LessWrong 2.0 Reader

If you do end up installing the Fatebook browser extension and making some predictions, let me know!

I did.

Yeah, if that was the only consideration I think I would have created the market myself.

Some reporting on this: https://www.vox.com/future-perfect/374275/openai-just-sold-you-out

When I write a post and select text, a menu appears where I can select  text appearance properties etc. However, in my latest post, this menu does not appear when I edit the post and select text. Any idea why that could be the case?

Okay, sounds like I have misunderstood you.

Sure, I can retry.

My next attempt to pass your ITT is thus:

Broadly when gaining new skills we go from doing what feels natural, to doing things differently within rigid structures, to getting good at them, to releasing the structures and then just doing what comes naturally. And often afterwards it is both more effective and also comes more naturally than it did before.

Some people seem trapped in the middle step on certain things. They always practice music with a metronome ticking in order to keep the beat, they never trust themselves to just feel it. They always leave the party without drinking, never trusting themselves to behave well and have fun with it. They always need an explicit theory guiding their overall trajectory in life (e.g. career decisions involving spreadsheets), they can never make a major life decision because it feels good in their gut. They always have to discuss purchases over $1,000 with their spouse and sleep on it, they never feel comfortable just going with something that feels right in the moment.

Such people have successfully found useful structures, but are also trapped in them, never venturing forward into the world themselves, always bound by the formalities. This limits their personhood and humanity from coming through, it bounds them to only be as good as the structures they've adopted.

Insofar as you name a structure or set of rules for living life, you are always bound by them and will never let your humanity outshine them.

How close is this to what you're saying, from 1 to 10?

Ah right, the decades part--I had written about the 1930 revolution, commune, and bourbon destitution, then checked the dates online and stupidly thought "ah, it must be just 1815 then" and only talked about that. Thanks

That would be great, but maybe it is covered much more in your bubble than in large newspapers etc? Moreover, if this is covered like the OpenAI-internal fight last year, the typical news outlet comment will be: "crazy sci-fi cult paranoid people are making noise about this totally sensible change in the institutional structure of this very productive firm!"

As a tangent to my question, I wonder how many AI companies are already using RLAIF and not even aware of it. From a recent WSJ story [LW(p) · GW(p)]:

Early last year, Meta Platforms asked the startup to create 27,000 question-and-answer pairs to help train its AI chatbots on Instagram and Facebook.

When Meta researchers received the data, they spotted something odd. Many answers sounded the same, or began with the phrase “as an AI language model…” It turns out the contractors had used ChatGPT to write-up their responses—a complete violation of Scale’s raison d’être.

So they detected the cheating that time, but in RLHF how would they know if contractors used AI to select which of two AI responses is more preferred?

BTW here's a poem(?) I wrote for Twitter, actually before coming across the above story:

The people try to align the board. The board tries to align the CEO. The CEO tries to align the managers. The managers try to align the employees. The employees try to align the contractors. The contractors sneak the work off to the AI. The AI tries to align the AI.

I surveyed some participants about their preferences. I believe this is nine generals plus a petrov (second from bottom).

9 people prefer peace; 1 deranged person says "peace = mutual destruction" and "I was kinda hoping the other side would launch the nukes"; fortunately we survived them.

If one side had admitted to launching nukes, it looks like at least one person on the other side would have favored retaliating; unclear whether they'd get a majority (and unclear whether they'd launch without a majority).

I agree with the long note. I think anonymity is ~necessary to get decent P(defection) from a small group of high-karma users. But there are issues with that and the ritual is fine with low P(destruction).

That $769 number might be more relevant than you expect for college undergrads participating in weird psychology research studies for $10 or $25 depending on the study.