There's no way that it hasn't existed for infinity, for the value where universe = literally everything. It can't come from nothing, so it has always existed. I am obviously not aware of the formal definition of universe, considering there are theories of our universe being one of many bubbling in the cosmic foam.
> There's no way that it hasn't existed for infinity
I'm not disagreeing, but this is _slightly_ shifting the goal posts. When scientists in this field talk about the 'age of the universe', it's nearly always the short form of 'the time distance from now back to the big bang, beyond which there is no way to see beyond'.
Sure, I explicitly shifted them from formal physics definitions, where "nothing" is actually "something", e.g. in the case of your cited book, the "nothing" is still a set of quantum laws and fields. How did those arise?
IIUC, There is no good reason to believe that they 'arose' at all, or inversely no good reason believe that they did not. It is likely unknowable. We don't know that there ever _could_ be a 'nothing' void of fields, etc. No one has demonstrated either that such a situation could arise, and no one has demonstrated that such a situation is necessary.
It's an interesting concept that I read a few years ago on HN. Basically, if you add 1 to a number, you get a bigger number. You never get to infinity. You can keep doing this until the heat death of the universe, and you still won't get to infinity. So why do we think infinities exist? We know the universe started, it's reasonable to conclude it will also end, and is therefore not infinite. And if the universe is not infinite, it cannot contain infinities, by definition. Which matches our observations; we don't see any infinities in the universe.
Time is part of spacetime, and spacetime is a property of this universe. So there's no "when" or "where" outside the universe. So it doesn't make any sense to talk about "before" the universe. Time started when the universe started. Causality is a property of time; one thing follows from another because of time. If there's no time, there can't be any causality.
> How could it start from absolutely nothing?
So this question doesn't really make any sense. There's no other way it could have started, because the very concept of "nothing" and "started" didn't apply outside of the thing we're saying started (sorry, it gets very difficult to talk about this using a language so firmly grounded in time). It doesn't make any sense to ask what caused the universe because causality itself is a property of the universe.
Like I said, interesting concept and thought. I have no idea if it actually describes reality.
What always confuses me about Anthropic is them insisting that they need more compute.
This feels wrong in several ways:
a) they're certainly not addressing their customers like they're compute-limited. One must ask, what would that look like? For one, I think it'd look like a wait list to sign up. It'd look like prices going _higher_, and not lower
b) IIUC, their revenues don't cover their _current_ compute costs. So, let's pretend coreweave magics up a new dc and hands it over to anthropic, which means presumably anthropic needs to start paying for it. How does that affect anthropic's revenue? AFAICT, it just adds costs, it doesn't make their revenue go up.
The only interpretation I can find is if Anthropic is _convinced_ that they need to train faster/harder, and that doing so would somehow significantly shift their TAM or costs, for example by coming up with a model that was simply 10+ years ahead of anyone else that costs the same as opus 5.5 to run, and they instantly nab the entirety of openai's market share. Seems pretty unlikely.
Anthropic’s unit economics are really good, so I don’t think (b) is true. Point (a) may be valid, but they’ve also grown consistently, and even their slower growth rate in Q3 of this year is still very strong. To keep up growth, they likely need to procure compute months ahead of time as well, so it makes sense for them to do this before they are actually limited.
I assume some of this is the escalating cost of training, indeed, and that being faster means abandoning efficiency.
But you’re right. I have often wondered why it is that such a small amount of extra compute (300MW) from SpaceX was entirely enough for them to uncap so many customers.
I think the (in-house) data centre burdens of ever-more complex models may be considerable, but I think the data centre businesses expecting to serve inference tasks are probably heading into a Chinese real estate bubble.
So I see an interesting side effect of all of this in your comment. The groupthink seems to conclude that the root of GH's issues are:
- A dramatic increase in its traffic due to vibecoding
- Azure just generally not being very reliable
Now, if you're not a paying customer, maybe you just ignore this and chalk it up to "I'm getting what I paid for". And if you're a paying customer, you migrate away. What happens to GH then? Well, you (and other paying customers) are probably only a small part of the first of the identified causes, and not at all a contributor to the second. So it's not like you moving away will make GH more reliable, but it does lessen MS's motivations for fixing things: GH as a business unit works by having a free tier and a premium tier and that's fine as long as the profit from the premium tier exceeds the cost of the free tier.
THat's long winded way of saying: I wonder if you/other paying customers leaving is the real start of a GH death spiral. Less paying customers == less MS investment == less paying customers == less MS investment, etc.
For us it's not just about reliability, but risk exposure. The last couple of years really changed what our security model looks like. And it's not in public clouds, on SaaS products. LLMs are killing SaaS for sure, not by replacing them but making them a concentrated risk.
> A human proof won't just be logically correct, it will be cognitively distilled and coherent. It may still take years to understand it, but the logical flow will be straightforward, not obfuscated.
Not a counterpoint, actually case in point, because:
>Mochizuki and a few other mathematicians claim that the theory indeed yields such a proof but this has so far not been accepted by the mathematical community.
A B300 has 8 gpus, 2? xeon processors, and 2 fancy nics. Assuming all of those computer chips are equal value cause averages, that's ~ $4k per chip which fits the 'thousands per chip' pretty well.
a) we don't actually know what the prices will look like yet,
b) what about same weights + huge context? or, same weights that you'd run on 128gb/256gb, but multiple models running for different tasks?
I think it's safe to start the conversation as about bad as the jump from 256 to 512 on the M3, which was a little more than double base to 256. If it's surprisingly different at launch then it can be a party, but there is no sense getting your hopes up for that at the moment.
Longer context also slows token prediction proportional to the context size. If it wasn't regularly referenced then there would be no need to keep it in RAM.
Usually the pitch for more memory is "I can run a massive model/context and get my answer in a while instead of next weekend from disk".
> And tyeah... open weight models cant compete with likes of astra yet
I think this is true, but also misses that a lot of us are just doing basic flask apps with a react front end. We don't need astra; Something sonnet 4.6 level locally is perfectly sufficient 95% of the time, and maybe 99% of the time.
This. People have convinced themselves that the absolute frontier is what is needed, anything below it is an unacceptable compromise, and we seem to be speaking different languages when it comes to discussing model capability.
It's like watching a discussion about cars available to take on a 100km road trip. A new car gets released that is on par with a Toyota Corolla but it is dismissed as completely useless for a 100km trip because it doesn't have the seat massagers and air ride suspension that the new Escalades have.
The reality is that something like Sonnet 4.6 is still amazingly capable for so many programming tasks, especially if you already have some reasonable level of experience to steer it in the right direction.
And if you think Sonnet 4.6 is still worthwhile, then it seems undeniable that something like Qwen 3.8-27B is also worthwhile.
I'd settle for any number of necks. Currently, when a corporation fucks something up, breaks the law, or hurts or even kills people, there aren't consequences besides a tiny token fine and a strongly worded letter telling them to not do it again or they'll get another tiny fine and letter, and their CEO might even have to sit down in front of Congress to say a few words and look sad.
I feel like our legislators would never get this far. They really don’t seem to care, maybe after “their emails get hacked”, but do you find it likely for this to actually pass into law?
Not any reputable ones. The consensus on the estimated age of the universe has shifted only slightly in the last several decades.
reply