CEO of Composer Technologies. opinions are my own / not investment advice
Toronto
Joined January 2011
- Tweets6.3K
- Following1.2K
- Followers3.2K
- Likes45.7K
Tokens have no intrinsic value. They are merely a means to an end. Tokens are useful if they do economically valuable work. If it takes less tokens to do a task, and/or the price of a token falls, then the value of a GPU hour *increases* because you can do more economically useful work with that capacity.
Why are people still hung up on falling token prices
Falling price per FLOP or GB over the last century didn’t exactly crush the computing industry
i think what's missing in the argument that LLMs are commodities that face disruption from China/open source is the tremendous heterogeneity of supply - totally unlike homogenous commodities like oil etc.
There is also extreme dispersion in willingness to pay because of heterogneity of demand/use case. There are lots of consumers who will always want to pay $0. There are also industrial scientists for whom a million $ is a small price of entry for potential ROI.
In equilibrium would expect a very diverse, segmented market with a lot of price discrimination.
You can just ignore token prices and look at marginal product per GPU hour to determine where price of compute will land. Just like wages converge at marginal product of labor.
Allegedly “the most important chart” for understanding the current state of the AI economy. The model layer seems to be going through a period of commodification as non-frontier models saturate everyday use while frontier models compete fiercely. Compute layer sees massive increase in total demand for compute regardless of the source or use case.
before enlightenment, watch cricket, label data.
after enlightenment, watch cricket, label data.
Mercor is paying $20/hour to watch Indian cricket matches and help train AI models.
This might be the coolest way to make easy money.
The chasm from seed to series A sucks in short term for founders, but is to their benefit in longer term. You don’t want to get buried in lit pref in something that can never have a venture scale outcome or you end up wasting years of your life.
Most businesses are simply not venture scale and better to discover that sooner than later.
like Supertake, certainly not a novel idea. But wonder if timing is finally right...
Robinhood and Revolut democratised access to financial markets.
But access ≠ sophistication. AI has the potential to change that.
Thrilled to share that @Sequoia has led @CatalystLabsX's $30M seed round.
Benjamin P Rollert retweeted
Quant firms are hiring idea guys, not quants
I've been skeptical about the whole - software is commoditized, ideas are the moat argument
But I've seen first hand the best quant and technical firms in the world hire thinkers, liberal arts grads, and generalists
I predict the trend accelerates
Letter from @AQRCapital below
i dont think i've ever liked the aesthetic of anything generated by AI ever, and I'm an AI maxi when it comes to its impact
Benjamin P Rollert retweeted
AI is primarily a production tool, not a consumer good. That means that AI will show up in GDP as part of the value added in other sectors like health care, entertainment, and manufacturing.
9% is definitely not unrealistic.
To justify today's AI capex, we'll have to spend 9% of GDP on AI services, a study finds. Is it plausible we'll spend as much on AI as on food? Twice as much as on energy? The law of diminishing returns would like a word. My column: wsj.com/tech/ai/will-america…
I've been going down this rabbit hole of people deconstructing Jimi Hendrix original version of Little Wing. It's one of the few recordings that not only doesn't sound dated with time, it actually gets more brilliant the more you listen to it and break it down. And even the best studio musicians admit they can't replicate Jimi's feel to this day. It's like Jimi connected with an alien intelligence that day, and left all these easter eggs in the form of a recording.
