advocating for writing exclusively in English. Anyone can imagine the main reasons to do so:

Improve your own English knowlodge;
Reach more people;
Help on your career growth;

And I guess that they are really valid! Also, as I consume content in English here on dev.to, I intereact a lot with people that doesn't speak my native language, so producing content in English seems like an excellent way to continue contributing.

But why write in your native language?

A comment that I saw in one of the articles defending writing in english says: "English is the language of the internet". While I acknowledge its important role in connecting people globally, isn't that segregating a large part of the world who lack access to resources for learning a new language? Unfortunely, in Brazil, for example, having access to education is a privilege that not everyone can have.
While I aspire to contribute globally, I also want to make a difference on the life of the people from my country (even if it's a small part). Wouldn't it be more meaningful to give back to the community I'm physically surrounded by and create accessible content for them?

Isn't writing in both languages an option?

Certainly, there's the option of writing the same content in two different languages, but I feel it might not be effective on the internet. I've never come across a successful YouTube channel, for instance, that produces content in two different languages on the same profile. I fear that duplicating all my posts, one in English and the other in my native language, might lead to losing both audiences and not reaching anyone at all.

Share your toughts

What are your toughts? Do you have any relavant experience on the matter? I'd love to hear back from you all :)

Diferenças entre Monólito e Microsserviços

Victor Bueno — Tue, 22 Feb 2022 12:10:47 +0000

Esta semana iniciei em uma nova empresa e ouvi, de um desenvolvedor mais experiente da equipe, que o principal produto que vou trabalhar havia sido desenvolvido como um Monólito e agora estavam sendo iniciados os esforços a fim de migrá-lo para uma arquitetura de Microsserviços. Eu já havia escutado esses termos mas não entendia bem o que eles implicavam, então fui buscar entendê-los melhor... Agora com esse conhecimento adquirido, venho compartilhar as diferenças que vi entre esses dois tipos de arquitetura.

O que é um Monólito?

A palavra monólito faz referência a um monumento formado por um único bloco de pedra e isso nos explica muita coisa!

Um software que adota uma arquitetura Monolítica tem todo seu código e funcionamento concentrados em um só processo. Logo, todas as funções do negócio estão dividindo os mesmos recursos de uma mesma máquina.

Vantagens

Novos desenvolvedores vão ter uma maior facilidade de entender o projeto, pois todo o código se concentra em um único lugar.
O desenvolvimento inicial é mais fácil e rápido. Além de, enquanto pequeno, possuir baixa complexidade.

Desvantagens

Acarreta a uma difícil manutenção no futuro, quando ao alterar parte de uma funcionalidade é criado um efeito cascata (que pode causar comportamentos inesperados) em outras funções que estão ligadas de alguma forma. Afinal, há uma alta dependência de componentes do código.
Seu principal ponto negativo: escalabilidade. Ou melhor, a falta dela... Por todo o funcionamento estar integrado, não podemos escalar uma função individualmente, é necessário que todo o produto se expanda em conjunto.

O que é um Microsserviço?

Um sistema que utiliza uma arquitetura de Microsserviços é composto por partes menores. Enquanto o Monólito é uma grande e única pedra, aqui temos diversas pedrinhas.

Nesse tipo de software vamos ter um código dividido em vários projetos que se comunicam entre si formando o produto final. Assim, cada serviço é criado de acordo com um conjunto de regras de negócio específicas, sendo implementado de forma independente.

Vantagens

A manutenção se torna mais fácil por se trabalhar com partes menores de código e que não vão gerar problemas nos outros serviços do sistema.
Cada parte do software pode ser desenvolvida com a tecnologia que melhor se encaixa nela, sem a necessidade de ter o projeto todo usando uma única linguagem, por exemplo.
Fazer o deploy de um microsserviço é muito mais simples, facilitando a chegada de atualizações à produção.
A escalabilidade está presente, de forma que se torna possível expandir apenas os microsserviços desejados conforme a necessidade daquele momento.

Desvantagens

Como temos um sistema distribuído, deve haver alguma forma de conectá-lo e trabalhar nisso traz uma dificuldade adicional ao projeto.
O aprendizado é dificultado para desenvolvedores que entrarem no projeto ao longo de sua criação tendo que compreender o funcionamento de diversos serviços.

Suas diferenças

Acredito que agora você já tenha entendido o conceito dessas arquiteturas e caso não, podemos resumir "Monólito vs Microsserviços" como "Centralizado vs Descentralizado". De um lado temos um grande bloco de código e, em outro, pequenos blocos interligados por conexões.

Ao olhar as vantagens e desvantagens de cada arquitetura entendemos que:

Enquanto Monólito é fácil de um novo integrante do time aprender, o Microsserviços pode complicar o aprendizado, inclusive na maneira que foi escolhida fazer a ligação entre os serviços.
Monólitos são eficientes para POCs (Prove of Concept) por terem baixa complexidade inicial. Porém, como não possuem boa escalabilidade, caso queira levar o projeto a frente faz sentido convertê-lo para microsserviços.
Prestar manutenção em um Monólito é mais trabalhoso, enquanto nos microsserviços a independência do código leva vantagem.

Conclusão

Se você está na dúvida entre qual arquitetura seu produto deve utilizar, acredito que uma boa alternativa é aproveitar o melhor dos dois mundos, iniciando o sistema com a utilização de uma arquitetura monolítica e tornando-a descentralizada quando a necessidade de escalabilidade chegar.
É isso que ocorreu com o produto da nova empresa em que estou trabalhando e após meus estudos somado à escrita desse artigo entendi o por quê dessa mudança estar sendo feita.

Este é o primeiro artigo que escrevo e o conhecimento compartilhado foi adquirido do meu próprio estudo então caso tenha algo a acrescentar ou corrigir, não hesite.
Espero que esse artigo contribua com a comunidade e até a próxima! 😉