I tried the Anthropic model on this dataset with roughly your prompt and it's much better in terms of KL divergence between its predictions and Manifold probabilities. Giving it 10 web search results in a prompt further improves the performance. But the difference no search -> search is smaller compared to GPT-3 -> Anthropic, I'd say mainly because of unhelpful search results.
I think it's not free books per se, but free books related to phrases "here's what's really important", "this is how to think about morality" that are problematic in the context of the Bible comparison
I tried the Anthropic model on this dataset with roughly your prompt and it's much better in terms of KL divergence between its predictions and Manifold probabilities. Giving it 10 web search results in a prompt further improves the performance. But the difference no search -> search is smaller compared to GPT-3 -> Anthropic, I'd say mainly because of unhelpful search results.
Nice visualization, I wonder how it's doing now? Curious about any post-mortem comments.
I think it's not free books per se, but free books related to phrases "here's what's really important", "this is how to think about morality" that are problematic in the context of the Bible comparison
Some people have been saying that Starlink system's limit is 0.5M consumers even after they release a whole lot more satellites: https://www.techdirt.com/articles/20200928/09175145397/report-notes-musks-starlink-wont-have-capacity-to-truly-disrupt-us-telecom.shtml. This would mean you can't expect it to turn even 0.1% of unconnected people to netizens.