This is a second kind of obvious point to make, but if you are interested in AI, AI safety, or cognition in general, it is likely worth going to top ML conferences, such as NeurIPS, ICML or ICLR. In this post I cover some reasons why, and some anecdotal stories.

1. Parts of AI alignment and safety are now completely mainstream

Looking at the "Best paper awards" at ICML,  you'll find these safety-relevant or alignment-relevant papers:

• Stealing part of a production language model by Carlini et al.
• Probabilistic Inference in Language Models via Twisted Sequential Monte Carlo by Zhao et al.
• Debating with More Persuasive LLMs Leads to More Truthful Answers by Khan et al.
• Genie: Generative Interactive Environments Bruce et al.

which amounts to about one-third (!). "Because of safety concerns" is part of the motivation for hundreds of papers. 

While the signal-to-noise ratio is even worse than on LessWrong, in total, the amount  you can learn is higher - my personal guess is there is maybe 2-3x as much prosaic AI safety relevant work at conferences than what you get by just following LessWrong, Alignment Forum and safety-oriented communication channels.

2. Conferences are an efficient way how to screen general ML research without spending a lot of time on X

Almost all papers are presented in the form of posters. In case of a big conference, this usually means many thousands of posters presented in huge poster sessions. 

My routine for engaging with this firehose of papers:

1. For each session, read all the titles. Usually, this prunes it by a factor of ten (i.e. from 600 papers to 60).
2. Read the abstracts. Prune it to things which I haven't noticed before and seem relevant. For me, this is usually by a factor of ~3-5.
3. Visit the posters. Posters with paper authors present are actually a highly efficient way how to digest research:
   • Sometimes, you suspect there is some assumption or choice hidden somewhere making the result approximately irrelevant - just asking can often resolve this in a matter of tens of seconds.
   • Posters themselves don't undergo peer review which makes the communication more honest, with less hedging.
   • Usually authors of a paper know significantly more about the problem than what's in the paper, and you can learn more about negative results, obstacles, or directions people are excited about.

Clear disadvantage of conferences is the time lag; by the time they are presented, some of the main results are old and well known, but in my view a lot of the value is the long tail of results which are sometimes very useful, but not attention grabbing.

3. ML research community as a control group

My vague impression is that in conceptual research, mainstream ML research lags behind LW/AI safety community by something between 1 to 5 years, rediscovering topics discussed here. Some examples:

• ICML poster & oral presentation The Platonic Representation Hypothesis is an independent version of Natural abstractions discussed here for about 4 years.
• A Roadmap to Pluralistic Alignment deals with Self-unalignment problem and  Coherent extrapolated volition
• Plenty of research on safety protocols like debate, IDA,...

Prior work published in the LW/AI safety community is almost never cited or acknowledged - in some cases because it is more convenient to claim the topic is completely novel, but I suspect in many cases researchers are genuinely not aware of the existing work, which makes their contribution a useful control: if someone starts thinking about these topics, unaware of the thousands hours spent on them by dozens of people, what will they arrive at?

 4. What 'experts' think

ML research community is the intellectual home of many people expressing public opinions about AI risk. In my view, background in technical ML alone actually does not give you significantly more expertise in understanding AI risk than let's say background in mountaineering methodology or theoretical evolutionary biology, but it is natural for the public to assume it does. This makes it useful to understand prevailing opinions and the broad epistemic landscape of the ML community.

As an anecdote, only after going to NeurIPS I fully realized how many researcher in NLP suffer from some internal conflict where part of them is really excited about AIs actually getting intelligent, but another part deeply hates this is largely due to scaling, with a place like OpenAI in the lead. 

 5. Examples

If the previous points haven't convinced you, here are five papers I discovered at conferences which I learned something from, but were not linked or noticed here and I would likely miss them while not visiting a conference.

• Learning Universal Predictors explores what happens if you take ideas from Solomonoff Induction and train actual neural network architectures like transformers on data generated from Universal Turing Machines. 
• Are Vision Transformers More Data Hungry Than Newborn Visual Systems? attempts to make a fair comparison of learning efficiency between newborn chicks and vision transformers, by training the transformers on first-person visual data similar to what newborn chicks experience.
• Prompting a Pretrained Transformer Can Be a Universal approximator asks if one can arbitrarily modify the behavior of a pretrained model by prompting or prefix-tuning it. Formally, whether prompting and prefix-tuning a pretrained model can universally approximate sequence-to-sequence functions.
• Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models provides nice theory for the intuition that robust watermarking is likely impossible
• Human Alignment of Large Language Models through Online Preference Optimisation explains a deep link between some recent alignment methods and  finding the Nash equilibrium of the preference model through self-play.

In each case, if someone re-wrote this is a LW post, I would expect it to be highly upvoted and read.

 6. Conclusion

In my view, if you tend to follow AI, AI safety or 'cognition in general' topics on safety community platforms, it is likely worth your time to go to a conference. If you don't go in person, you can still do some of of the described steps - skim titles, select abstracts, discover new things.

I would also be in favor of work that makes the community boundaries more permeable. In one direction, by converting some LW posts into conference papers - in particular, pieces explaining conceptual shortcomings and limits of safety methods people are likely to arrive. In the other direction, by distilling what's relevant but not safety-branded.

ACS would probably be happy to sponsor conference participation (like, tickets and travel) for someone in exchange for distillation work with regard to topics we are interested in - i.e. going through the abstracts, engaging with the papers, writing blogpost summaries of relevant research.