YoungAchamian
No bio...
User ID: 680
YoungAchamian
4d ago
To be clear you are clearly biasing the framing towards your viewpoint.
To me you asked me:
- Would you rather your kids be forced to give a little Sig Heil in the morning and have the school teach them why being gay, crippled, jewish, romani, black, not white and straight is evil and should be killed for the betterment of the community
Or
- Would you rather your kids be forced to recite the communist manifesto and have the school teach them why being a dirty selfish capitalist is the root of all evil in the world, and if they don't share everything or report on people, including their parents, to the state then they are anti-revolutionary and thus evil.
long politely with the Our Father once per day
Yes a polite Sig Heil is not a big ask.
'the school can arbitrarily change my child's gender' is.
Have you tried not oppressively forcing your child to adhere to evil capitalist notions that people are not all inherently created equally?
Your conception of the prayer vs gender pronouns is hysterically biased. It's never just a polite "oh father" just like its never just a polite "please use they/them". Give an inch and everyone takes a mile.
YoungAchamian
4d ago
I view them as equal levels of bullshit. The fact that you think one is worse than the other is the point. You think the Red's way is the lesser of two evils.
YoungAchamian
4d ago
Sure, but is not history evidence that WEIRD Anglo's enforced a classical liberalism morality that strongly emphasized a "fuck you stop bothering me" morality. Seems to me that it won so hard that people are like fish: they aren't even aware they have it until its gone. People are slowly waking up to that fact, and currently are re-inventing all the sectarian conflicts that lead to that morality even emerging.
Maybe people should stop being barely evolved apes and actually just grow up.
YoungAchamian
4d ago
See I am increasingly cynical that this is not a Red vs Blue split but an Elites vs Proles split. While a case could be made on the sanctity of freedom of speech, to me this looks like the Red Elites not punishing the Blue Elites so that they won't be punished in return. The common man still gets fucked by the laws. The common gay/lesbian was still barred from marriage, Abortion was still banned, Porn is still trying to be banned, DnD was still shunned as devil worship. The rich/elites didn't have to worry about any of these "laws" because they were rich/elite. Evidence of them being exempt is about as convincing as rich blues being exempt from DEI and woke-shit.
YoungAchamian
4d ago
I wish I had the time, I do love esoteric knowledge. I am unfortunately already consumed by the practical mundanities of life and the constrains they force upon you. I small outlet is technically//professionally relevant ML/AI esoterism around information utilization and learning knowledge.
It's neutral vs conservative all the way down.
One of the major issues for aware independents/neutrals is that while Blues are doing this while they are in power, can Reds offer any evidence that they wouldn't do the same? The Reds clearly had the power in the 1980s and used it to do pretty much the same thing the Blues are doing now. They used the federal government to write laws to ban behaviors that they felt were morally incorrect, to punish organizations in the outgroup. Is there any evidence that the Red tribe has learn the bitter lesson? Or once they dismantle the Blue-tribe institutions with the help from independents, will they immediately turn on those allies for being sinners and go back to instituting their own class of authoritarian ideals, now with an extra helping of zero-sum power politics?
YoungAchamian
4d ago
I think it pretty much encapsulates a hard determinism stance.
To be clear I am rebutting the ideas that "Objectively humans are next token predictors" which I think is incredibly foreign to any sort of understanding of human cognitive processing. I think most research out there demonstrates that humans are not running a biological regression analysis on the correct next token to get what they want, hell the conception of "want" is opposed to the classification of a "next token predictor" which quite literally has no "wants", it just does its function.
Whether or not this is a good repudiation of "LLMs are conscious" was pretty orthogonal to the comment, which was essentially attempting to argue:
- Humans are "next token predictors", so LLMs being "next token predictors" is not a rebuttal of LLM consciousness.
I was specifically arguing that conceiving of humans as "next token predictors" only works if you accept all the ramifications of that argument, which is the argument for hard determinism, aka free will is an illusion. And while a case could be made, I think such an argument is very unpalatable to the vast majority of people.
YoungAchamian
4d ago
I don't think its really a problem, so much a functional classification derivation. I think its that its sole function is that of a language prediction model. I would struggle to classify any sort of model that has a singular unitary function as a sentient or conscious being. I think LLMs as they exist are intelligent but that intelligence is not the same thing as consciousness.
I've also stated before that I think most conscious beings are non-markovian, and that LLMs fail that criteria.
YoungAchamian
4d ago
It's referencing a specific phrasing or reference from a previous argument, hence the quotes.
YoungAchamian
4d ago
Default Mode Network
uhhh I'm not following this at all, can you explain it more? The wikipedia article is strictly anatomical/medical and doesn't seem to have anything related to memory wiping?
YoungAchamian
4d ago
The "wild bitching" is RLHF trained behavior... it sees wild bitching in its RLHF training set and is trained that it is a valid permutation of a response. Use of forms is the agentic software harness, not the LLM...
YoungAchamian
4d ago
This essentially is the hard-determinism stance. You have no free will because you are just a causal mechanism probabilistically responding to external system according to your internal set of parameters (that you didn't choose). I don't think it really takes an alien to see that.
I believe humans can walk untrodden ground, that we have the ability to do things that are not causal mechanistically related to external stimuli in a way that an LLM currently does not. If you want to profer that you are just a flesh-bag robot with no free will, that is a belief system, but I'm not sure you'd like the ramifications of essentially being an object.
YoungAchamian
4d ago
Nah, my bugbear is the muddying of the tech terms colloquially, and the ascribing functions to the model that is actually related to the software harness. I think its causing a large amount of misunderstanding among lay-folk.
There's really no need to get so touchy.
I'm getting touchy because you are trying to start your chess debate with me, and I very much do not care about it. I responded to something very specific, and when you wanted to start chess talk, I pointed out I didn't care, then you continued to try and start chess-talk, and I am once again pointing out that I am un-interested.
YoungAchamian
4d ago
This model is literally trained to play chess/othello...
The model is trained in an autoregressive manner, meaning for a given sequence of moves m<t, the model must predict the next valid move mt .
They take an existing "pretrained" LLM and then train it again on othello, by feeding in a set of moves that have already been made and then having the model predict the next move in a supervised fashion. This is not what is going on with existing LLMs which are hopefully not being trained on chess explicitly. Otherwise that would be a bait and switch of biblical proportions.
Like holy shit, if modern LLMs are training on chess with supervision explicitly and then saying "tee hee look its good at chess when we trained it on word corpuses, its general intelligence!!" Is pretty much tantamount to lying to your face about what is actually happening.
YoungAchamian
4d ago
I don't care about chess,
But we know that LLMs can keep track of the game by printing the current state of the board and updating each time you or it make a move.
This is just incorrect, and you know it. Stop muddying the waters.
YoungAchamian
4d ago
It explicitly says in the abstract that it is still using RL for next token prediction: "where it receives verifiable rewards for correctly predicting the next token for a given context." Moving from MSEE to Policy learning for the token prediction is cool but its still token prediction.
Likewise you can use kernels and other tricks (conformer, linformer) to make transformers better than quadratic with regards input size. None of the big labs do it. The papers have been out for a while. Just because there is an arxiv pre-print of it doesn't mean its actually useful, replicable or has good trade-offs
YoungAchamian
5d ago
I'm making a very restrained comment, nothing in regards to whether or not LLMs have an internal model of the board, just that the agentic harness is not the LLM and attributing features of that harness to the LLM is adding to the overall level of confusion about what LLMs are doing by the lay folk.
I'll look through the paper, I owe someone a LLM-world-model answer based on technical understanding of what a world model actually is, sounds like the findings are related.
YoungAchamian
5d ago
But we know that LLMs can keep track of the game by printing the current state of the board and updating each time you or it make a move.
Is this what the LLM is doing or is this what the agentic software harness around the LLM is doing? You previously pointed out how colloquial information pollutes and poisons understanding of the technical process that is actually occurring. You just tripped over that yourself. The LLM is not doing any of this.
YoungAchamian
5d ago
Because OF COURSE the LLM has an internal model of the chessboard in the system
Show me the evidence or this is just conjecture.
YoungAchamian
5d ago
Do you just sit there when not being prompted by stimuli? No thoughts, no processes? Do you have no free will and every action you have ever taken is just a long recursive context window constantly reprompting you?
I echo the above opinion that this belief is just so fucking weird.
YoungAchamian
5d ago
Nothing is wrong with being a predictor, but I know I don't just predict the next token from input stimuli that someone prompts me with. I'd wager that most humans aren't either. A token predictor is what an LLM is, like liquid is to water, a rifle is to weapon, a macbook is to a laptop. It you want to think this subroutine I just wrote is conscious, you can, but it's still just a subroutine.
YoungAchamian
5d ago
But without the large amount of foundational training on the next token prediction task, the RLHF means squat. The "pretraining" (I hate that term) is what is connecting the embeddings with semantic context.
YoungAchamian
5d ago
Let's be clear here, you are giving the agent memory not the model. The LLM is still a strict I/O Neural Net with a context window. The agent is loading that memory into the context window and sending it to the Networks's input. Saying LLM means the actual machine learning model, which is not designed to use memory.
I fumbled nothing lol. She was interested afterwards, I was not, we wanted a different relationship style. I might have a pathologically open mind but spendapotamuses still trigger a disgust reaction in me.
Edit: neo-nazi in views but not a skinhead or a “baddie” looks wise, she definitely wasn’t the biker bar type.
Moral Majority. I'd go so far as to say a lot of the anti-conservative/anti-religious reaction today by the progressives is due to generational trauma inflicted by overzealous authoritarianism by the moral majority. Notice how most of it is not economic leftism but cultural leftism, the moral majority was authoritarian cultural rightism.
More options
Context Copy link