Judge feedback

Judge feedback is anonymous and shown in a random order.

• I am very interested in the issue of AI art, and I was excited to see you tackle it. It was particularly interesting to me to see confirmed that AI art has been a catalyst of negative sentiment, and also - and more surprisingly - that the movement spiked but quickly receded. I think this was a good first attempt, but hardly polished - the article's scope starts too broad, it is not clearly connected to strategic issues, the analysis is more speculative than well supported and it neglects many important questions such as what prompted the decline in interest or what drove the positive sentiment in AI art during the period. There is also little engagement with the content that drove the negative sentiment, and the measures that they are likely to lobby for in the coming months. If I was judging this as a work trial for Epoch, it would rank close to the 50% percentile - better than most we receive, but insufficient to advance to the next stage of the hiring process.
• Why I think is this project is strategically important: If we had good forecasts of 1) how public sentiment will change in the future and 2) how this will impact regulation, it would be extremely useful for deciding which policies to research and advocate. Why I think this project does not have high information value: The data on public sentiment is interesting and informative, as is the fact that AI art is the main driver of the recent negative sentiment, and that the timing did not match the first releases of AI art tools. However, the analysis of the causes of this change in sentiment was not entirely convincing to me. A lot of the proposed underlying factors seem shared with ChatGPT (it can also produce art, like poetry, and it is also not well understood by the public) Similarly, the predictions on policy effects seem too coarse-grained to be actionable. The main policy option discussed is banning this technology, especially in the context of China. Why I think this project is novel: I personally have not seen any analysis of public sentiment about AI in the context of AI risk. In addition, the basic facts about sentiment are somewhat surprising (eg: I didn't expect negative sentiment to subside so quickly). Some things that I miss: A section on next steps, or how this analysis could be improved with more time and effort. Do you think this methodology could allow us to produce more precise predictions, especially about the possible influence on policy?