Thomas MyrdenMar 11, 2023
sanae   (SF)

I have a theory that, while LLMs (Large language models, like ChatGPT) don't really reflect how most people think, reason, and write

However, they do reflect how VCs tend to think, reason and write: they don't actually understand what they are saying or what they are doing, and their abilities are largely an illusion based on a mix of having a staggering number of resources at their disposal, and also on vacuuming up other people's work and claiming it as their own

Mar 10, 2023 at 11:57pmWeb
Show threadsuldrew 🚲🏳️‍🌈Mar 10, 2023
@sanae also, they depend on a huge amount of cheap compute, preferably paid for by a sponsor who also happens to operate a giant cloud service
Show threadCassandrichMar 11, 2023
@sanae 💯
Show threadKevin RiggleMar 11, 2023
@sanae yes but I think the lack of ground-truthung/need right same is the key
Show threadLauraMar 11, 2023
@sanae Dan McQuillan on Tech Won't Save Us said a similar thing, he said they are "bullshit engines"
Show threadSam WronskiMar 11, 2023
@sanae I don't think OKRs are compatible with widespread AI adoption either. That's going to need to change or we're going to over optimize on metrics not people.
Show threadit's B! Cavello 🐝Mar 11, 2023
@sanae this is probably meant as a 🌶️ funny take, but ngl it resonates with me.
I’m not a VC, but I also do relate to feeling empathy to some ML architectures. The more I learn, the more I find recognition. When I learnt about latent space, for instance, it made many of my own thought processes feel more legible. It’s not to say that we actually are the same, but I wonder if some people do really think more in these kinds of architectures, while others maybe totally don’t!
Show threadsanae   (SF)Mar 11, 2023
@b_cavello I think there are things that every human brain does which resembles what LLMs do, but a LLM couldn't come up with the post you just wrote unless a lot of other people had written something similar and then it could repeat the ideas back. Introspecting on the ways your thought process do and don't resemble a particular architecture is definitely not something LLMs are designed to do!
Show threadsanae   (SF)Mar 11, 2023

@b_cavello

I think people are quite capable of imitating a body of writing they have already seen in a kind of LLM like fashion, but that's just one thing that people can do. Some of it is snark, but also I do think that the kind of tech startup thought leader space is largely about observing what other people think success looks like and then imitating its characteristics in a convincing manner. It's a thing everyone does, but it's a lot more core to that industry.

Show threadJaycieMar 11, 2023
@sanae The hunger of the Ouroboros grows.
Show threadDan Cassidy 🦌Mar 11, 2023
@sanae I've met several actual humans who "think" and speak exactly like ChatGPT. They're convincing when you first meet them but in time it becomes clear that they're just regurgitating a mashup of things other people have said without understanding any of it. All of them have been business owners, investors, or upper management.
Show threadDeanna DayMar 13, 2023
I keep saying that ChatGPT doesn’t tell you the “truth,” it tells you what a particularly smooth Penn undergrad would tell you if you asked them a question in class and they didn’t know the answer. In other words, it’s just bullshitting.
Show threadleading cackle influencerMar 16, 2023
@sanae Burnnnn
Show threadleading cackle influencerMar 16, 2023
@sanae but also: accurate
Show threadR. L. Dane  May 6, 2023

@sanae

#ApplyAloeToBurn ;)

Show threadDaniel AndrlikMay 7, 2023
@sanae All of THIS. Spot on.