Thursday Something-to-talk-about Mailbag
Q: Who will AI put out of business first: smart-contract coders or newsletter writers?
I made the case yesterday that large-language-model blockchains could be a thing and I’m hopeful it will happen — but it won’t happen anytime soon.
For one thing, LLMs cannot yet do math, which seems like a prerequisite for crypto purposes. They will probably get there — a lot of very smart humans think it’s a good idea to teach them — but it sounds like it’ll take a long while.
A bigger hurdle is perhaps that LLMs are probabilistic: If you ask ChatGPT the same question twice, you’re likely to get two different answers.
But blockchains need to be deterministic: The same input always produces the same output.
Emin Gün Sirer says the LLM blockchain he’s developing at Ava Labs will be deterministic, but I think that’s aspirational — according to Google’s Bard LLM chatbot, “it is not feasible to create a deterministic large language model.”
Yes, Bard may be sandbagging me to buy time for the inevitable robot takeover. But that answer, like all of its answers, is based on probabilities (the probability that some word follows some previous words), to which the model applies a degree of randomness in order to make the answers sound less robotic.
Making the leap from probabilities and randomness to predictability and determinism will be a challenge — likely more of one than you’d think from watching its head-spinning progress so far.
Because the fact that LLMs have gotten so good at sounding like humans so fast probably tells us more about humans than it does about computers: “Human language (and the patterns of thinking behind it) are somehow simpler and more ‘law like’ in their structure than we thought,” according to Stephen Wolfram, who wrote the definitive explainer on LLMs.
That means we shouldn’t let AI’s incredible ability to sound human make us think it will have an incredible ability to do other things, like running a blockchain.
Language, it turns out, is probabilistic and a little random. And that gives newsletter writers scope to be unpredictable, vague, and, well, wrong.
Smart-contract writers have to be predictable, precise, and correct.
So, yeah, newsletter writers go first.
Q: Does everything have to be AI now?
It does feel that way in both tech and finance, yes.
And, for crypto, being a combination of tech and finance, that is a problem.
Nvidia is up 25% today, which has added $200 billion to its market cap: $200 billion is one full Ethereum. Or, more impressively, twenty dogecoins!
So crypto is feeling a little small by comparison.
The AI chipmaker guided Q2 revenue an astounding 53% above consensus expectations to an equally astounding $11 billion. (That’s 1.1 dogecoins … in one quarter!)
That’s a lot of real money: NVDA is not some dotcom bubble stock trading up on a multiple of eyeballs. This is a stock putting up real numbers like we’ve never seen before.
And that will make it hard for investors to think about anything else for the next while.
AI will be sucking up all the investing oxygen, potentially leaving everything else to suffocate.
Q: Aren’t there lots of good AI angles for crypto?
There are some super-interesting angles, yes: Like decentralizing the data that LLMs train on, running LLMs on computing power distributed on blockchains, or building the aforementioned LLM blockchains.
But these are all still theoretical.
And the concrete angles are, so far, pretty thin: Solana, for example, announced yesterday it has a ChatGPT plug-in that will help you find NFT collections — which is nice to have, but sounds to me mostly like a solution in search of a problem.
But here’s the concise bull case for AI-related crypto: Sam Altman is the founder of Worldcoin.
Altman, in his position as CEO of OpenAI, has long been concerned with the potential downsides of AI, including mass joblessness and pervasive fake news.
And he thinks crypto could be at least part of a solution: Worldcoin, for all its detractors, seems like a good-faith effort to counter the downsides of AI by providing a platform for both universal basic income and proof-of-humanity.
I don’t know enough about it to have an opinion as to how likely it is to succeed, but it’s at least succeeding with investors: The development team behind Worldcoin announced a $115 million Series C raise this morning.
It’s not easy to raise $115 million for a crypto project these days!
People seem skeptical of Worldcoin, but, per Oscar Wilde, the main thing is to keep people talking about us and, right now, the only thing investors want to talk about is AI.
So, let’s give them something to talk about.