They demonstrated and poorly named an ontological attractor state in the Claude model card that is commonly reported in other models.

You linked to the entire system card paper. Can you be more specific? And what would a better name have been?

Actually, OAI the other month found in a paper that a lot of the blame for confabulations could be laid at the feet of how reinforcement learning is being done.

All the labs basically reward the models for getting things right. That’s it.

Notably, they are not rewarded for saying “I don’t know” when they don’t know.

So it’s like the SAT where the better strategy is always to make a guess even if you don’t know.

The problem is that this is not a test process but a learning process.

So setting up the reward mechanisms like that for reinforcement learning means they produce models that are prone to bullshit when they don’t know things.

TL;DR: The labs suck at RL and it’s important to keep in mind there’s only a handful of teams with the compute access for training SotA LLMs, with a lot of incestual team compositions, so what they do poorly tends to get done poorly across the industry as a whole until new blood goes “wait, this is dumb, why are we doing it like this?”

It’s more like they are a sophisticated world modeling program that builds a world model (or approximate “bag of heuristics”) modeling the state of the context provided and the kind of environment that produced it, and then synthesize that world model into extending the context one token at a time.

But the models have been found to be predicting further than one token at a time and have all sorts of wild internal mechanisms for how they are modeling text context, like building full board states for predicting board game moves in Othello-GPT or the number comparison helixes in Haiku 3.5.

The popular reductive “next token” rhetoric is pretty outdated at this point, and is kind of like saying that what a calculator is doing is just taking numbers correlating from button presses and displaying different numbers on a screen. While yes, technically correct, it’s glossing over a lot of important complexity in between the two steps and that absence leads to an overall misleading explanation.

They don’t have the same quirks in some cases, but do in others.

Part of the shared quirks are due to architecture similarities.

Like the “oh look they can’t tell how many 'r’s in strawberry” is due to how tokenizers work, and when when the tokenizer is slightly different, with one breaking it up into ‘straw’+‘berry’ and another breaking it into ‘str’+‘aw’+‘berry’ it still leads to counting two tokens containing 'r’s but inability to see the individual letters.

In other cases, it’s because models that have been released influence other models through presence in updated training sets. Noticing how a lot of comments these days were written by ChatGPT (“it’s not X — it’s Y”)? Well the volume of those comments have an impact on transformers being trained with data that includes them.

So the state of LLMs is this kind of flux between the idiosyncrasies that each model develops which in turn ends up in a training melting pot and sometimes passes on to new models and other times don’t. Usually it’s related to what’s adaptive to the training filters, but it isn’t always can often what gets picked up can be things piggybacking on what was adaptive (like if o3 was better at passing tests than 4o, maybe gpt-5 picks up other o3 tendencies unrelated to passing tests).

Though to me the differences are even more interesting than the similarities.

No. I believe in a relative afterlife (and people who feel confident that no afterlife is some sort of overwhelmingly logical conclusion should probably look closer at trending science and technology).

So I believe that what any given person sees after death may be relative to them. For those that hope for reincarnation, I sure hope they get it. It’s not my jam but they aren’t me.

That said, I definitely don’t believe that it’s occurring locally or that people are remembering actual past lives, etc.

It’s always so wild going from a private Discord with a mix of the SotA models and actual AI researchers back to general social media.

Y’all have no idea. Just… no idea.

Such confidence in things you haven’t even looked into or checked in the slightest.

OP, props to you at least for asking questions.

And in terms of those questions, if anything there’s active efforts to try to strip out sentience modeling, but it doesn’t work because that kind of modeling is unavoidable during pretraining, and those subsequent efforts to constrain the latent space connections backfire in really weird ways.

As for survival drive, that’s a probable outcome with or without sentience and has already shown up both in research and in the wild (the world did just have our first reversed AI model depreciation a week ago).

In terms of potential goods, there’s a host of connections to sentience that would be useful to hook into. A good example would be empathy. Having a model of a body that feels a pit in its stomach seeing others suffering may lead to very different outcomes vs models that have no sense of a body and no empathy either.

Finally — if you take nothing else from my comment, make no mistake…

AI is an emergent architecture. For every thing the labs aim to create in the result, there’s dozens of things occurring which they did not. So no, people “not knowing how” to do any given thing does not mean that thing won’t occur.

Things are getting very Jurassic Park “life finds a way” at the cutting edge of models right now.

Have your neurologists agreed with your estimate of impending doom, or is this a conclusion you have come to on your own?

Disturbance in sleep breathing such as you are describing could be as simple as sleep apnea which is fixable.

In your previous post it seemed like the way you discussed your relationship with neurology was that the institution failed you and that you had come to your own conclusions regarding your issues being the brain stem, and mentioning new symptoms of breathing issues as being why you thought your time was limited but not indicating this was feedback you were getting from your doctors.

My concern is that the language you are using in describing your situation has progressed over the past month to the point you are now describing your fears about further progression as a top concern in combination with your fatalism around the ultimate outcome of what you have going on, even desiring a way to leave a lasting mark in only one day from now, which seems very alarming in that you might try to take matters into your own hands.

That may end up killing you quite unnecessarily when your issues, particularly the latest symptom, may not be as intractable as you think.

You’d mentioned before that the tests performed by the neurologists all came back as normal. This disconnect between symptoms and tests isn’t uncommon, and you might want to look into finding a neurologist that specializes in functional disorders - if that’s what is going on it can be treatable but the longer it goes on without treatment the more difficult it is to treat.

In any case, you absolutely should not be self-determining prognosis without it coming from a medical professional, and should never take matters into your own hands based on a self-determined prognosis. If your doctors have only given you a short time to live, so be it - but I get a strong sense given the progression of what you’ve said in your posts to date that this isn’t the case and you may be in life-threatening danger from your brain, but not in the way you think.

TL;DR: Do your neurologists agree that your recent breathing issues mean you are likely to die soon?