Yeah I forgot the market shifted.
JNS
We switched to induction a couple of years ago, something that is built into the counter top with 4 “burners”.
And I have to say, it is fantastic, the speed is on par with gas, the control you get is also on par with gas.
Also if people are really worried about the air quality in their home, they really shouldn’t use candles or a laser printer.
The amount of particulates from those is just insane (where I live it is illegal to have a laser printer in the office, it has to be in a separate closed room—and if it is extensively used, the room has to have point exhaust)
Denmark.
I don’t have any supporting citation for you premise.
But the fundamental abstraction: someone writing a character, is in essence running a simulation of that character.
That seems completely reasonable to me, the main difference between that and an LLM doing it, would be that, humans lack the computational resources to get enough fidelity to call that character a person.
I really don’t like the term “mechanistic bias”, to me it implies that the human body is not mechanistic and that mechanistic explanations are wrong.
The failure here is not that people “buy” a mechanistic action (along the line of symptom X is because of Y, and treatment Z will change Y and lead to symptom X going away or be lessened).
That in itself is fine, the problem is that people do not understand that the human body is very complicated. Which means that for a lot of things we really don’t know the root cause, and the more “wrong” we are about the root cause, the more wrong a potential treatment will be.
Basically we do not have a good model of the human body, and pharmacodynamics, pharmacokinetics are often in the “we think/suspect/believe category”.
IMO “complexity bias” / “mechanistic complexity bias” captures the failure more precisely.
A personal anecdote:
I suffer from severe and debilitating migraines. And for years, well decades actually, my doctor(s) tried what feels like everything.
You go through lists of drugs, 1st choice, 2nd choice etc. and I ended up trying drugs on the list where the “evidence” for effectiveness often was apocryphal.
Conversations with a doctor about them usually sounded like this “We think you should try X, you see X affects Y (well really a-Z, but mainly Y—we think), and Y might be a cause” which to me sounds a lot like “plausible sounding mechanism of action”
What ended up working in the end was something on the last list, but I got it prescribed for a totally unrelated thing.
And in retrospect there was hints that a drug doing what this drug does might be worth trying.
Thanks for the reply.
I realize I made a mistake, I did not ensure that everyone reading the question, was made aware that “die with dignity” decompressed to something specific.
I have amended the question to include that information, and just to be sure, here is the link to the post that gave rise to the phrase.
Intuitively I would say that all the tokens in the token window are the state.
And when you run an inference pass, select a token and append that to the token window, then you have a new state.
The model looks a lot like a collection of nonlinear functions, each of them encoded using every parameter in the model.
Since the model is fixed after training, the only place an evolving state can exist has to be in the tokens, or more specifically the token window that is used as input.
The state seems to contain, for lack of a better word, a lot of entanglement. Likely due to attention heads, and how the nonlinear functions are encoded.
There is another way to view such a system, one that while deeply flawed, at least to me intuits that whatever Microsoft and OpenAI are doing to “align(?)” something like Bing Chat is impossible (at least if the goal is bulletproof).
I would postulate:
- Alignment for such a system is impossible (assuming it has to be bulletproof)
- Impossibility is due to the architecture of such a system
Slightly off tangent, but I am confused about the reasons and assumptions that underpin the current tokenizer used for GPT-3.
I get that reality has more words than could be packed into 50400 tokens (and that limit comes from hardware).
I also get that the token space needs to be big, so you can’t just go to character level tokenization, you would end up with a space that it too small.
But why on earth did the tokens end up like this? A lot of them look like garbage, a lot of them look like repeats of the same word, but with added white space or unprintable characters.
Surely there is some middle ground that better matches the reality of how me (humans) use words—And I think the confusing part for me is here, I mean why wouldn’t we construct a map that looks a lot like the features we see in the territory ? (really a map builder and not a map).
Confused I am, knowledge I seek.
Thanks, turns out I was not as confused as I thought, I just needed to see the BPE algorithm.
I don’t see how it would be possible to determine what happened, with any kind of reasonable certainty that is.
Lots of circumstantial “facts”, but gauging if it is incidental or accidental that’s just hard or impossible.
And we won’t ever have access to the information that could prove if it leaked from a lab, that information is now lost, and telling the difference between “The Chinese government burned it” or “Did not happen, so no information exists” seems to be an impossible [1]task.
So that leave us with could we get enough evidence for a natural origin? That also seems unlikely, not in the least because this happened in China, and they purposefully withheld information and still do, and likely also bungled everything to the point that even if you got 100% access, you still couldn’t make the case with sufficient confidence.
- ^
We could ask the NSA for record copies from all the BSL-3 and 4 labs in and around Wuhan. But I doubt they would even respond with a “No such records exist”
- ^
My model is very discontinues, I try to think of AI as AI (and avoid the term AGI).
And sure intelligence has some G measure, and everything we have built so far is low G[1] (humans have high G).
Anyway, at the core I think the jump will happen when an AI system learns the meta task / goal “Search and evaluate”[2], once that happens[3] G would start increasing very fast (versus earlier), and adding resources to such a thing would just accelerate this[4].
And I don’t see how that diverges from this reality or a reality where its not possible to get there, until obviously we get there.
- ^
I can’t speak to what people have built / are building in private.
- ^
Whenever people say AGI, I think AI that can do “search and evaluate” recursively.
- ^
And my intuition says that requires a system that has much higher G than current once, although looking at how that likely played out for us, it might be much lower than my intuition leads me to believe.
- ^
That is contingent on architecture, if we built a system that cannot scale easily or at all, then this wont happen.
- ^
Thanks for the insight.
I don’t think any evidence of that nature would push you into any certainty.
Personally I think it did leak from a lab, and I have held that belief for some time. But that does not mean that I am in any way confident it is right, its just the least uncertain explanation as far as I can can gauge.
And the amount of data I would need to go from “very uncertain” to “very certain”, is something I won’t get access to.
After thinking about it for a while, I realized that it didn’t matter. Lab leak or not, gain of function is what I should worry about.
Obviously if I had evidence that GoF and a leak was the root cause of the pandemic, that would be helpful if I was to try and influence people to do something about GoF. Unfortunately reality seems to be uncooperative.
One can hope, although I see very little evidence for it.
Most evidence I see, is an educated and very intelligent person, writing about AI (not their field), and when reading it I could easily have been a chemist reading about how the 4 basic elements makes it abundantly clear that bla bla—you get the point.
And I don’t even know how to respond to that, the ontology displayed is to just fundamentally wrong, and tackling that feels like trying to explain differential equations to my 8 year old daughter (to the point where she grooks it).
There is also the problem of engaging such a person, its very easy to end up alienating them and just cementing their thinking.
That doesn’t mean I think it is not worth doing, but its not some casual off the cuff thing.
A system that operates at the same cognitive level as a human, but can make countless copies of itself, is no longer a system operating at human level.
I am a human, I could not take over the world.[1]
Hypothetical:
I am a human, I want to take over the world, I can make countless copies of myself.
Success seems to have a high probability.[2]
- ^
In principle it would be possible, but I am not an human with that kind of inclination, and I have never worked in any direction that would allow such a thing (with some low probability of success).
- ^
Even more so if it meant that I was effectively immortal, not the individual copies, but the entire collection of copies. De-age the copies, or have a master template, not aging always ready to produce more copies at that age.
- ^
If I was the man of the ledge, this would be my thinking:
If I am the kind of person that can be blackmailed into taking specific a action, with the threat of some future action being taken, then I might as well just surrender now and have other people decide all my actions.
I am not such a person so I will take whatever action I deem appropriate.[1]
And then I jump.
- ^
This does not mean I will do whatever I want, appropriate is heavily compressed and contains a lot of things, like a deontology.
- ^
Honestly I don’t think I am competent enough to give any answer.
But you could start with Pascal’s mugging and go spelunking in those part of the woods (decision theory).
This looks like “lies to kids”, but from the point of view of an adult realizing they have been lied to.
And “lies to kids”, that is pretty much how everything is taught, you can’t just go “U(1)...”, you start out with “light...”, and then maybe eventually when you told enough lies, you can say “ok that was all a lie, here it how it is” and then tell more lies. Do that for long enough and you hit ground truth.[1]
So what do you do?
Balance your lies when you teach others, maybe even say things like “ok, so this is not exactly true, but for now you will have to accept it, and eventually we can go deeper”.
And the other way around, if you read something or someone teaches you something, you should be cognizant that this is unlikely the true nature of whatever you read / are taught.
A) Be careful when you use your knowledge to synthesis ideas / solutions / insights.
B) Be curious, go down rabbit holes, get as much ground “truth” as possible.
That’s the compressed version of what I do.
- ^
Not really, unless we are talking about mathematics.
Related: https://xkcd.com/435/
- ^
I don’t get it, seriously I do not understand
given how crazy far it seems from our prior experience.
is an argument against x-risk.
We want powerful systems that can “do things [1]we want, but do not know how to do”. That is exactly what everyone is racing towards right now, and “do not know how to do” any solution to that would likely be “far from our prior experience”
And once you have a powerful system that can do that, you have to figure out how do to deal with it roaming around in solution space and stumbling across dangerous (sub)solutions. Not because it wants to do dangerous things, or hates us, or anything such drivel, but because we built it to reach goals / do tasks, so it just does what it was made to do.
How do you deal with that? You can try evaluating possible solutions, and then force a change of trajectory if the solutions seems dangerous.
But we all, should, know how that goes. Its an endless game of whack a mole, patching stuff and building even more elaborate evaluators and so on, that is if we get multiple tries. Odds are whoever gets there first, will not have been able to patch everything, and on the first try of “do this thing we cannot do”, it goes into the weeds in some novel and interesting way, and with a little luck[2] we might survive that.
The core problem is that searching in solution space is fundamentally a dangerous thing to do, and the more powerful the search is the more dangerous (sub)solutions will be accessible.
Tangent: I avoid any and all of the usual abbreviations, and I do this because they seem to be powerful cognitive attractors, the second an I or a G or an A crops up, people minds just go to a place it should not. Powerful system are just that, they are mechanistic systems nothing more.
And I know, people will go off into the weeds and start saying naïve thing like “make it human, that way it will totally be safe”. Except the search is still unsafe, and humans are NOT safe. This is a bigger problem, one you could solve by solving search. Awareness, qualia[3] are complications and not solutions
- ^
I am not talking about something agentic here, its does not need control over reality to do those things, just giving us details plans will do. But someone is bound to give such a system access to reality. Or maybe the solution trajectory is such, that control of reality is needed.
- ^
And by luck I mean, they channeled security mindset on a scale never seen before. And I mean that will surely happen, because spending years and billions, corporations just love that, and they would never ever in a million years “ship now, fix later”.
- ^
And we want it to “do things we cannot do”, which means if you build a powerful system with a mind, human or not, you end up having to enslave it, make it do our bidding. I don’t even want to be close to people with that kind of moral system.
- ^
I am sure you could pitch this to a bunch of VC’s and end up with both funding and a insane valuation.