I think it’s an intentional pun, like, “whether forecasters” are people who predict whether something will happen or not.
npostavs
Maybe I should have asked: In what sense are machines “fully doing” first-order logic? I think I understand the part where first logic formulas are recursively enumerable, in theory, but isn’t that intractable to the point of being useless and irrelevant in practice?
Unlike first-order logic, second-order logic is not recursively enumerable—less computationally tractable, more fluid, more human. It operates in a space that, for now, remains beyond the reach of machines still bound to the strict determinism of their logic gates.
In what sense is second-order logic “beyond the reach of machines”? Is it non-deterministic? Or what are you trying to say here? (Maybe some examples would help)
What about tuning the fiddle strings down 1 tone?
You say this:
If you’re thinking, “Wait no, I’m pretty sure my group is fundamentally about X, which is fundamentally good,” then you’re probably still in Red or Blue.
But you also say this:
First, the Grey tribe is about something, [...] things that people already think are good in themselves.
Doesn’t the first statement completely undermine the second one?
I guess you meant jukebox, not jutebox. Unless there is some kind of record-playing box made of jute fiber that I haven’t heard of...
but I recently tried again to see if it could learn at runtime not to lose in the same way multiple times. It couldn’t. I was able to play the same strategy over and over again in the same chat history and win every time.
I wonder if having the losses in the chat history would instead be training/reinforcing it to lose every time.
Yes, my understanding is that the system prompt isn’t really priviledged in any way by the LLM itself, just in the scaffolding around it.
But regardless, this sounds to me less like maintaining or forming a sense of purpose, and more like retrieving information from the context window.
That is, if the LLM has previously seen (through system prompt or first instruction or whatever) “your purpose is to assist the user”, and later sees “what is your purpose?” an answer saying “my purpose is to assist the user” doesn’t seem like evidence of purposefulness. Same if you run the exercise with “flurbles are purple”, and later “what color are flurbles?” with the answer “purple”.
#2: Purposefulness. The Big 3 LLMs typically maintain or can at least form a sense of purpose or intention throughout a conversation with you, such as to assist you.
Isn’t this just because the system prompt is always saying something along the lines of “your purpose is to assist the user”?
by saying their name aloud: [...] …but it’s a lot more difficult to use active recall to remember people’s names.
I’m confused, isn’t saying their name in a sentence an example of active recall?
Finding two bugs in a large codebase doesn’t seem especially suspicious to me.
I don’t think I understand, what is the strawman?
I think the AI gave the expected answer here, that is, it agreed with and expanded on the opinions given in the prompt. I wouldn’t say it’s great or dumb, it’s just something to be aware of when reading AI output.
It looks like you are measuring smartness by how much it agrees with your opinions? I guess you will find that Claude is not only smarter than LessWrong, but it’s also smarter any human alive (except yourself) by this measure.
Entries 1a and 1b are obviously not not relevant to the OP, which is mainly about the sense in 3b (maybe a little bit the 3a sense too, since it is “merged with or coloured by sense 3b”).
Entry 3b looks (to me) sufficiently broad and vague that it doesn’t really rule anything out. Do you think it contradicts anything that’s in the OP?
The OED defines ‘gender’, excluding obsolete meanings, as follows:
Okay? Why are you telling us this?
Maybe if you solve for equilibrium you get that after releasing the tool, the tool is defeated reasonably quickly?
I believe it’s already known that running the text through another (possibly smaller and cheaper) LLM to reword it can remove the watermarking. So for catching cheaters it’s only a tiny bit stronger than searching for “as a large language model” in the text.
Why release a phone with 5 new features when you can just R&D one and put it in a new case?
In the ideal case of a competitive market, you don’t release just one new feature, because any of your competitors could release a phone with two new features and eat your lunch. But the real-world smartphone market is surely much closer to oligopoly than perfect competition.
The costs of the competition of the market are almost invisible, but we have been seeing them over decades get more and more obvious.
How sure are you that this isn’t rather the costs of lack of competition?
Maybe, although what is “sufficient” depends a lot on the rate of catching the evaders. I don’t have a good guess as to what that rate is.
I don’t think this interpretation can hold up: the body of titotal’s post doesn’t deal with the good vs bad timeline. It’s just about the uncertainty of modelling AI progress which applies for both the good and bad timelines.