wordcloud: randomized social choice—causal modeling—string diagrams—AI strategy «cancel heat death» Research Fellow @FHIOxford + R&D Strategy @protocollabs

Austin, TX
Joined July 2008
welcome to nKindergarten! today we are learning about shapes
5
25
3
178
Show this thread
davidad 🎇 retweeted
Replying to @SrdjanMiletic1
I think of moving towards moral realism as the process of deriving more and more moral conclusions from fewer and fewer premises (even if we're never able to derive them from nothing). See this blog post for more:
Replying to @RichardMCNgo
For more details, here's a blog post I wrote on the topic (will probably do a thread on it later). lesswrong.com/posts/LeXhzj7m…
Show this thread
1
2
5
sometimes people talk about a difference between “science the good process for obtaining knowledge” and “science the contemporary cultural institution”; one aspect of this difference is that only in the former does it sometimes make sense to be your own test subject
Here's a list of nobel prize winners who experimented on themselves, something which is now highly taboo in the medical research community.
Show this thread
1
1
10
davidad 🎇 retweeted
Replying to @davidad @rhaegar227
to be fair, “pretty please uwu” is obviously a straw-man attempt; the design space of prompts for normative alignment is probably worth exploring, especially for teams like @AnthropicAI that are already betting on LLMs
1
1
6
timeline in which the final breakthrough that precipitates AGI could plausibly be literally some one-sentence platitude about general problem-solving
Large Language Models are Zero-Shot Reasoners Simply adding “Let’s think step by step” before each answer increases the accuracy on MultiArith from 17.7% to 78.7% and GSM8K from 10.4% to 40.7% with GPT-3. arxiv.org/abs/2205.11916
Show this thread
23
66
9
504
hard-takeoff vibes
3
11
313
thread of experiments with different operationalisations of "responsive cohesion" via mutual information, by optimising bitmaps for them and seeing what happens
1
1
1
6
224
"formally" has experienced a similar kind of semantic inflation within CS as "literally" in everyday English. unfortunately, this leads to difficulty in properly denoting the stronger concepts of a literal quote or a formal proof.
I *really* dislike the way many cryptography papers use the term "formally prove" to mean "wrote an argument down in words with some math thrown in"
3
1
11
clearly, they should instead be asking for π₀(_ : ∃n. 1 + 2 = n)
Replying to @kurt2001 @davidad
Or in elementary when the use = as an operator. 1 + 2 = _ I tell my kids that the answer to this question is false.
4
Has anyone else noticed that Beethoven Piano Sonata No. 3 appears to contain a video game sound effect? It really subjectively stands out every time I listen to the Beethoven piano sonatas. First instance is at 2:39 here: piped.kavin.rocks/ThvKrhqh3fE?t=2m38s Compare: piped.kavin.rocks/3BsBXp6VkvU
1
1
6