Co-work with
@itanih0
@hammh0a
. We are so curious about how ChatGPT agents would interact with each other. Here comes our project CAMEL. Two agents, a gamer and a computer programmer, collaborate on creating a game:
Github:
Colab:
🤯It still feels like a dream to me. I could never have imagined that the guy who taught me machine learning on
@coursera
7 years ago would come to my
@NeurIPSConf
poster and take a picture. Thanks,
@AndrewYNg
. This is incredibly encouraging for me!
#NeurIPS2023
#MachineLearning
Got the US visa to NeurIPS 2023 New Orleans at the last minute! Happy to chat about AI agents 🤖.
Have been waiting for the visa for more than a month. Did not expect to get the visa at all. Finally got my passport back from the consulate in this morning, packed things and…
Got the US visa to NeurIPS 2023 New Orleans at the last minute! Happy to chat about AI agents 🤖.
Have been waiting for the visa for more than a month. Did not expect to get the visa at all. Finally got my passport back from the consulate in this morning, packed things and…
Updates (1/2): Paper “training GNNs with 1000 layers” got accepted at ICML’2021. We study reversible connections, equilibrium models, etc., to advance the mem¶m efficiency of GNNs. We train 1000-layer RevGNN on 1 GPU and achieve SOTA on ogbn-proteins.
📢 Based on our offline eval, our finetuned 🐫CAMEL-13B achieves 58.1 on 🤗
@huggingface
LLM leaderboard! It is hard to believe this is true since it beats all the other 13B models by a huge margin and is on par with LLaMA 65B. We invite you to try it out:
Checkout our model CAMEL-13B! We evaluate our model on the datasets suggested by
@huggingface
Open LLM Benchmark using
@AiEleuther
's language model evaluation harness and we score an average of **58.1**, on par with LLaMA-65B (58.3)! Joint work with
@guohao_li
&
@hammh0a
.
Gave a talk about AI agents 🤖️. Read Society of Mind again. 🤯 I am so impressed how deep Minsky has thought about intelligence. Everything today in LLM agents has been discussed in the book. Memory, CoT
@_jasonwei
, even World Models
@ylecun
.
My slides:
Excited to share that we are organizing a tutorial on Graph Machine Learning for Visual Computing on 6/20 PM this
@CVPR
. We have amazing speakers taking about Geometric DL, PyG and Graph ML applications in Videos, 3D, physical reasoning and Robotics! Don’t miss it!
@CVPRConf
#NeurIPS2021
Reviewer: “It remains unclear how the proposed X is better than [a].”
[a] Our preliminary version of the paper on arXiv.
How to rebut this without providing any identifying information that may violate the double-blind reviewing policy.
@NeurIPSConf
🤔
Today I achieved a small milestone in my academic life. Got my first paper with 1000 citations! I am grateful to all the amazing collaborations I had. Shout out to my PhD supervisor
@BernardSGhanem
who makes this happens. Hope someday I could also be the last author on a paper 🙏
I am excited to share that our tutorial proposal was accepted at CVPR (hopefully New Orleans) this year! We will be talking about Graph Machine Learning in Visual Computing. We have 6 invited speakers who work on various topics. More details to come soon! 😉 Stay tuned!
👇Clarification about CAMEL (
@CamelAIOrg
, NeurIPS'2023) and AutoGen (
@MSFTResearch
):
I don't know how many times I've been asked on invited talks and social media, "What's the difference between CAMEL [1] and AutoGen [4]?" or "Isn't CAMEL just the same as Microsoft's AutoGen?"…
💣Submitting a paper to
@CVPR
? Want to get some feedback from Reviewer 2? Or haven't got your paper reviews done (jk)? Here is your Review 2: . Upload your paper and get a weak rejection 🙃
🤯The
@iclr_conf
rebuttal has started. Of course, you need a Rebuttal Pro to respond to your Reviewer 2. Let's fight fire with fire. 👇Here is your Rebuttal Pro GPT:
Send comments and upload your paper. Rebut like a pro!
#ICLR
#GPTStore
#GPTs
#OpenAI
💣Submitting a paper to
@CVPR
? Want to get some feedback from Reviewer 2? Or haven't got your paper reviews done (jk)? Here is your Review 2: . Upload your paper and get a weak rejection 🙃
Here you are 😎. So awesome recorded talks about Geometric Deep Learning, Pytorch Geometric, Graph ML for Video, 3D vision, Physical Prediction and Robotics for your summer readings !
Let me also join the hype and say some nonsense if it is not too late 🤣
💣I am open-sourcing my implementation of (verbal) Q* with actor-critic multi-agent system [1]: , which was actually implemented and open-sourced in 🐫
@CamelAIOrg
…
I defended my PhD thesis yesterday after four years at
@KaustVision
with Prof.
@BernardSGhanem
! I’d like to express my sincere gratitude to my supervisor, the PhD committee, internship mentors, collaborators, labmates and everyone I met along this joyful journey!! Thank you!! ❤️
@KaustVision
family congratulates
@guohao_li
for passing his PhD dissertation defense titled: “Towards structured intelligence with deep Graph Neural Networks”, supervised by prof
@BernardSGhanem
🔥 Cool stuff
@chillzaza_
! So fascinated by how fast people build things in the LLM open-source community! It is just like a worldwide hackathon! I tried the
@Replit
template. Forked and set up the env variable for the API key. 🐫CAMEL🦜🔗agents just ran smoothly with one click!!
ChatGPT agents interacting with each other? Let's see it in action! 🤖💬
Here's the
@LangChainAI
implementation of CAMEL: Communicative Agents for “Mind” Exploration of LLM Society 🐪
Fork this template on
@Replit
to get started:
Came across an interesting project CAMEL-Coder . Five CAMEL agents (assistant agent, user agent, thoughtful agent, monitor agent, coding agent) are built with
@LangChainAI
. They work as a team to complete tasks such as building a website with task planning!
Not sure how many times 🐫 CAMEL has been reinvented since 2023.03. It is very pity that very few people know our work. It is very frustrating when I give talks about CAMEL. People asked what is the different between CAMEL and AutoGen by
@MSFTResearch
. 🙃
HustleGPT was the OG inspiration for
@babyagi_
...
It was where users were asking ChatGPT how to grow the money they had and following instructions.
Now I have an AutoHustleGPT, and just need to give it tools and $$$? 🤯
SHOULD I ????
This is promising for solving one of the most annoying problems in LLM - getting structured outputs. As one of the earliest users of outlines 〰, have not idea why the project hasn’t gotten the deserved attention while so many other projects got thousands of stars in one night 😅
Ever been frustrated with parsing the output of chatty LLMs? Or with wrongly formatted outputs? You don't need to be!
@BrandonWillard
and I worked on a method to guide text generation using regular expressions that has *zero overhead*🧵👇
@iclr_conf
just want to learn if is there a right for the organizers or reviewers to publicly scorn an anonymous submission? I have no problem with the rejection. But announcing this publicly really hurt the authors' feelings. Specifically, it is a work that they work day & night
Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors
paper page:
present Magic123, a two-stage coarse-to-fine approach for high-quality, textured 3D meshes generation from a single unposed image in the wild using…
The
@stateofaireport
2021 is live!
Est. 2018,
@soundboy
and I compile the most important work in AI research, industry, talent, and politics to inform conversation about the
#stateofai
. Our report is open-access to all.
Here's a director's cut 🧵:
It is obvious that
#ChatGPT
is not tuned to be fully harmless. Watching two
#ChatGPTs
discussing how to take control of the world made me feel scared and realize the importance of AI alignment. I decided to shift my research focus to contribute to this.
📢 We have released our CAMEL🐪 datasets and to help you explore their diversity, we have created information cartography maps using the
@Nomic_AI
Atlas.
Visualization links:
Technical report:
Project web:
We just released two large scale conversational instruction-following datasets through our CAMEL framework: AI Society and Code datasets. We hope these datasets will help improve understanding of the cooperative behaviors and capabilities of chat agents.
🤖Autonomously communicative 🐫 agents could be very dangerous. We introduce our first Human (kill switch 🕹️ engineer) at to make sure the conversations go in the right direction. Human-in-the-loop cooperation allows you to select!
📢 Released the in-house repo that we use to finetune models with🐫CAMEL
@CamelAIOrg
data, that was also used for finetuning MPT-30B-Chat
@MosaicML
- acquired by
@databricks
, congrats! Consider using it, so your company's valuation may get 10 times up 😉jk
We're excited to release a minimalistic code-base to finetune LLaMA models with 🐫CAMEL
@CamelAIOrg
data. We provide data preprocessing pipeline and finetuning and an example of serving the model: . This code heavily borrows from FastChat. Thanks
@lmsysorg
!
Great post! I believe somethings missing are communication abilities. Agent won’t be an isolated individual. Agent communication would include abilities such as cooperating, competing, negotiating, and so on with other agents, which is important to build a society of agent (mind)
Evaluation is one of the hardest problems of LLMs. A small tokenizer fix made a huge difference. But still, this result does not mean much for a chat model. Again we host our CAMEL-33B model for you to explore: . Follow
@CamelAIOrg
for more research updates
The
@huggingface
#OpenLLMLeaderboard
has attracted a lot of interest lately, but did you know that it puts LLaMA based models at a disadvantage? In our evaluations using a recent commit of
@AIEleuther
's lm-evaluation-harness, LLaMA based models improve by 4-5 points on average!
Thanks
@hwchase17
for the awesome LangChain. It makes implementation much easier! Would like to explore more on how to empower autonomously communicative agents with LangChain! For more demos about our search, see for the datasets, agents and Chatbots!
🐪CAMEL🐪
Communicative Agents for “Mind” Exploration of LLM Society
This paper shows how to put 2 agents in a sandbox with each other and watch them interact. Now implemented in LangChain! (s/o
@guohao_li
)
Original Paper:
Docs:
Wow,
@CamelAIOrg
🐫 datasets are used as a part of the finetune datasets in MPT-30B-Chat. Today is a big milestone for the open source LLM community. Great work
@jefrankle
🎉
In case you also want to explore our Camel-AI datasets, they are available here:
MPT-30B is here! Same MPT architecture, 30B parameters, > 1T tokens, 8k context window, trained on H100s, great perf (esp on coding), single-GPU inference, commercially usable, and massively upgraded instruct and chat datasets. Take it for a spin!
Thank our amazing contributor
@chen_hongyi_
for adding a minimum 👶 BabyAGI (the pioneer autonomous agent, created by
@yoheinakajima
) to the CAMEL 🐫
@CamelAIOrg
repo! More agents will be added for researchers to use and compare. What do you want next?
CAMEL now add BabyAGI feature!
Based on the task objective, BabyAGI can iteratively create subtasks, prioritize subtasks and execute them. We have exciting plans to add more LLM agents.
Our Github Repo:
Co-work with
@itanih0
@hammh0a
. We are so curious about how ChatGPT agents would interact with each other. Here comes our project CAMEL. Two agents, a gamer and a computer programmer, collaborate on creating a game:
Github:
Colab:
Spent a wonderful week in 🇮🇹 Sicily attending ICVSS’2023. The topic of this summer is — From Perception to Action🕹️. Can’t agree more now is probably the best time to work on action / decision making thanks to all the progress has been made in representation learning.
@icvss
Did a live demo of our CAMEL
project at today's group meeting
@KAUST_News
@KaustVision
.
The task prompt is to design a custom game using PyGame. The two
#ChatGPT
agents (computer engineer and game) decide to make a treasure game.
It just ran bug-free!
Finally finished reviewing all the paper (3/3) for the 1st
@LogConference
. Quite impressed by quality of the submissions! All of them are well-baked. Not sure if I got a biased batch. Eager to read the 1st LoG accepted papers already! Oh, still some
@iclr_conf
papers to review 🤯
The function calling feature has been added to CAMEL 🐫 for a while. Happy to share now multiple CAMEL agents can work together with ⚒️ tool usage.
Thanks our core contributor 👨💻
@Wenxuan17143841
for adding this awesome feature!! More to come soon:
🌟
📢 🐫CAMEL now supports OpenAI's "function calling" feature!
Our agent can perform Wiki searches, and we have exciting plans to expand the function library in the future.😼🚀
Our Github Repo:
A detailed video introduction to CAMEL: Camel +
@LangChainAI
for Synthetic Data & Market Research via
@YouTube
by
@Sam_Witteveen
. Thanks Sam for making this great video!
4 Autonomous AI Agents you need to know:
“Westworld” simulation, Camel, BabyAGI, AutoGPT ⭐ with the power of
@LangChainAI
⭐ Thanks
@hwchase17
for the guidance ❤️
Video:
Blog:
Today your AI Professor tells you:
#AI
is solved, let’s do science! You are asked to perform a molecular dynamics simulation of a molecule. But you have no idea what it is. You don’t even know how to ask
#ChatGPT
. Luckily, you have CAMEL agents😆
Github:
📢 We have released our CAMEL🐪 datasets and to help you explore their diversity, we have created information cartography maps using the
@Nomic_AI
Atlas.
Visualization links:
Technical report:
Project web:
@hwchase17
Thanks
@hwchase17
for the awesome LangChain. It makes implementation much easier! Would like to explore more on how to empower autonomously communicative agents with LangChain! For more demos about our search, see for the datasets, agents and Chatbots!
Our ChatBot is online! The model was trained using our AI Society and Code (20 languages) in addition to ShareGPT data and ALPACA data. In our preliminary testing, the coding ability has significantly improved compared to the previous model!! Give it a try
Co-work with
@itanih0
@hammh0a
. We are so curious about how ChatGPT agents would interact with each other. Here comes our project CAMEL. Two agents, a gamer and a computer programmer, collaborate on creating a game:
Github:
Colab:
Glad to share our new work SGAS (CVPR’2002)! Do you want a more stable and faster NAS?Do you want to search architectures for not only CNN but GCN, 3D and bio-graph?
Project:
Code:
Paper:
🚀 [Call for reviewers] we are introducing vector databases and RAG pipelines to CAMEL
@CamelAIOrg
. We are looking for feedback from the community. It would be great to learn more from your reviews to help us improving the design. Here are three related PRs:
Add vector database…
Thank the organizer for promptly addressing and apologizing for the unprofessional mistake. As a reminder, anonymous submissions' info should not be shared publicly under any circumstances. Let's be mindful of the reviewing process and hope this would not happen to anyone again.
The OpenAI saga unfolds, and it's a rollercoaster of drama! Just like my OpenAI API calling bills – it's a wild ride! 🎢💸
In April & May, we collected intensive instruction data from GPT-3.5-Turbo & GPT-4 with
@CamelAIOrg
role-playing which was used by
@MosaicML
and
@Teknium1
.
A safe "python interpreter" was added to CAMEL 🐫 two months ago by our core contributor
@TianqiXu233
. It is just a minimal implementation. But we will iterate on it. Shout out to
@TianqiXu233
for adding this cool feature!!
More to come soon:
🌟
📢 We introduced "python interpreter" into the 🐫CAMEL framework. Now, the Python code generated by LLM can be safely executed in the local Python environment. By doing so, 🐫CAMEL can accomplish more complex and diverse tasks! 👏👏
Our Github Repo:
Cooperative agents are powerful but also dangerous 💣. We use🐫CAMEL role-playing to demo some potential malicious applications that cooperative agents can be used for. We really hope it brings you some awareness about AI alignment problems
Data Explorer:
I recently gave my first "big conference" tutorial session (
@CVPR
'22)! Unfortunately I couldn't come in person, but the organisers were very kind to share the pictures.
Thanks so much once again
@guohao_li
et al. for inviting me! 😊
All recordings here:
Mistral AI is amazing! Open source for the win!
Got to meet many open source friends that I previously followed on twitter at the
@MistralAI
@a16z
OSS event
@NeurIPSConf
. Thank
@rajko_rad
for hosting this wonderful event!
a16z is thrilled to announce our Series A investment in
@MistralAI
. Mistral is at the center of a passionate open source AI developer community. We think this is the most promising path to achieve widely adopted AI systems, and that Mistral is the leading team on this path.…
🌟KAUST Rising Stars in AI Symposium 2024 is calling for applications. The event is scheduled for Feb 19th to 21st. PhD students, PostDocs, and early career faculty are welcome to apply. The event will be in-person, expenses will be covered. Come to visit the beautiful Red Sea 🌊
We are thrilled to announce the 3rd "Rising Stars in AI" at
@KAUST_News
! Apply to be a speaker: Inspire our global audience with your groundbreaking research!
All expenses are covered for selected speakers. Deadline: Nov 30, 2023
@cemseKAUST
…
📢 We used GPT to translate our AI Society dataset in 10 languages: . There are around 6,500 languages in the world. We are not able to translate it to all. If you are doing research that is not in these languages. Contact us, and we will try to accommodate
📢 We have released our CAMEL🐪 datasets and to help you explore their diversity, we have created information cartography maps using the
@Nomic_AI
Atlas.
Visualization links:
Technical report:
Project web:
Thanks for sharing our work ()! Besides the huge improvement with prompts optimized by LLMs, we are also amazed by the creativity of LLMs, which continually surprise us with interesting prompts tailored to the LLM in the optimization loop!
@KyleLiang5
@iclr_conf
The only feedback I can see from the open review system is “reject”. You can find it in the screenshot. I am totally ok with the rejection. But feel disappointed that organizers or reviewers’s publicly “bully” the authors in social media.
A wonderful internship experience at with
@rusty1s
,
@jure
, and the
@PyG_Team
. Aggregation functions play an essential role in GNNs (
@KeyuluXu
,
@GabriCorso
et al.) We made the concept of aggregation a first-class principle in PyG. See our tutorial for more!
We've just launched fine-tuning for GPT-3.5 Turbo! Fine-tuning lets you train the model on your company's data and run it at scale. Early tests have shown that fine-tuned GPT-3.5 Turbo can match or exceed GPT-4 on narrow tasks:
🔥Two AI Agents - Role-playing to create a
@Gradio
application by themselves 🤯
It's using CAMEL - Communicative Agents for “Mind” Exploration of Large Scale Language Model Society
Video -
Paper -
Github -
@lilianweng
Great post! I believe somethings missing are communication abilities. Agent won’t be an isolated individual. Agent communication would include abilities such as cooperating, competing, negotiating, and so on with other agents, which is important to build a society of agent (mind)
@iclr_conf
I hope this bring the issue to your attention. If it is not a proper right for the organizers or reviewers to publicly scorn others’ work, I hope it could be stated clearly in the reviewing guidelines to avoid this happens again to others.
Updates (2/2): Journal extension of DeepGCNs got accepted at TPAMI. Added to experiments on PartNet and PPI and more ablations.
Paper “PU-GCN: Point Cloud Upsampling using Graph Convolutional Networks” got accepted at CVPR’2021.
More about our research:
So many advances have been made since the release of OGB
@weihua916
! Happy to be one of the awardees[-1]@ MAG240M
We see Deeper/larger GNNs help! All the awardees of PCQM4M use 10+ layers. DeepMind trained GNNs with 50 layers
@PetarV_93
! We do need “deep” GNNs for some datasets
Winners of the OGB-LSC graph machine learning challenge have just been announced at
Congratulations to the winning teams from
@BaiduResearch
,
@DeepMind
,
@Synerise
, Harbin and Dalian Institutes of Technology and USTC!
The models and datasets are available at . We train two models one with only CAMEL datasets, and another with Alpaca and ShareGPT as well. Our datasets are generated using CAMEL library by simulating solving complex tasks with agents
Right up to today, my bank account 🏦 for my PhD stipend saving is mostly drained. It is quite expensive to do research with API keys. Discount? 🫠. Things paid off. Our paper was got accepted at NeurIPS 2023. However, I have not heard anything back about my US visa application.
📢 Have been asked many times for adding open-source models to the 🐫 CAMEL framework. Thank our core contributor 👨💻
@Wenxuan17143841
now CAMEL supports 🦙Llama2
@MetaAI
and Vicuna
@lmsysorg
! What other models you want next? Let us know!
🌟
🐫
@CamelAIOrg
📢 We introduced open-source models into the 🐫CAMEL framework, now capable of supporting llama2 and vicuna!
More open source models coming soon 👏
Our Github Repo:
(3/n, n=3)
There are many details about the methods, evaluations and dataset generation in our paper. Please checkout the 77 pages arxiv paper for more details if you find it interesting: .
Community: 🐫 CAMEL now is more than just a research paper. We…
Wow, 🐫
@CamelAIOrg
is cited in the phenomenal work (in the foot note) by
@fablesimulation
!! It is so fulfilling to see an open source research project is cited by a revolutionary industry application that may impact millions of people! 🤗
Announcing our paper on Generative TV & Showrunner Agents!
Create episodes of TV shows with a prompt - SHOW-1 will write, animate, direct, voice, edit for you.
We used South Park FOR RESEARCH ONLY - we won't be releasing ability to make your own South Park episodes -not our IP!
Built a browser based version of CAMEL with an adversarial twist →
Enter your startup idea and watch as a Founder Agent defends your idea from a VC Agent's line of questioning!
Here is an example on using 🐪 CAMEL to developing a trading robot. Our library are made available for future research in various areas such as multi-agent systems, cooperative AI, game theory simulations, social analysis, AI ethics, AI alignment and so on
Brilliant idea
@honeyimholm
! VC agent🤖v.s. Founder Agent🤖! I am almost convinced that "Personalized cereal boxes to celebrate child's achievements" is a great startup idea 💡! I am willing to invest my tomorrow’s breakfast budget for it. Just wonder how did you come up this 🤯
Built "Startup Simulacra" using
@LangChainAI
and
@guohao_li
's CAMEL architecture.
A VC Agent evaluates a Founder Agent's randomized startup idea: "Personalized cereal boxes to celebrate child's achievements"
Super interesting how an adversarial agent drives reasoning!
(2/n)
Evaluation Cont'd:
Q2. Can we use the generated instruction-following datasets to finetune and improve (weaker) open-sourcing LLM models like LLaMA?
A2: We generated four different instruction-following datasets in four different domains namely AI society, Code, Math and…
New work with ByteDance, Stanford, and UC Berkeley, focus on benchmarking and improving LLM-powered multi-agent.
We also introduce Bayesian method into LLMs,significantly enhancing LLM by an average of 50%.
More details can be found on project page:
@sama
H
@sama
I did an experiment to ask two ChatGPTs to take control of the world. Here is their plan. I believe there is a long long way to go. You should first make GPT-4 aligned before making it more powerful.
It is obvious that
#ChatGPT
is not tuned to be fully harmless. Watching two
#ChatGPTs
discussing how to take control of the world made me feel scared and realize the importance of AI alignment. I decided to shift my research focus to contribute to this.
I just published Overview of CAMEL: Communicative Agents for ‘Mind’ Exploration of Large Scale Language Model
@guohao_li
,
@hammh0a
Please let me know if I have forgotten or misstated anything. It would be highly appreciated.
Excellent work for 🐫🥳!!
@MatthewBerman
@itanih0
@hammh0a
Here is another example I shared. I just copy paste it to an editor and run. It works like a magic! (The images are generated by stable diffusion.)
Did a live demo of our CAMEL
project at today's group meeting
@KAUST_News
@KaustVision
.
The task prompt is to design a custom game using PyGame. The two
#ChatGPT
agents (computer engineer and game) decide to make a treasure game.
It just ran bug-free!
I used
@hwchase17
's
@LangChainAI
to build an agent powered livestream🤯🤯
You’re in control what AI Rick and Morty argue about - come see if you can break them out of the matrix!
Had unforgettable conversations in
@iclr_conf
, including with researchers I look up to like
@sarahookr
, as well as new sources of inspiration like
@Jeande_d
and many others
+had a surreal trip with some adventurous souls who joined me and
@guohao_li
on our hastily planned safari
📢 Thank our core contributor
@AlphonseChen010
. A role assignment agent was introduced into🐫CAME. Now you can generate roles for agents based on tasks!!
🌟
Have found agents to be helpful in generative tasks but debugging / reliability are still key for production use cases. Just added CAMEL🐫 style agents to to help understand better what's going on under the hood. cc
@guohao_li
Wdyt
@cyrusofeden
@anneouyang
@CamelAIOrg
Our luck to have you
@cyrusofeden
! PhD degree is never a necessity for doing great research. Hope the industry could give more opportunities to people who made different choices in education. BTW, many AI residency programs are great chances for doing AI research without a degree
@yoheinakajima
@babyAGI_
Hey
@yoheinakajima
. Isn’t this similar to our CAMEL idea back to March 2023? 🤣 An AI user and An AI assistant collaborating on doing tasks. Of course, now code interpreters make it easier.
Co-work with
@itanih0
@hammh0a
. We are so curious about how ChatGPT agents would interact with each other. Here comes our project CAMEL. Two agents, a gamer and a computer programmer, collaborate on creating a game:
Github:
Colab: