I. Like most people, my teens and twenties have been confusing and not always the most fun. I’ve struggled to make friends. In high school and university, I didn’t have as many romantic relationships as I wanted. When I was 24, I met a beautiful, wonderful woman who became my...
I reproduced results from ARC Evals' recent report, Evaluating Language-Model Agents on Realistic Autonomous Tasks. For the report, ARC Evals built a set of language model agents, combining a language model like GPT-4 with scaffolding software that lets the language model execute shell commands and interact with a browser. Then,...
Like most people, my teens and twenties have been confusing and not always the most fun. I’ve struggled to make friends. In high school and university, I didn’t have as many romantic relationships as I wanted. When I was 24, I met a beautiful, wonderful woman who became my wife, but I still feel like I have a lot of room to be a better husband. I lucked into a relatively stable and interesting career, but my day-to-day experience with work has involved a lot of emotional swings and occasional disillusionment. In general, I’ve struggled to feel consistently happy.
I haven’t really figured out what to do about all this. I’ve thought about... (read 2852 more words →)
Hey, I'm Thomas. I just moved from Ontario, Canada to Burbank. I'm planning to run ~weekly rationality meetups in the San Fernando Valley in Los Angeles. I invite folks from Burbank, Glendale, Pasadena, North Hollywood, the rest of the SFV, and LA at large to attend.
Let's meet in the common area at my apartment complex. If you'd like to swim, feel free to bring a bathing suit and a towel and use the pool or hot tub.
There's street parking on the residential streets north of Empire Ave and west of Buena Vista St. You can also park in my building's garage for $7 -- please let me know beforehand if you'd like to do that.
You can contact me at htbroadley@outlook.com or on Discord at newwwworld.
Hey, I'm Thomas. I just moved from Ontario, Canada to Burbank. I'm planning to run ~weekly rationality meetups in the San Fernando Valley in Los Angeles. I invite folks from Burbank, Glendale, Pasadena, North Hollywood, the rest of the SFV, and LA at large to attend.
Let's meet in the common area at my apartment complex. If you'd like to swim, feel free to bring a bathing suit and a towel and use the pool or hot tub.
There's street parking on the residential streets north of Empire Ave and west of Buena Vista St. You can also park in my building's garage for $7 -- please let me know beforehand if you'd like to do that.
You can contact me at htbroadley@outlook.com.
I have arrived!
Come to the first rationalist meetup in the San Fernando Valley! (That I could find evidence of online :P ) I invite folks from Burbank, Glendale, Pasadena, the rest of the SFV, and LA at large to attend.
Conversation starter: "Does Capitalism Beat Charity?" by Scott Alexander. Please read the article and, if you have extra time, feel free to browse the comments or read the associated "Highlights from the Comments" post.
Let's meet in the Burbank Town Center food court. I'll be sitting at a table near the Panda Express. I'll have a sign with me. Google Maps link: https://maps.app.goo.gl/CWQ8Afhsgx3RtrGE9
I neglected to update my comment here -- the agent I built for this replication is now publicly available as part of the METR task workbench, here: https://drive.google.com/drive/folders/1-m1y0_Akunqq5AWcFoEH2_-BeKwsodPf
That's me on the bass! Thank you for hosting, it was really fun to jam with everyone.
Yeah, I definitely could! It's on my to-do list. I'll let you know when I complete it.
Thank you! No, I'm not building custom prompts for the different tasks. I wrote a single prompt template -- the only difference between runs is the task description, which gets plugged into the template. I think ARC Evals did the same thing.
I have been improving the prompt as I worked through the tasks. I probably spent 2-3 hours working on the prompt to try and improve the agent's performance on some tasks. I'll definitely rerun all the tasks with the current version of my prompt, just to check that it can still perform the easier tasks.
You're right that getting the agent to attempt the last three tasks is relatively simple. Still, I was thinking that it wasn't worth the time or money. I think it's very unlikely that the agent will succeed at any of the last three tasks. Still, maybe it's worth getting a conclusive negative result.
Thank you for the kind comment! You have lots of good ideas for how to improve this. I especially like the idea of testing with different cloud providers. I could add programming languages in there: Maybe GPT-4 is better at writing Node.JS than Python (the language I prompted it to use).
I agree, a fully reproducible version would have benefits. Differences in prompt quality between evaluations is a problem.
Also agreed that it's important to allow the agent to try and complete the tasks without assistance. I did that for this reproduction. The only changes I made to the agent's commands were to restrict it to accessing files in a particular directory on my... (read more)
EDIT: The agent I built for this replication is now publicly available as part of the METR task workbench: https://drive.google.com/drive/folders/1-m1y0_Akunqq5AWcFoEH2_-BeKwsodPf
I'm torn! I think that better LLM scaffolding accelerates capabilities as much as it accelerates alignment. On the other hand, a programmer (or a non-programmer with help from ChatGPT) could easily reproduce my current scaffolding code. Maybe open-sourcing the current state of the project is fine. What do you think?
I reproduced results from ARC Evals' recent report, Evaluating Language-Model Agents on Realistic Autonomous Tasks. For the report, ARC Evals built a set of language model agents, combining a language model like GPT-4 with scaffolding software that lets the language model execute shell commands and interact with a browser. Then, it asked the agents to complete a set of 12 computer-related tasks, from searching Wikipedia for information to conducting a phishing campaign. The goal is to test how close the agents are to being able to make money, obtain computing power, make copies of themselves, and adapt to changes in their environment.
To reproduce these results, I wrote my own language model agent.... (read 889 more words →)
since private goods are non-rival it is efficient to exclude consumers who aren't willing to pay
Should this be, "since private goods are rival it is efficient..."?
Here is a submission: https://ai-safety-conversational-agent.thomasbroadley.com
Source code here: https://github.com/tbroadley/ai-safety-conversational-agent
I followed @Max H's suggestion of using chat-langchain. To start, I created an embedding based on the articles from https://aisafety.info and have the submission using that embedding.
I'll get in touch with Stampy about working on their conversational agent.
Sorry, I'm having trouble understanding. If you'd like to give more detail, I'd appreciate it! Are you saying it's true that these views/concepts aren't resolvable? Or that it takes a lifetime of work to resolve them?
It also sounds like you're saying that, even after reaching enlightenment, you'll still have mental habits that become maladaptive over time. That's interesting, it wasn't my impression of what enlightenment was like.