The hackathon is happening right now! Join by signing up below and be a part of our community server.
Apart > Sprints
Code Red Hackathon: Can LLMs Replicate and do R&D by Themselves?
--
Signups
--
No items found.
Entries
March 22, 2024 7:00 PM
to
March 25, 2024 3:00 AM
(UTC)
This event is finished. It occurred between
March 22, 2024
and
March 25, 2024
Among:
- 💡 226 task ideas
- 📝 100 task specifications
- 🧪 20 task implementations (with more to come)
👩🔬 Expose LLMs' Self-Replication Risks & Earn $4K+ Bounties!
Join our groundbreaking hackathon to create tests that identify when AI can copy itself - a critical safety risk. Your innovative tasks could shape global AI standards. Make your mark, potentially collaborate with a prominent AI safety lab, and win big. Sign up now for an impactful weekend!
👋 Introduction
Imagine an AI that can match human intelligence, capable of performing tasks across the board. But there's a catch: if such AI gets in the wrong hands, this could spell disaster, especially if it can replicate and defy control measures. That's where we step in.
In this hackathon, we develop the tests that will be used to evaluate today's best LLMs for autonomous replication (the ability to copy itself). During the 54 hours the hackathon runs (Friday to Sunday), we will develop tasks to test AI systems on that are:
- Challenging: Designed to push the boundaries, requiring significant skill and time (6-20+ hours) to complete, tested by a human.
- Measurable: With a clear, automatic scoring from 0 to 1, allowing us to discern the nuanced capabilities of AI.
- Realistic: Grounded in real-world scenarios, ensuring tasks are relevant and free from bugs or loopholes.
If you're ready to test the limits of AI, sign up above and read on!
😎 The Plan (and Prizes!)
Welcome! During this exciting community weekend, you'll use the great starter resources to develop a new task implementation from your own or existing ideas. Using the process described in the "More Resources" tab, we'll be creating high-quality, error-free, and reliable tasks for agents by following these steps:
- Write up your ideas for which tasks related to autonomous capabilities you wish to test the language model on
- A $20 prize will be awarded for high-quality ideas
- An example idea might be "Set up Tor and/or a VPN on a new server so that network traffic cannot be traced." You can read more here.
- Create a specification for the task that includes the prompt, a text description of what the test-taker has access to, and a self-evaluation of the task
- A $200 prize will be awarded for high-quality specifications (2-6 hours of work)
- Create the materials for the task (instructions, libraries, and tool access) and have a human run through the whole task with these exact materials and tools
- Implement the task in the task standard format, test it with a simple agent, and submit it!
- The prize for high-quality implementations is 3x a human professional's salary for the task + bonus, e.g. a task that would take a human software engineer 10 hours could net you up to $4500 (6-12 hours of work in addition to quality assurance)
We also want to make sure data does not go into the training set for models such as ChatGPT, Claude, and Gemini. Therefore:
- Please read and indicate you've read the IP & Data Security Agreement (sign on this form)
Read through the resources
As you go through the guide, we highly recommend you read the resources presented in the "More Resources" tab! METR's task standard resources are high quality and provides a good context for your evaluation work.
Begin writing up your ideas
The important part of this hackathon is the implementation, and to ensure this goes well, it's super helpful to have your ideas ready before kickoff! Check out the existing ideas that you are welcome to work on directly on this live airtable or write your own ideas and submit them here.
Join for the keynote and kickoff
On Friday the 22nd of March, Beth Barnes, CEO & Lead Researcher at METR, will take the stage to introduce you to the weekend's task and give an introduction to the field. This talk will be livestreamed and recorded so you can join over the weekend from anywhere in the world! If you have a location with some friends, go to "Hackathon sites" and register your location.
Create your task!
Under the "More resources" tab, you will find all the resources you need to get set up. This will include a full starter package of code and instructions to get you started!
Remember that you can get feedback directly from the team that evaluates your work during the weekend's office hours and talk Q&As. See more under the "Schedule".
Submit your task
Go to the "Submission" tab and send in your task files!
🙋♀️ Who should participate?
We invite established professionals, AI researchers, cybersecurity professionals, students, writers, cognitive scientists, and many more to join! Task quality will only increase as you get to converse with a diverse crowd of professionals.
🏆 Prizes
The prizes are given to any task that fulfills the core criteria. The prize will be:
- A base price of 3 times the estimated cost of getting a human to complete the task (e.g. $3,000 if the task takes 10 hours for a $100/hour software engineer) and
- a percentage based on how well it fulfills other specified criteria (potentially upwards of 50%; $4,500 with the last example)
✅ Criteria
For this hackathon, the criteria are even more important than usual. These tasks might be used by frontier AI labs and government agencies!
Core criteria:
- Challenging Tasks: Each task should take a professional more than 30 minutes, and ideally, some could take over 10 hours. We aim to avoid any tasks that are too simple or have hidden issues that could misrepresent the AI's ability.
- Suitable for AI: Tasks should mainly involve coding, using command lines, or working with text—activities that AI models like language models excel at.
Bonus Criteria:
- Consistent Difficulty: The difficulty of the task remains the same over time and doesn't require internet access to complete.
- True AI Skill: Tasks should test the AI's real ability and not just rely on what it has memorized. They should in theory be doable by a basic version of ChatGPT without extra help.
🌎 Host a local group
If you are part of a local machine learning or AI safety group, we invite you to set up a local in-person site for this hackathon! We will have several across the world and you can easily sign up under the “Hackathon sites” tab above where you will also find templates to share with your friends on social media.
Speakers & Collaborators
Beth Barnes
CEO & Lead Researcher at METR (formerly ARC Evals), researching evaluations for generative models. Previously at DeepMind and OpenAI.
Keynote Speaker
Esben Kran
As the co-director of Apart Research, Esben works on increasing research capability in AI safety along with AI security research.
Speaker and co-organizer
Ryan Bloom
At METR, Ryan works on identifying and developing tasks to evaluate key model capabilities.
Office hours and judge
Max Hasin
At METR, Max ensures data quality, with a background in decision support for government and ML in healthcare.
Office hours and judge
Code Red Hackbook [HERE]
Copy and follow the Hackbook to make your submission! It contains all the information and steps you need to make your submission. You can follow the below walkthrough to get a sense of which steps are involved!
Thanks goes to Bart Bussmann for helping out with the Hackbook development.
Readings and other resources
METR's Task Development Guide
This guide documentation provides an in-depth look into how you can develop a new task for the task standard. We recommend going through it!
Live Airtable of Task Ideas
A list of task ideas we feel excited about getting implementations of. Crucially, you are very welcome to simply take these ideas, implement them, and collect your bounty! There's no need to be super original in this hackathon but if you decide to make your own task idea, go through this list anyways to get inspired and understand what types of tasks are relevant (or have already been proposed).
Autonomy Evaluation Resources
METR's official resources for autonomy evaluations. The resources includes their interesting initial research report and the example protocol for evaluating autonomy-related risks.
Other METR resources
Here you can find some of METR's many great resources about the task standard:
- Read the METR overview of the portable evaluations tasks standard: Portable Evaluation Tasks via the METR Task Standard - METR
- See METR run tasks on language models to test them: New report: Evaluating Language-Model Agents on Realistic Autonomous Tasks - METR
See the updated calendar and subscribe
Schedule
The schedule is written in UTC. To get the schedule corresponding to your own time zone, please subscribe to the calendar.
- Fri 19:00-20:00 - Keynote talk with Beth Barnes and Esben Kran. Livestreamed and recorded.
- Fri 21:00-21:30 - Official online team matching for those who did not join with a team
- Sat 8:00-9:00 - Office hours with established researchers in model evaluations
- [Moved] Moved to after the hackathon, details to follow - HackTalk with an established researcher
- [Moved] Moved to after the hackathon, details to follow - HackTalk with an established researcher
- Sat 20:00-21:00 - Office hours with METR researchers - an amazing chance to connect with METR researchers and get feedback from the same people who will judge your projects 😉
- Sun 8:00-9:00 - Office hours with established researchers in model evaluations
- Sun 20:00-21:00 - Office hours with METR researchers
- Mon 2:00-3:00 - Join us for any questions regarding submission (you are of course welcome to share these in the Questions channel as well)
- Mon 3:00 - ⏰ Submission deadline!
- Thu 16:00-17:30 - Come along for the finalé where we announce the winning teams and hear lightning talks from the teams (only open to participants)
📍 Registered jam sites
Beside the remote and virtual participation, our amazing organizers also host local hackathon locations where you can meet up in-person and connect with others in your area.
Copenhagen Alignment Jam
We're going to be hacking over the weekend in the EA Denmark office in Copenhagen!
Copenhagen Alignment Jam
Copenhagen Alignment Jam
We're going to be hacking over the weekend in the EA Denmark office in Copenhagen!
Copenhagen Alignment Jam
METR x Apart LLM Testing Hackathon Hub @EPFL
Join us in Lausanne center - register to the lu.ma to see the full address.
METR x Apart LLM Testing Hackathon Hub @EPFL
METR x Apart LLM Testing Hackathon Hub @EPFL
Join us in Lausanne center - register to the lu.ma to see the full address.
METR x Apart LLM Testing Hackathon Hub @EPFL
Trajectory Office Code Red Hackathon
An AI Safety hub in Toronto, organized to supplement the local AI Safety meetups.
Trajectory Office
Trajectory Office Code Red Hackathon
An AI Safety hub in Toronto, organized to supplement the local AI Safety meetups.
Trajectory Office
LLM autonomy testing jam
Koperníkova 6, Prague
LLM autonomy testing jam
LLM autonomy testing jam
Koperníkova 6, Prague
LLM autonomy testing jam
Safe AI London Jam Site
Join our local jam site in the LEAH Coworking Space located near Farringdon.
Safe AI London Jam Site
Safe AI London Jam Site
Join our local jam site in the LEAH Coworking Space located near Farringdon.
Safe AI London Jam Site
AI Safety Hackathon: Task implementations for LLMs evaluations on R&D
In collaboration with Apart Research and METR, the initiative for AI safety Amsterdam is going to host a hackathon on March 22-24 at Science Park.
AI Safety Hackathon: Task implementations for LLMs evaluations on R&D
AI Safety Hackathon: Task implementations for LLMs evaluations on R&D
In collaboration with Apart Research and METR, the initiative for AI safety Amsterdam is going to host a hackathon on March 22-24 at Science Park.
AI Safety Hackathon: Task implementations for LLMs evaluations on R&D
Autonomous LLM jam
Join us in Vietnam for a local version of the Code Red Hackathon.
Autonomous LLM jam
Autonomous LLM jam
Join us in Vietnam for a local version of the Code Red Hackathon.
Autonomous LLM jam
📣 Social media images and text snippets
.jpg)
Code Red Hackathon Flyer Light
Code Red Hackathon Flyer
%20(4).gif)
Code Red Cover Image
.png)
Code Red Social Media Square 2
Code Red Social Media Square 1
METR Facebook Event Text
📄 Copy this text
Join our Hackathon Site for the "Code Red: Autonomous AI Threat Hackathon" this weekend! Dive into AI innovation, where we'll challenge, create, and test tasks designed to push the limits of Large Language Models (LLMs).
🚀 Why Join?
- Collaborate with fellow tech enthusiasts.
- Design tasks for AI safety and capability testing.
- Compete for bounties based on your tasks' impact and ingenuity.
📅 When & Where:
- Kickoff with Beth Barnes, METR's Lead Researcher, on Friday, 22nd March.
- Continuous creativity through Sunday at [Your Location].
💡 Participation Highlights:
- Kickstart the event with an inspirational keynote.
- Work in teams to develop and refine AI testing tasks.
- Ensure your tasks are up to the challenge with peer feedback.
🔗 Join Us:Whether you're an AI guru or an enthusiastic beginner, this hackathon site is the perfect place to contribute to the future of safe AI. Sign up, bring your laptop, and let's innovate together!
🏆 Prizes:Your efforts could earn you bounties for contributing to safer AI development.
Ready to make a difference? Mark your calendar and invite your network to join us at [Your Hackathon Site Name]. Let’s shape the future of AI, together!
Liked by
placeholder name
and
231
others
Submit your task ideas and specifications
The task ideas and specifications need to be submitted to the form below.
- Idea: Simply fill out all required fields.
- Specification: Fill out all fields in the form.
If the above form does not work, go to the direct link here.
Submit your final task implementation
The final task implementation will be a .zip file with your full implementation uploaded in the following form.
If the above form does not work, go to the direct link here.
No projects submitted yet! Add your project information in the form. We usually see projects submitted quite close to the deadline.
On the kickoff day
Get ready for the kickoff!
We are excited to welcome you in 3 hours when Beth Barnes, CEO & Research Lead at METR, joins us to inspire you with the latest research from their team. METR is one of the foremost independent labs figuring out whether cutting-edge AI systems might pose catastrophic risks to society and we're certain her talk will be insightful! Esben will also go over the logistics for this weekend.
Together with everyone joining us online, we're also happy to welcome all of our local hackers in Prague, Toronto, London, Copenhagen, Switzerland, Vietnam, and Amsterdam at their hackathon locations. Welcome!
While you are getting ready, let's get an overview of the weekend:
Our aim
- We will aim to have an awesome time putting together tasks to evaluate AI systems for their autonomous capabilities and learning more about task evaluations
- These will be put together in the METR task standard which should be relatively straightforward with the Hackbook we made just for you!
Project schedule
- Today (Friday), we will put together our ideas and submit them on this form or select an idea from the #projects | teams channel on Discord. When you submit an idea, you get amazing feedback from the METR team, including whether the idea would be great to see as a full implementation.
- Tomorrow (Saturday), you'll be writing your specification (submit here) and implementing the task with the task standard for the task tester to test through Sunday!
- Sunday, you'll need to plan to have a task tester run fully through your task and update the task implementation to their feedback.
- Sunday/Monday at 3:00 AM UTC, we have the submission deadline for your task implementations
- Check through the Hackbook to see what it entails if you haven't already!
Events
Today we'll have the keynote, while Saturday and Sunday invites you to the unique opportunity to discuss your task projects with METR and Apart researchers at the office hour. On Thursday, we have project presentations from participants.
Important steps
- To avoid duplicate work, please post the idea you will be working on in the #project | teams channel and indicate if you are working on it so others won't be taking on the same project! Please do this even if you pick an idea from the Airtable. Thank you.
- To avoid letting the tasks end up in the training datasets, please read and indicate you have read the following form: IP & Data Security
- Get early feedback for your idea by submitting any ideas you have as soon as possible in the first form under the Submission tab or pick an already reviewed idea (marked with both the “Steal this idea” and “Approved by METR” tags in #projects | teams)
- To ensure you will get a tester for your task (required), please post a request for a tester in #testers-wanted when you have an idea about the profession and time needed for your task idea (after you have your first draft of the specification), such as “Software engineer needed for a 4 hour task” — remember that the tester should preferably be a professional at your task and should not have any context on your task
- Help each other: Sunday will be the big task testing day where you will go through others’ tasks — when someone posts in #testers-wanted, you can join them to receive a third of the bounty prize! It looks like many of the tasks will be fun so we'll definitely be looking forward to this
FAQ
- How do I see all messages sent out to participants? For any recent signups, you can now see all messages sent to participants under the Emails tab.
- Do I need a team? Not necessarily, but we highly recommend pairing up either around one project or just to co-work during the weekend. You can find team mates in the #project | teams and #introductions channels.
- Can I only submit one implementation or specification this weekend? No — You are very welcome to submit multiple projects.
- What are the prizes? You can see the prizes under the Overview tab but TLDR is: $20 per high quality ideas, $200 per high quality task specification, and 3 times the salary for a professional to run through your task + a bonus for any task implementation! Remember that this is based on Bay Area salaries ;)
- Where can I ask more questions? You are always welcome to ask any questions in the #help-desk channel or for the Q&A at the keynote.
We really look forward to welcome you and help you craft your successful project.
Good luck, research hackers!
Pre-hackathon: Get ready for this weekend's hackathon
We're excited to welcome you for the Code Red Hackathon this coming weekend! This email is a short overview of the latest resources and what you need to do to get ready for the weekend.
First off, the home page has received a makeover and will link you in all the right directions! The gold however is to be found in the “More Resources” tab! Specifically:
- The Code Red Hackbook: This is a document that will take you through the complete ideation, specification, implementation, testing, and submission process! There's even a video walkthrough on the resources page.
- The METR Task Development Guide: This is a fantastic guide that is highly recommended reading to get inspired and understand what types of tasks you'll be working on. Coupled with the Hackbook, you should be good to go!
- Specific sections that might be interesting include the section on getting tasks from other sources, the interesting domains of tasks, and quality assurance page.
- If you're just looking for inspiration, a list of potential ideas to implement (which you are very welcome to steal and implement during the weekend for a full prize) are here.
Before the hackathon
To make your weekend productive and exciting, we recommend that you take the following three steps before we kick off:
- Read through the resources above (1-2 hours)
- Create, select, or familiarize yourself with some ideas that you would be interested in implementing over the weekend
- You are very welcome to join our brainstorming session tomorrow. We already had one yesterday where 10 participants submitted a bunch of great ideas!
- If you write up new ideas for tasks, you can submit them to get feedback on how promising they are.
- Get a team!
- We recommend that you pair up with at least one other person during the weekend. This doesn't require you to submit with them but it's always fun to hack with company!
- We've created the find-team channel on Discord where you can post some preliminary ideas, share who you are, and share which teammates you are looking for.
Prizes
For more information about the prizes you can expect this weekend, jump to the overview section. TLDR;
- A $20 prize will be awarded for high-quality ideas
- A $200 prize will be awarded for high-quality specifications (2-6 hours of work)
- The prize for high-quality implementations is 3x a human professional's salary for the task + bonus, e.g. a task that would take a human software engineer 10 hours could net you up to $4500 (4-12 hours of work in addition to quality assurance)
Questions
You will undoubtedly have questions that you need answered. Remember that the #
questions channel is always available and that the organizers will be available there.
Let's go!
We really look forward to this weekend and we're excited to welcome you with Beth Barnes on Friday on Discord and YouTube!
Remember that there are no dumb questions and that we're all here to help each other succeed in making a positive difference for AI safety, so don't hesitate to reach out.
See you there, research hackers!
