Frankly I think it’s kind of childish to just put up a massive Uk wide block on your website. “Call your representatives”, ok dude, can I give you a list of things I want to change about your country’s policies?
Needs a [2010] tag. In almost all modern hardware development you'll have coding guidelines along the lines of "Always use blocking assignments for comb logic, always use non-blocking for sequential logic". You end up back at the same place as VHDL, by nature SystemVerilog is much weaker typed than VHDL. So you have to just have conventions in order to regain some level of safety.
What do you mean by simulate? Do you want the language to be aware of the temperature of the silicon? Because I can build you circuits whose behaviour changes due to variation in the temperature of the silicon. Essentially all these languages are not timing aware. So you design your circuit with combinatorial logic and a clock, and then hope (pray) that your compiler makes it meet timing.
The fundamental problem is that we're trying to create a simulation model of real hardware that is (a) realistic enough to tell us something reasonable about how to expect the hardware to behave and (b) computationally efficient enough to tell us about a in a reasonable period of time.
I guess this really depends on your view of the world. Was Marc Andreessen some visionary without whom no one would've ever figured out images could appear on websites. Some kind of Albert Einstein of cat gifs. Or was the img tag an inevitability once the web had enough bandwidth to transfer images.
There are obviously tonnes of accurate stereotypes in the TV show Silicon Valley, but one of the ones I think about often is when Richard calculates how much money Russ Hanneman has made investing his billions... and it works out to less than sticking it in the bank.
You've got all these silicon valley guys running around "venture investing", the truth is it's more of a life style than a money making exercise. They made their money decades ago, and now they're just sort of hanging around desperately trying to tell everyone how clever they are.
I'm always a little surprised at how these Tech CEOs are willing to go on TV and just spout nonsense. Firstly, 40% of college educated white women voted for Trump at the last election. Secondly, isn't the entire theory of Trump's support amongst working class voters an appeal to economic populism due to an erosion of their economic position? Aren't you literally describing a process that last time lead to a massive political shift in favour of those who were negatively economically impacted? Oh and you think all the white collar workers are going to lose their jobs, but you don't think that's just directly going to cause a recession that wipes out blue collar republican jobs?
It's difficult to (a) see how he can say this having given any real thought at all and (b) understand why he's going to on news interviews and winging it.
Whilst this is interesting I find the topic bought up on odd lots is more interesting. The idea was this: Once you've built a model, if you can sell tokens for a profit, this is a great business - just sell more tokens. But you can't just build a model and sell tokens. You need to build the best model to sell new tokens. So the question is much more "How much does it cost you to build a new SotA model" and then "How effectively can you monetize it". And since you need a SotA model, your only option if you have a bad model that isn't selling is to invest billions more into building a better model whose tokens you can sell.
So this turns into a death march.
If you are behind, the only thing you can do is make massive capital investments to catch up. Once you're ahead you can sell tokens until someone else catches up. And, breaking the model of normal of places like chip fabrication, your billions of investment may only keep you ahead for 2 months. So you have a tiny window to sell those tokens.
What you are talking about isn't inference cost. Yes, fundamentally what matters is all the work that goes into the models, including R&D, training, and inference.
But we talk about inference separately for a reason: largely inference cost is the scaling cost. Once you have a model the margin on your inference is how you get to profitability, as long as your margin is positive you can make the entire enterprise profitable by just selling more tokens. This is the same fundamental business that chip fabs work on. Yes it costs them a lot to get to the next node, but what's important is the margin they can get on the wafers they sell, because they sell tonnes of wafers.
It's pretty core to the concept of SAAS businesses that yes, you do consider all costs. But you want to focus on the margin of the bit that scales. This is why WeWork exploded, the thing they were scaling only scaled up at negative margin.
The point is that if their inference margin is positive, they can "just" scale up and become profitable. If their inference margin is negative, then scaling up the business actually causes problems.
There's two points here. The first is that a strategy of monetizing models to fund the goal of reaching AI is indistinguishable from just running a business selling LLM model access, you don't actually need to be trying to reach AGI you can just run an LLM company and that is probably what these companies are largely doing. The AGI talk is just a recruiting/marketing strategy.
Secondly, it's not clear that the current LLMs are a run up to AGI. That's what LeCun is betting - that the LLM labs are chasing a local maxima.
reply