> but is slower than the baseline. Validator says pass, result is useless You should read this blog, they cover this exact scenario - https://www.weco.ai/blog/first-evidence-of-recursive-self-im... > One domain that…
And today we call it transmutation - https://en.wikipedia.org/wiki/Nuclear_transmutation
Fascinating. I think it's the first time I've heard it put that way. For me it's more intuitive the other way around, as the "outer" loops increase in complexity (and can have additional separate loops running inside…
There's two variants of this (or, as the joke goes, for very big values of bit): Ternary Bonsai 27B uses ternary {−1, 0, +1} weights with FP16 group-wise scaling, giving a true 1.71 effective bits per weight. 1-bit…
> isn't it an obvious fact Just below your question is a very confidently incorrect take about "parroting"... So, not obvious at all, at least for some people :)
Not that one. This one is a start-up founded by ex-oAI CTO Mira Murati. Last I heard they were mainly doing hosted finetunes with a few clicks on popular open models.
> Using ciphertext for inference would mean it's not a very secure ciphertext. Inference is done in plain text. It's just that some parts of the response can be encrypted. While I haven't looked into this specific…
> morality matching or exceeding Western Not sure if troll or misguided, but you know there are people still alive today that can talk about that moral superiority from experience, right? Just ask them, read their…
> anyone ever needs a pot to do is hold fluids Yes, but this might also be a counter-point to your position. In a world of rising baselines, having a "pot but without all the bells and whistles" might be a thing people…
> Wishful thinking by the managerial class. Heh. I wouldn't be so fast in calling this one, yet. It might go either way, or a combination of the two, who knows... Things are moving and progressing fast enough to at…
I try to get past the claudisms if the subject is interesting, in the hopes of learning something new, and this article had many such claudisms throughout, but it was an interesting read nonetheless. It could have been…
No, training data has nothing to do with this. My point was that "you can't take over the world with tokens" is not a valid counterpoint in the doomer's position. Because their view is that singularity can happen. And…
Haha. The time travel is part of the thought experiment, just a means to explore an idea. We don't have to actually do it. Let me put it another way. Would you be able to "control the world with tokens" today if you had…
> You cannot take over the world with tokens. I have a nice thought experiment I like to do with people when confronted with "AI can't do x". Let's go back in time. How much do we need to go for this to become true? So…
There's 0 chance of that. You can maybe say that for things like material science, nuclear stuff, weird physics, and so on (basically anything relating to making big booms, delivering big booms, or ensuring others can't…
> Not proof, but a high bar for evidence that we may be in that explosion now. I agree, that's why I said "we're seeing things we have working towards this". I think that the "jagged intelligence" that is often used to…
> Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. That's a pretty early definition of what we now call ASI (Artificial Super…
> haggling culture. There's haggling where you actually enjoy the process (i.e. Turkey, visiting the bazaars, you get to haggle, then you get invited for some cay w/ the vendor, talk a bit, that's really enjoyable) and…
> in chess engines that literally have endgame databases for example You / Carlsen / anyone will not beat a top chess engine even without the endgame databases. In the vast majority of cases you / anyone won't even…
Thanks. I was asking because I couldn't find even their previous 1.0 model there.
How are people trying this? I don't see it on openrouter. Any ways of testing this without subscribing to meta stuff?
> I've read multiple times that this approach is harmful in training. There's a lot of nuance here. Note that I said "prepare" datasets and not just "generate" datasets. First, the "model collapse" paper(s) were highly…
(from Cursor's blog) > Training included trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools. This dataset lets the model learn both from existing…
> so shouldn't it also be open weight? Should as in "would it be nice?" - yeah. Should as in they have to? No. > Permission is hereby granted, free of charge, to any person obtaining a copy of this software and…
It is already here. Not humanoid (yet, but it's in the works) but tracked robots with bolted on machine guns have both held and captured positions in UA.
> but is slower than the baseline. Validator says pass, result is useless You should read this blog, they cover this exact scenario - https://www.weco.ai/blog/first-evidence-of-recursive-self-im... > One domain that…
And today we call it transmutation - https://en.wikipedia.org/wiki/Nuclear_transmutation
Fascinating. I think it's the first time I've heard it put that way. For me it's more intuitive the other way around, as the "outer" loops increase in complexity (and can have additional separate loops running inside…
There's two variants of this (or, as the joke goes, for very big values of bit): Ternary Bonsai 27B uses ternary {−1, 0, +1} weights with FP16 group-wise scaling, giving a true 1.71 effective bits per weight. 1-bit…
> isn't it an obvious fact Just below your question is a very confidently incorrect take about "parroting"... So, not obvious at all, at least for some people :)
Not that one. This one is a start-up founded by ex-oAI CTO Mira Murati. Last I heard they were mainly doing hosted finetunes with a few clicks on popular open models.
> Using ciphertext for inference would mean it's not a very secure ciphertext. Inference is done in plain text. It's just that some parts of the response can be encrypted. While I haven't looked into this specific…
> morality matching or exceeding Western Not sure if troll or misguided, but you know there are people still alive today that can talk about that moral superiority from experience, right? Just ask them, read their…
> anyone ever needs a pot to do is hold fluids Yes, but this might also be a counter-point to your position. In a world of rising baselines, having a "pot but without all the bells and whistles" might be a thing people…
> Wishful thinking by the managerial class. Heh. I wouldn't be so fast in calling this one, yet. It might go either way, or a combination of the two, who knows... Things are moving and progressing fast enough to at…
I try to get past the claudisms if the subject is interesting, in the hopes of learning something new, and this article had many such claudisms throughout, but it was an interesting read nonetheless. It could have been…
No, training data has nothing to do with this. My point was that "you can't take over the world with tokens" is not a valid counterpoint in the doomer's position. Because their view is that singularity can happen. And…
Haha. The time travel is part of the thought experiment, just a means to explore an idea. We don't have to actually do it. Let me put it another way. Would you be able to "control the world with tokens" today if you had…
> You cannot take over the world with tokens. I have a nice thought experiment I like to do with people when confronted with "AI can't do x". Let's go back in time. How much do we need to go for this to become true? So…
There's 0 chance of that. You can maybe say that for things like material science, nuclear stuff, weird physics, and so on (basically anything relating to making big booms, delivering big booms, or ensuring others can't…
> Not proof, but a high bar for evidence that we may be in that explosion now. I agree, that's why I said "we're seeing things we have working towards this". I think that the "jagged intelligence" that is often used to…
> Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. That's a pretty early definition of what we now call ASI (Artificial Super…
> haggling culture. There's haggling where you actually enjoy the process (i.e. Turkey, visiting the bazaars, you get to haggle, then you get invited for some cay w/ the vendor, talk a bit, that's really enjoyable) and…
> in chess engines that literally have endgame databases for example You / Carlsen / anyone will not beat a top chess engine even without the endgame databases. In the vast majority of cases you / anyone won't even…
Thanks. I was asking because I couldn't find even their previous 1.0 model there.
How are people trying this? I don't see it on openrouter. Any ways of testing this without subscribing to meta stuff?
> I've read multiple times that this approach is harmful in training. There's a lot of nuance here. Note that I said "prepare" datasets and not just "generate" datasets. First, the "model collapse" paper(s) were highly…
(from Cursor's blog) > Training included trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools. This dataset lets the model learn both from existing…
> so shouldn't it also be open weight? Should as in "would it be nice?" - yeah. Should as in they have to? No. > Permission is hereby granted, free of charge, to any person obtaining a copy of this software and…
It is already here. Not humanoid (yet, but it's in the works) but tracked robots with bolted on machine guns have both held and captured positions in UA.