Ariel Flint Ashery 🧜‍♂️

The first pre-print to come from my work on populations of LLMs! Critically, we argue that AI alignment requires *collective alignment* 👨‍👨‍👧‍👧. Much more is in the works, stay tuned!

Join Complexity72h in Madrid!

🏅Ready to coach their #Complexity72h team, meet Javier M. Buldú (@JmBuldu) from URJC, Brennan Klein (@jkbren) & Maddalena Torricelli (@___Maddalena___) from @NUnetsi. ⚽️Their project: "Leveraging Complexity Sciences to Analyze Soccer Performance Evolution Over the Last Decade"

🚨 Complexity72h 2025 is OPEN for applications! 🚨 Join us in Madrid for a 72-hour workshop, solving cutting-edge challenges in complex systems alongside 17 tutors across 12 projects! Stay tuned as we reveal the lineup. 🌟 📍Apply now: complexity72h.com #Complexity72h

Do you have a great research idea but need the right people to bring it to life? Help us make Complexity72h an unforgettable experience again by applying to be a tutor in Madrid next June. The call for tutors is now open: forms.gle/q6APB672iXRzMe…

This reminds me of a time during my undergrad when we had to write a short essay about any physics related topic. My favourite one was an essay about the most optimal way to get ketchup out of a glass bottle.

@elisa_omodei @CompSysSoc Congratulations Elisa!!

Such an incredible honor! For which I would like to thank all my mentors & collaborators, as well as the wonderful @CompSysSoc #ComplexSystems community, which is the most welcoming, encouraging & motivating scientific home I could ever find 🙏

Ariel Flint Ashery experimentally shows LLMs can spontaneously generate conventions through interactions among themselves (without human intervention) #CCS2024 @ConfCompSys

I'm very happy to take part in the 2024 edition of the Computational Modelling of Cognition Summer school in the stunning mountains of Hinterstoder, Austria 😁 I'll pass by Vienna on the 17th of August, if anyone would like to catch up over coffee and a strudel, let me know!

That’s all folks! …until next time!

35 hours to go... @Complexity72h

The clock has started ticking at #Complexity72h, less than 65h to go! 📸: @SantoroArnaldo @RELflintashery @pires_dl @CassellsDuncan @Lorenzo89350671 @msanchezc_ & @gshreyas5 working hard on the food insecurity & migration project

M. Ángeles Serrano discussing hidden metrics spaces and the S¹ model @Complexity72h

@a_baronca I'll take that pay!

Seemingly legit message: "Hi Andrea, <company> is looking for mathematics experts to help train AI systems and LLMs. Your profile stood out", "Earn up to $50 per hour"

Multi-agent collaboration has emerged as a key AI agentic design pattern. Given a complex task like writing software, a multi-agent approach would break down the task into subtasks to be executed by different roles -- such as a software engineer, product manager, designer, QA (quality assurance) engineer, and so on -- and have different agents accomplish different subtasks. Different agents might be built by prompting one LLM (or, if you prefer, different LLMs) to carry out different tasks. For example, to build a software engineer agent, we might prompt the LLM: "You are an expert in writing clear, efficient code. Write code to perform the task …". It might seem counterintuitive that, although we are making multiple calls to the same LLM, we apply the programming abstraction of using multiple agents. I'd like to offer a few reasons: - It works! Many teams are getting good results with this method, and there's nothing like results! Further, ablation studies (for example, in the AutoGen paper cited below) show that multiple agents give superior performance to a single agent. - Even though some LLMs today can accept very long input contexts (for instance, Gemini 1.5 Pro accepts 1 million tokens), their ability to truly understand long, complex inputs is mixed. An agentic workflow in which the LLM is prompted to focus on one thing at a time can give better performance. By telling it when it should play software engineer, we can also specify what is important in that subtask: For example, the prompt above emphasized clear, efficient code as opposed to, say, scalable and highly secure code. By decomposing the overall task into subtasks, we can optimize the subtasks better. - Perhaps most important, the multi-agent design pattern gives us, as developers, a framework for breaking down complex tasks into subtasks. When writing code to run on a single CPU, we often break our program up into different processes or threads. This is a useful abstraction that lets us decompose a task -- like implementing a web browser -- into subtasks that are easier to code. I find thinking through multi-agents roles to be a useful abstraction. In many companies, managers routinely decide what roles to hire, and then how to split complex projects -- like writing a large piece of software or preparing a research report -- into smaller tasks to assign to employees with different specialties. Using multiple agents is analogous. Each agent implements its own workflow, has its own memory (itself a rapidly evolving area in agentic technologies -- how can an agent remember enough of its past interactions to perform better on upcoming ones?), and may ask other agents for help. Agents themselves can also engage in Planning and Tool Use. This results in a cacophony of LLM calls and message passing between agents that can result in very complex workflows. While managing people is hard, it's a sufficiently familiar idea that it gives us a mental framework for how to "hire" and assign tasks to our AI agents. Fortunately, the damage from mismanaging an AI agent is much lower than that from mismanaging humans! Emerging frameworks like AutoGen, Crew AI, and LangGraph, provide rich ways to build multi-agent solutions to problems. If you're interested in playing with a fun multi-agent system, also check out ChatDev, an open source implementation of a set of agents that run a virtual software company. I encourage you to check out their github repo and perhaps even clone the repo and run the system yourself. While it may not always produce what you want, you might be amazed at how well it does! Like the design pattern of Planning, I find the output quality of multi-agent collaboration hard to predict. The more mature patterns of Reflection and Tool use are more reliable. I hope you enjoy playing with these agentic design patterns and that they produce amazing results for you! If you're interested in learning more, I recommend: - Communicative Agents for Software Development, Qian et al. (2023) (the ChatDev paper) - AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation, Wu et al. (2023) - MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework, Hong et al. (2023) [Original text: deeplearning.ai/the-batch/issu… ]

Frozen sound waves distributed throughout our universe, with galaxies forming at their peaks. Physics doesn't get much cooler than that!

Awesome programme, join us next Friday (5/4) Worth stressing: we’re in the Franklin building, 5 mins from main campus: 124 Goswell Road, London, EC1V 7DP maps.google.com/maps/place//da…

It's finally here, the full schedule for Data Natives 2024 😱 We're super excited to welcome our keynote speaker Eleanor Power (@eleanorapower), who will talk to us about 'Studying the dynamics of social inequality'. See you all @CityUniLondon (room R204), April 5th!

We will be hosting our first ever virtual "open house" on Weds 3rd April! Been following the Raphtory project and want to hear more? Want some inspiration on digging into your new graphy dataset? Got feature requests? All are welcome to join us for a virtual coffee ☕️👇

@chanwoopark20 First time it ever happened to me that I googled a question, and the perfect paper for it had been released 10 hours earlier. Crazy coincidence!

@RELflintashery Thank you for your kind introduction to our paper on twitter!!

Do LLM Agents Have Regret? A Case Study in Online Learning and Games arxiv.org/abs/2403.16843 Awesome work @chanwoopark20, Xiangyu Liu and team!