I am sympathetic to the revolutionary vanguard approach. But MAPLE seems to have even less success in the approach than the SRs of the russian revolution. At least the SRs tried going to the people first—telling them “peasants, you are being exploited and you could change your condition!”.
From the outside, I don’t see any saintly accomplishments that look like “trying to apprehend alignment” at a robust mechanical level.
I think read the docs is a fine advice. But docs rot, and most docs are bad. If possible, read the code?
Reading the code doesn’t mean grokking the entire cpp code base while reading by candlelight in 1000-day monk mode. Reading the code means understanding the code path—descending from your first contact point down to the stateless functions that does the heavy lifting. This can be done with an actual debugger and stepping through each stage and see how the sausage is made. Reading the code is slow, sometimes painful. The alternatives are worse.
A bad advice is to use a “better” language. I guess that Raemon uses weakly typed languages beyond the base rate of js/python users. Many of the complaints here don’t really make sense if you write in apl, haskell or even rust. When you create the right abstraction, or best, have no abstraction, there is no meta state of the program to keep in your mind.
You probably don’t like the term LLM because it doesn’t describe capability. And most model are multimodal these days, so it is not just natural language.
You also wouldn’t like the term Autoregressive/Next-token predictor. Still because it says what it does, not what it is capable of.
AI is a pretty good term. As overloaded as it is.
go-away is my personal choice.
Doesn’t require weird js and text mode browsing like Anubis. Widely(ish) used. Not nuclear like anubis.
Not a downvoter, but I am put off by things like:
| Runs 93x faster than Zephyr 7B
On a…. What? A potato? A consumer gpu that doesn’t fit all of the 7B model so it is mem-moribund? Things with “patent pending” (nothing wrong with patents!) and permitting grad students to use it “for their degrees”. Just enough little vibe nudges that I feel confused and unmotivated to actually read the code/paper.
FYI there is branching on Claude desktop.
Fair. In Sarah Constantin’s terminology, it seems you aspire to “potentially take a stand on the controversy, but only when a conclusion emerges from an impartial process that a priori could have come out either way”. I… really don’t know if I’d call that neutrality in the sense of the normal daily usage of neutrality. But I think it is a worthy and good goal.
I don’t think Cole is wrong.
Lesswrong is not neutral because it is built on the principle of where a walled garden ought to be defended from pests and uncharitable principles. Where politics can kill minds. Out of all possible distribution of human interactions we could have on the internet, we pick this narrow band because that’s what makes high quality interaction. It makes us well calibrated (relative to baseline). It makes us more willing to ignore status plays and disagree with our idols.
All these things I love are not neutrality. They are deliberate policies for a less wrong discourse. Lesswrong is all the better because it is not neutral. And just because neutrality is a high-status word where a impartial judge may seem to be—doesn’t mean we should lay claim to it.
The claim about “no systematic attempt at making a good [prompt]” is just not true?
See:
Is this earth shattering as an observation? No.
But it is a good retelling of a phenomenon that most of us experience personally (density of time lived corresponding to novelty) tied in with interesting personal backstory. This is a good post that I enjoyed reading.
Good suggestion. Thou it was mostly as a tribute to a programming blog that I really appreciate. I expected no one else to read this but me.
TLDR: random is roughly 477 elo. (See details and caveats below)
See link, where someone matches weird chess algos against various dilutions of stockfish (stockfish at lvl n with x% of random move mixed in). So this elo is of course, relative to various stockfish dilutions at the time of recording.
https://www.youtube.com/watch?v=DpXy041BIlA
Revolutions
I like how it is a focused podcast on just the most important 10 revolutions of human history. Told humorously and researched thoroughly, it aims to teach the general truth through indepth case studies.
It teaches history very well. By using the revolution as the narrative skeleton, you are kept on track by the chronological evolution of the event. While learning the people and systems through context and fractal expansion on the details of the said participants.
Yes, mostly.
People with a vague interest in history and likes stories told well.
“Winning” is underpractied certainly, but undervalued!? I find that a bit hard to believe. (Currently reading the transcript, will change opinion if winning means differently than what Eliezer oft mentions)
> Rationality is Systematized Winning by Eliezer Yudkowsky
Sure—i am currently on my phone but I can paint a quick picture.
Local Memory—I keep my own internal predictions on fatebook and have it synced locally to my obsidian (a local markdown file manager). Then, I use Claude’s obsidian MCP to help me write down my daily notes from work and a jumbled context of my messages with coworkers, random web comments and other messaging services so it can help me to keep my profiles on my friends and projects up to date. (It is again, glued together with more MCPs that have limited access to my chatlogs with my friends). Ofc, with human in the loop.
Delphi—I wrote a simple MCP that basically just does the Delphi method with LLMs. Usually facilitated by Claude, it calls a panel of experts. These experts are the topK ranked models on LLM arena. And it does the questionaire generation based on my question, hand them out, aggregate the consensus, and decide if one is reached! Again, it has the context needed from me through my Obsidian. I use this for questions that are more personal or that there are not good liquidity for on prediction markets.
I will bite.
First of all, I appreciate the effort of trying to communicate better and hammering down the neat borders of word and how they used across domains—especially for words that are often used interchangeably and carelessly.
TLDR: Sometimes posts just get unlucky! And your style is on the verbose side and I am still somewhat confused about your value props.
It seems like your frustration is from a lack of responses—often, a lack of response might just be luck based and how the LW algorithm works (exponential time decay). Maybe you posted at a time that most forum users are asleep, and/or you are running against the headwind of literally one of the most popular article of all time. Sometimes, you just get unlucky! Twice even!
However, I can also say that the content is not written very legibly. It took me a long time to get the actual punch line and understand what you really want—“Can someone technical say something about how you want these terms to be used?”.
In addition, the post is written very verbosely. It takes a long time to get to the point, it is not clear about what it wants until the very end. It doesn’t say how you are going to this, or that if it is even worth engaging with you because it is not clear how you’d help to delineate the word boundary from your position. I am still unsure about the exact value proposition on how better delineating these words would lead to reduced p(doom).
Keep trying!
I mostly use Claude desktop client with MCPs (like additional plugins and tooling for Claude to use) for:
2-iter Delphi method involving calling Gemini2.5pro+whatever is top at the llm arena of the day through open router.
Metaculus, Kalshi and Manifold search for quick intuition on subjects
Smart fetch (for ocr’ing pdf, images, etc)
Local memory
I like where this wants to go, but I don’t want to get there with bad arguments.
To me, ank, this proposal is neither complete, nor comprehensive, and also not 100% safe.
It is not complete because “Ideally those non-agentic GPUs are like sandboxes that run an LLM internally and can only spit out the text or image as safe output” is just a rephrasing of the oracle[1] problem. Rephrasing does not a solution make.
It is not comprehensive because it rely on GPU manufacturers to do the right thing. If Nvidia realizes it can just bypass the protocol and takeover the world itself, the pressure for responsible actions would just suddenly evaporate.
It is not 100% safe because it is not complete nor comprehensive.
In addition, “But lets not make the perfect be the enemy of good” from the your comment below seems like a subtle bait and switch. In the original post, 100% safety is waxed poetic about. And yet in your response to buck, that goal is truncated to softer (and IMO more reasonable) stance that we’d want a “50% sure we’ll have 99% security in most cases”. The charitable reading is that your proposal is the only way that get 100% comprehensive safety, but it doesn’t mean we can get there right away. However, the ending of your comment—“Let’s not give up because we are not 100% sure we’ll have 100% security” feels too motte-and-bailey argumenty for me; you are the person who suggested that this is the “only” way towards “100% safe[ty]”, not us.
My intuition was that it was a semi-public meeting. My memory is hazy but there was a lot of people, and they were on twitter spaces (a mostly “public forum” style audio chatroom). So I don’t think it is a secret per say.
As someone who has both had a professional career in and out of China, I am utterly confused by the statement that the China today thinks ”rest of the world [is] low in quality”.
Care to share why you think that is the case?