Matthias Plappert

I’m actually quite bullish on Europe: I think it’s culturally well equipped to deal with the onset of increasing automation due to AI. We already have strong social security systems that can buffer the disruptions and peoples lives are less focused on work.

Essays like the one from Dario (darioamodei.com/on-deepseek-...) miss the point on why DeepSeek matters so much to investors. The key issue for OpenAI and Anthropic (and any other VC model company in this space) is that they absolutely require outside investments. (1/n)

So how valuable really is the second part? Surely you have to discount this now as well. (Side note: DeepSeek also intends to build AGI so the first point gets indirectly attacked as well because there’s more competition; Dario realizes this and wants more export control for this reason) (5/n)

Essays like the one from Dario (darioamodei.com/on-deepseek-...) miss the point on why DeepSeek matters so much to investors. The key issue for OpenAI and Anthropic (and any other VC model company in this space) is that they absolutely require outside investments. (1/n)

So you have two possible angles: we’ll build AGI and this will be very valuable and/or what we’re building on the way is also very valuable. The problem is that the first angle is very risky because you don’t know who will build AGI and when. So even though valuable, you need to discount it. (3/n)

Essays like the one from Dario (darioamodei.com/on-deepseek-...) miss the point on why DeepSeek matters so much to investors. The key issue for OpenAI and Anthropic (and any other VC model company in this space) is that they absolutely require outside investments. (1/n)

Finally got around to read the DeepSeek-v3 and r1 papers in detail and I’m very impressed. They keep things simple and pragmatic but go deep and optimize when worthwhile. I think access to compute remains key but they demonstrate how much more can be squeezed out when faced with constraints.

It also increasingly looks to me like these models will get commoditized very rapidly. They also depreciate in value extremely quickly: Arguably the value of a frontier model is now only a few million dollars but it likely cost OpenAI et al many times more than that only a few months ago.

Finally got around to read the DeepSeek-v3 and r1 papers in detail and I’m very impressed. They keep things simple and pragmatic but go deep and optimize when worthwhile. I think access to compute remains key but they demonstrate how much more can be squeezed out when faced with constraints.

Excellent (old but still very relevant) article by @molly.wiki on crypto market caps and the pitfalls with these numbers. So if your friends or the mainstream media say things like “Trump made $10B in a few days” you can be the person that goes “Well, actually …”

I can now solve IMO problems thanks to AI but if I have to do any chore (schedule a meeting, make a reservation at a restaurant, collect and forward invoices to my accountant) it’s completely on me. This obviously sucks since I never want to solve IMO problems in the first place.

My main point being here: Most of the time I don’t need a PhD-level AI, I need something that can reliably solve real-world, mundane tasks that I don’t want to bother with. This is like building a self-driving car and benchmarking it on a race track when what I really care about is city driving.

I’m genuinely confused why we’re still arguing about the stochastic parrot claim. It never made much sense to me (or tbh most people who worked on these models) even back then. There’s also a lot of evidence against it. I think it’s also pretty straightforward to see why it’s unlikely to be true.

Basic sketch of the argument: 1. You have an extremely large, complex dataset 2. You have a large but much smaller model 3. You try to fit your data but you can’t fit it all -> need to compress

This Lex Fridman interview of Dario, Amanda and Chris is quite good (though unreasonably long): youtu.be/ugvHCXCOmm4?...

I wish Python had enums with associated values, similar to how enums work in Swift (literally the only feature in Swift I actually like)

In case you don't know what the f I'm talking about: docs.swift.org/swift-book/d... Again, I can only warn you about Swift beyond this one feature that I like.