Hide table of contents

Sample Article

One of the exciting things about the future is that we’ll be able to accelerate our own flourishing and the flourishing of the world. But this will only become possible if we do the work needed to make ourselves more agentic. I think we’re already very close to being able to do this. But to become unstoppably agentic we need to make a series of small, non-trivial changes to ourselves. I’ve tried to capture these in the following list. I’ve also added some of my own insights into how to make such adjustments, as well as a few other thoughts.

I’m interested in this topic, so I’d love to hear your thoughts!

* * *

1. Resist the temptation to be the smartest person in the room

Many of us have this sense that if we can just figure out the answer, connect the dots, we can get whatever we want. The truth is that when it comes to some of the most important questions, we don’t know what the answer is. This isn’t an excuse for laziness, it’s an opportunity. The world is full of opportunities for us to learn, grow, and contribute. For example, I’m not sure of the answer to climate change, but I’m sure that the people who know it best need my help. Similarly, I don’t know the answers to most of the problems we’re trying to solve, but I’m sure that there are people who do, and that I can help them.

2. Look for people who are acting from a place of humility and curiosity

We can tell when someone is thinking about a problem from a place of humility and curiosity. If they’re not, they’re probably going to be wrong. And if they are, they’re probably onto something. This is true even if what they’re talking about is outside of their area of expertise. I’d love to be able to do more of this, but I often feel like I can’t. I can only be so curious and humble, and when I’m doing my own work I can’t seem to accommodate much more of either of those. But I’m excited to see people who are able to do this. In particular, I’m excited to see people who are able to ask the questions that seem obvious to me, but that others have been less excited about.

3. Slow down

This is an attitude, not a specific action. You might think I’m saying ‘don’t speed’. I’m not. I mean that we should slow down our pace of life, and take the time to really think about what we’re doing, to think deeper and more thoroughly. For example, maybe we should slow down when we’re trying to make a decision. Maybe we could do more to carefully weigh the pros and cons. Maybe we could take the time to get better at reading other people. We should all think about this. But I think it’s more important for those of us who do a lot of thinking.

4. Seek out opportunities to be a little weird

I’m not sure how to define weirdness, but I know that it’s something we should seek out. It’s something that can be hard to find. And it’s something that can be appreciated when you find it. Weirdness is a bit of a necessary evil. We want to be able to create value for ourselves and for the world. And to do that we need to be able to ask questions that are uncomfortable, or explore ideas that are unfamiliar. And weirdness can be uncomfortable. But it can also be that one of the best ways to develop new ideas is to look at ideas from another perspective. I’m excited to see people who are comfortable with weirdness, and who look for opportunities to explore it.

5. Try new things

We should all try new things. I’m excited by the people who are able to try new things, and then share what they learned, either by writing or through sharing their findings with others.

I’m also excited by the people who are able to share their findings with others. I’m excited by the people who are able to show others how to try new things.

6. Acknowledge the power of routine

This is a tough one.

We often think that there are big gains to be made by doing things differently. But there is some truth in the idea that we’re not going to be able to meaningfully improve without trying to improve our routines. For example, I think that if we try to change one thing at a time, we’re much more likely to be able to maintain those changes. This may not always be true, but I think it’s worth trying.

7. Be generous

I’ve spent a lot of time around people who are generous, and I’ve seen a lot of the impact that this can have. It’s not just about money. I’ve seen the impact that generosity has on people’s engagement with life. And I’ve seen how people who are generous inspire others. I’m excited to see more people who are able to be generous.

8. Ask for help

I’ve seen the power of asking for help. Because of this, I advocate for the power of asking for help. I’ve worked with a lot of people who don’t ask for help, and I’ve seen how their work suffers. I’ve seen the power of asking for help, and I’m excited to see more people who are willing to ask for help.

9. Hold yourself to high standards

This is a tough one.

It can be easy to set low standards for ourselves.

It can be easy to judge ourselves harshly, and to think that we can’t improve

ourselves.

But it’s important to hold ourselves to high standards.

It’s important to push ourselves to be better than we were.

It’s important to not be satisfied with just a little bit better than we were.

And it’s especially important to do this when we’re working on ourselves.

It’s easy to get stuck in patterns of behavior, and to hold ourselves to those patterns.

But it’s good to keep pushing ourselves to be better, even when it’s hard.

10. Look for the silver lining

I’ve seen the power of this in a few different areas.

I’ve seen it with people who are facing difficult times in their life.

I’ve seen it with people who are facing difficult times in the world.

I’ve seen it with people who are facing difficult times in their work.

And I’ve seen it with people who are facing difficult times in their personal

relationships.

I’m excited to see more people who are able to look for the silver lining.

Technical Information

GPT-EA-Forum-v1 is a very early prototype of GPT-EA. GPT-EA is a project to create a descriptive large language model reflecting the effective altruism community. Posts were randomly sampled in proportion to their karma, with each post being sampled at most once with no duplicates, resulting in 3000 total posts. The GraphQL API was used. This API appears to only return 8000 posts, compared to the estimated 15,000 that exist. A future iteration will use jacquesthibs' scrape of the Effective Altruism Forum and include more than 3000 posts.

The base model was GPT-NeoX-20B by EleutherAI trained on The Pile. Finetuning was done for 3 epochs.

The prompt the article below was given was: "Seven ways to become unstoppably agentic," as shamelessly stolen from this article. Aaron Gelter's name, signature, and email address were removed to avoid confusion. The model's output is otherwise completely unedited. The sampling settings used were temperature=0.7, top_k=1000.

Conclusion

This early prototype appears capable of generating plausible, coherent text in the prosaic style of an effective altruist. I'm excited to see what comes next as more forum posts, books and websites (OpenPhil.org, GivingWhatWeCan.org, potentially RethinkPriorities, potentially Center for Long-Term Risk, potentially GiveWell blog) are added to the training mix.

24

0
0

Reactions

0
0

More posts like this

Comments8
Sorted by Click to highlight new comments since:

Nice work, but I wonder at the consequences. Sure, it's inevitable that forums like ours will soon be bot-ridden and will need to bring in clunky authentication features, but aren't you speeding that up?

Are you trying to spook people, or just playing, or trying to give us a tool to make writing easier? The last two of those seem not worth it to me in expectation.

One goal is to make it easier to understand Effective Altruism through an interactive model.

I'm sick with COVID right now. I might respond in greater depth when I'm not sick.

This model performance is really impressive, and I'm glad you're interested in large language models. But I share some of Gavin's concerns, and I think it would be a great use of your time to write up a full theory of impact for this project. You could share it, get some feedback, and think about how to make this the most impactful while reducing risks of harm. 

One popular argument for short-term risks from advanced AI are the risks from AI persuasion. Beth Barnes has a great writeup, as does Daniel Kokotajlo. The most succinct case I can make is that the internet is already full of bots, they spread all kinds of harmful misinformation, they reduce trust and increase divisiveness, and we shouldn't be playing around with more advanced bots without seriously considering the possible consequences. 

I don't think anybody would make the argument that this project is literally an existential threat to humanity, but that shouldn't be the bar. Just as much as you need the technical skills of LLM training and the creativity and drive to pursue your ideas, you need to be able to faithfully and diligently evaluate the impact of your projects. I haven't thought about it nearly enough to say the final word on the project's impact, but before you keep publishing results, I would suggest spending some time to think and write about your impact. 

As a countervailing perspective, Dan Hendrycks thinks that it would be valuable to have automated moral philosophy research assistance to "help us reduce risks of value lock-in by improving our moral precedents earlier rather than later" (though I don't know if he would endorse this project). Likewise, some AI alignment researchers think it would be valuable to have automated assistance with AI alignment research. If EAs could write a nice EA Forum post just by giving GPT-EA-Forum a nice prompt and revising the resulting post, that could help EAs save time and explore a broader space of research directions. Still, I think some risks are:

  • This bot would write content similar to what the EA Forum has already written, rather than advancing EA philosophy
  • The content produced is less likely to be well-reasoned, lowering the quality of content on the EA Forum

This advice totally applies here: https://forum.effectivealtruism.org/posts/KFMMRyk6sTFReaWjs/you-don-t-have-to-respond-to-every-comment

Good luck with your projects, hope you’re feeling better soon.

Wow haha this is pretty cool! And also an entertaining read

Wow. Fine tuning a GPT-J model, especially the 20B, is much much, more work than running GPT-3.

I’m curious if you did this by grabbing the model and rolling it on the cloud, on some local GPUs (really, really expensive?) or used a service and never touched the model directly? All of these seem at least a little tricky.

Uh, feel free to not answer until you are much better.

Curated and popular this week
Relevant opportunities