Yeah absolutely, I have no objections to you looking at my thought experiments and saying "No, if you replaced every neuron with silicon I would stop being conscious". That's a totally valid conclusion and is a mainstream philosophical position. That's why they're thought experiments.
My issue was more just that people see the concept of a thinking machine and immediately reject it because it sounds silly, without thinking about it.
I shouldn't get in arguments about this stuff online, but have you actually sat down and thought about this in-depth? It's pretty normal to have a gut reaction that this is insane, but it's worth thinking about.
If we replaced every neuron in your brain, one by one, with a silicon equivalent, would you stop being conscious? At what point? Would you be able to tell?
If yes, what about if we took those silicon neurons and instead simulated them, IE perfectly simulated an entire human brain, neuron-for-neuron, would that be conscious?
If yes, where is the line at which code starts being conscious? Can you categorically say that existing LLMs are on one side or the other? Would scaling up or down change that?
There are robust philosophical positions on all sides of this argument.
The set of models that are pareto-optimal, IE for some set of variables, no other model strictly dominates them = no other model is better than them on every variable.
So like, on a cost-intelligence graph, the cheapest and most intelligent models are pareto optimal. Then in-between those if you have
- cost $3 intelligence 6
- cost $1 intelligence 5
- cost $2 intelligence 4
The 1st and 2nd are pareto optimal, the 3rd is not, because it's dominated by the 2nd (2nd is cheaper AND more intelligent at the same time)
I'm pretty sure that's the plan. Currently they're legally bound to keep it within 0.9s of solar noon (or something like that) but in 2035 it's changing to +-1 minute, which basically kicks the can down the road for another century or so
I say we let it reach 15 minutes then countries can solve it themselves by shifting timezone by 15 mins. Since making sure solar noon matches noon on the clock, is the point of timezones existing in the first place
I thought about this more and realised your question might have been "what's the difference between knowing and learning". IE, how can we say the model believes something without having been taught it.
I think you're right that they're basically the same thing. I'd argue they're very slightly different because what an AI model ends up knowing isn't perfectly predictable based on what they were taught (emergent intelligence), but the sentence you quoted is using believing and learning to mean the same thing, it's just trying to draw attention to the fact that the training process structurally enforces "cheat as much as possible without getting caught".
IE, the contrast in the original sentence wasn't "believe" vs "learn", it was "good" vs "permissible"
Believing and knowing are overlapping sets, imagine what you think of when someone says an AI "knows" something, it's the same mechanism (I'd describe it as something along the lines of "encoded abstractly in the weights")
I think we'll get there. Right now it works for me, because I'm naturally pretty verbose in my prompts, and know the codebase well, so I know what it needs to look at. Plus subagents for anything exploratory.
I think deepseek v4 pro has 1m context and does pretty well up to around 600k. But if you have the hardware to run that locally, you already know
Even then if there's a smaller model with 1M context, you'll need a ton of RAM to actually run it at full 1M. I guess that's why you don't see it too much. Anyone that could run Qwen 3.6 27B with 1m context would be better off running a much bigger model with smaller context instead, in the same amount of VRAM.
In terms of optimizing further, huge context + KV quantization sounds like a terrible idea, but there's some decent innovation in sparse attention, KV cache rotation allowing Q8 to perform nearly as well as full 16-bit precision, plus some ideas around offloading KV cache to system RAM (but I'm skeptical)
Yep, I daily drive Qwen3.6-27B (including for work), have done pretty much since it came out. IMO it's the only (small-ish, local) model worth using, if you can run it. It might not be as good as Opus at "add X large feature" but I don't want that in a model. I want to do the thinking while it does the typing. And Qwen 3.6 27B is perfectly good at that (while in my experience models like the 35A3B and gemma are significant downgrades)
Plus, I never have to worry about rate limits, quotas, or sitting in a queue during peak time. And I can always see its full thoughts, don't have to worry about where my data is getting sent, and know it can't get secretly nerfed.
Running on 2x 3090, 500-1000tok/s prefill and 60tok/s output at Q6_K_XL with MTP on llama.cpp, 220k tokens context window (starts to get a bit dumb above 160k ish), no KV quantization
The problem is, people see "they're not profitable once you account for training" and equate that to "AI will go away soon"
But if all the AI companies stopped training new models, they would all instantly become profitable (and stick around)
The thing that makes them unprofitable, is having to compete (which means training models). If / when enough companies exit the market, the cost to compete goes down and you end up in an equilibrium
If you have ASI that follows instructions, you can just instruct it to not get stolen and then it won't get stolen. Most logic / intuition breaks down with ASI.
My first instinct was that the essay would just be "67" as a stupid and harmless but nonsensical response.
Somewhat amusingly, mine depends on the examiner knowing how advanced AIs are. In the 1960s mine would just look like a trickle AI. It feeling human demands we assume the ai would actually be competent
Yours is even more effective. Both hinge on the solution being "be as unexpected and out-of-distribution as possible"
I somehow imagine they wouldn't like your essay that is made of 100% slurs though, regardless of how effective it is at the stated task
This thread started because of "the cheapest bridge that just barely won't fail"
My point was that safety factors are a part of this. A safety factor of 1.0, designing bridges so that they can perfectly withstand the expectations of intended use, means that some unacceptable % of those bridges will fall down in practice.
In other words, it's true that you can explain safety factors as:
> Assuming perfect construction, and no defects, under designed maximum load, make sure that this bridge really stays up by a wide margin
But that misses the point of why we use safety factors. Nobody is paying for a bridge to really stay up by a wide margin. Because there's no material difference between a bridge that stays up, and a bridge that really stays up, right up until the point that the weaker one falls down due to inevitable over-loading or defects in construction / materials.
[email protected]