All of corruptedCatapillar's Comments + Replies

"The beginning of wisdom is the definition of terms." - (attributed to) Socrates

I'd argue that for any conversation between people to make progress, they have to have some agreeance on what they're talking about. A counter example, if I were to order a 1/4" bolt from someone and we have different measurement standards of what 1/4" is, I won't be able to build on what I've received from them. Consistency is the key characteristic that allows interoperability between multiple parties.

I appreciate greatly that Gwern uses confidence tags to convey this aspect,... (read more)

1eggsyntax
Oh, I hadn't seen Gwern's confidence tags. There's an interesting difference between the Kesselman list that Gwern draws from (screenshotted below) and my proposal above. The Kesselman terms refer to non-overlapping, roughly equal ranges, whereas the climatology usage I give above describe overlapping ranges of different sizes, with one end of the range at 0 or 100%.  Both are certainly useful. I think I'd personally lean toward the climatology version as the better one to adopt, especially for talking about risk; I think in general researchers are less likely to want to say that something is 56-70% likely than that it is 66-100% likely, ie that it's at least 66% likely. If I were trying to identify a relatively narrow specific range, I think I'd be more inclined to just give an actual approximate number. The climatology version, it seems to me is more primarily for indicating a level of uncertainty. When saying something is 'likely' (66-100%) vs 'very likely', in both cases you're saying that you think something will probably happen, but you're saying different things about your confidence level. (@gwern I'd love to get your take on that question as well, since you settled on the Kesselman list) I've just posted a quick take getting a bit further into the range of possible things that probabilities can mean; our ways of communicating about probabilities and probability ranges seem hideously inadequate to me.  

Following up on this; do you have these notebooks available?

1jain18ayush
This repository has some of the code: https://github.com/er537/whisper_interpretability/tree/master. I do not know if you can rerun the same experiments but it has a lot of useful tools for analysis. 

Possibly of interest: https://arxiv.org/abs/2403.14380

"We found that participants who debated GPT-4 with access to their personal information had 81.7% (p < 0.01; N = 820 unique participants) higher odds of increased agreement with their opponents compared to partici- pants who debated humans. Without personalization, GPT-4 still outperforms humans, but the effect is lower and statistically non-significant (p = 0.31)."

1eggsyntax
Extremely of interest! Thanks very much for sharing, I hadn't seen it.

I really liked this essay a lot, especially how it shows the importance and necessity of creativity in such a rigorous field like mathematics for being able to explore and potentially end up wrong. This sprung up many thoughts:

  • Advice I had heard from Jacob Lagerros, "If you're not embarrassed by what you ship, you ship too late."

    • Contra the first technique the author mentions, having the audacity to put out a bad product seems good for feeling the fear but continuing to work anyways, continuing to explore what's unknown to you.
  • The etymology of the

... (read more)

Paul Graham


Link for the curious http://www.paulgraham.com/greatwork.html

I've heard this comment as, "if you're not paying for the product, you are the product."

I've read this book and tried to read it again as I thought I was missing something, but my impression of the book is that it's somewhat sloppy, a bit preachy of ZK being a cure-all, makes much more complicated a very simple system to the point of obfuscating the main point.

To my understanding, all the Zettlekasten is is having notes with:

1. individual names (if you look for one name, one note comes up), 
2. creating links between associated ideas (if you think, "wow, this reminds me of..." you may forget that connection later, so you link them), and&n... (read more)

I think I've done similar explorations as you've mentioned and have been curious to develop a framework of how to go about this more generally; specifically I get lost in your first example's preservation of structure (I think this is shallow vs. deep dives, could be wrong about the terminology.)

To my understanding, the overall objective is "to get a feel for what is out there." After running through it a couple times, I think I see a general pattern of...

  1. Getting in front of you "what is out there"
    - 100 companies who had 50% of the non-financial assets (NF
... (read more)
2johnswentworth
This sounds right.

Thanks for posting this! I'd also tag to the idea of the "Illusion of Transparency" that it may seem like common knowledge of how to be a part of online communities but has been fairly foreign to me. It's nice to get explicit steps/suggestions such as this.

On posting related ideas, I see this as incredibly helpful as has been noted by the "zettlekasten" method and being able to develop a highly connected network of knowledge. It's really cool to think of a network growing in the direction of ideas and people by this sort of act.