Andreas Kirsch 🇺🇦

My opinions only here.
👨‍🔬 RS DeepMind 1.8y, Midjourney 1y 🧑‍🎓 DPhil AIMS 4.5y 🧙‍♂️ RE DeepMind 1y 📺 SWE Google 3y 🎓 TUM
👤 @nwspk

[Andreas Kirsch 🇺🇦 @BlackHC]

I work at Google DeepMind. This won't make me popular. But it's all public reporting:

2014: DeepMind reportedly sold to Google on conditions: no military use, independent oversight

2026: a Pentagon contract for "any lawful government purpose"

Not one safeguard survived intact

[Andreas Kirsch 🇺🇦 @BlackHC]

@DaveShapi @grok So why is pharma not open-source? What's the difference? Do you get that?

[Andreas Kirsch 🇺🇦 @BlackHC]

Person says the truth and gets crucified for it 🤷‍♂️

100% logical that open models are deaccelerationist. They'll squeeze everyone's R&D budget and are a good way to slow things. That's another great reason to support open-weight models

(It's crazy how many bad takes and attacks have been written on these observations)

[Dean W. Ball @deanwball]

Some observations on Kimi:

1. It's a very good model! I don't think its performance can be explained away by distillation or anything like that. In agentic coding sessions, it seems pretty much on par with the best public models of Q1 2026. In my fairly limited use, it also seemed very token hungry. It's not obvious to me that this model is actually that cheap to run.

2. I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. To be clear, I *myself* might be fine with models presenting this level of marginal risk being open weight, but I am surprised that China is fine with it. I suspect the reason they are is 75% explained by strategic blindness/lack of AGI-pilledness (the CCP is very Yann Lecun-y in its views of AI). The other 25% or so is their lack of compute for customer inference (making China's open-weight strategy an unintended byproduct of US export controls) and the normal Chinese strategy of aggressive exports. For the companies, as opposed to the government, the decision to open source is partially ideological and partially because they are behind, and they know that very few people would pay for sub-frontier models from China.

3. Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models. I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models create over the whole of AI. It's not a bad strategy; it reminds me of James Scott's recounting of the hill people in "the art of not being governed." Still, in the end, open-weight models deter further AI capex.

4. One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state as a kind of "digital public infrastructure." This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end. You'd be surprised how many 'accelerationists' lobbied me, while I was in government, to support an eleven or twelve-figure federally funded data center so that startups could train models at a subsidy and then give them away for free. There was no other way for AI to progress, they said. Perhaps this is the logical end state of things. Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome. I think many of them just don't know what they're doing. Many accelerationists do not view the creation and serving of frontier models as a legitimate business.

5. I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don't need to "ban open source" (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There's a happy middle ground here. I'd assume they will do some version of this.

6. It's probably true that open-weight models of this capability make the world a bit more dangerous, but not so much more that you'll really notice. At some point the models will be capable enough that you will notice. "A nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a Chinese lab," you say? Color me shocked.

[Andreas Kirsch 🇺🇦 @BlackHC]

@lxjren I'm pro competition. Open-source is closer to "communism" than free market for obvious reasons. And very clear when you look at the origins of it and the copyleft movement. I'm not saying that's bad btw. Just your argument is a bit muddled

[Andreas Kirsch 🇺🇦 @BlackHC]

@radagaisus I can only think of where slightly closing things down sped up progress and that's the patent system when you look at its history (not always obviously)

[Andreas Kirsch 🇺🇦 @BlackHC]

@sdmat123 Different concerns about safety risks. So on the one hand, it slows down new risks, but on the other it proliferates/diffused existing ones

[Andreas Kirsch 🇺🇦 @BlackHC]

I guess the "crux" is what we mean by acceleration/accelerationism. Obviously distribution benefits, and derivative research efforts as well. Otoh, if eg open models are sufficiently good to capture most economic value then there is little incentive to pay for closed ones that subsidize research research efforts beyond that. Also from a safety perspective that is great bc a lot of concerns around AGI/ASI come from developing models that go way beyond being sufficiently economically useful right now

[Andreas Kirsch 🇺🇦 @BlackHC]

@Dave_Kayac @michalwols @aran_nayebi How do you define acceleration?

[Andreas Kirsch 🇺🇦 @BlackHC]

@neuralisa We currently also have advertisement communism in the ecosystem of you think about it 😅

https://t.co/oo3sXFwHpH

[Andreas Kirsch 🇺🇦 @BlackHC]

I think he makes quite a few good points also re "AI communism" which iirc refers to horizonal(?) investments if direct revenues collapse. The only ones making a lot of money will be inference and compute providers in the case that the best models are open (or sufficiently good to extract most economic value) iiuc

[Andreas Kirsch 🇺🇦 @BlackHC]

@heavyhelium11 https://t.co/oo3sXFwHpH

[Andreas Kirsch 🇺🇦 @BlackHC]

I think he makes quite a few good points also re "AI communism" which iirc refers to horizonal(?) investments if direct revenues collapse. The only ones making a lot of money will be inference and compute providers in the case that the best models are open (or sufficiently good to extract most economic value) iiuc

[Andreas Kirsch 🇺🇦 @BlackHC]

Up to a point. Some innovation requires big upfront commitments, while many other innovations do not. The latter obviously benefit but if most of the economic value is provided by open models then the only ones making a ton of money in the end are the compute providers which then changes how R&D is invested (they get to decide who to invest in and/or will be the ones being able to make big commitments)?

I think this indirectness is what he refers to as AI communism? Maybe similar to how we have advertisement communism in other parts of the ecosystem where ads cross-finance totally different products that users care about with less direct revenue mechanisms

[Andreas Kirsch 🇺🇦 @BlackHC]

I think he makes quite a few good points also re "AI communism" which iirc refers to horizonal(?) investments if direct revenues collapse. The only ones making a lot of money will be inference and compute providers in the case that the best models are open (or sufficiently good to extract most economic value) iiuc

[Andreas Kirsch 🇺🇦 @BlackHC]

@Dave_Kayac What does accelerate mean to you here?

Maybe we're talking about different effects with different saturation points?

How many open models and how much access does the ecosystem need to get most of the benefits of it re research inventions etc?

[Andreas Kirsch 🇺🇦 @BlackHC]

@michalwols @recurseparadox Bad take

[Andreas Kirsch 🇺🇦 @BlackHC]

@recurseparadox Really? It still works as intended in many areas. Software patents have issues bc of the way trolls can abuse it but patents were historically very effective at the very least

[Andreas Kirsch 🇺🇦 @BlackHC]

@recurseparadox Why was the patent system invented again? Patents were actually accelerationist in their effect (and intention)

So try again? 🙃

[Andreas Kirsch 🇺🇦 @BlackHC]

@anonid3430 @ArielKwiat Who would do that? And invest in what precisely?

Compute spent on inference is not compute spent on pretraining

For "acceleration" you need research and pretraining runs that require coordination

[Andreas Kirsch 🇺🇦 @BlackHC]

@michalwols @aran_nayebi I agree that for post-training this is true

I think for pre-training, it's difficult to see how to make back the money at some point?

[Andreas Kirsch 🇺🇦 @BlackHC]

Yes true but you also need to be able to spend huge amounts on new training runs and for that to be worth it, you need sufficient margins? Not sure but how do you see these investments happening

Most scaling breakthroughs didn't come from academia because it's impossible without larger/dedicated budgets (GPT, Chinchilla, AlphaGo, even DQN)

(Only my personal opinions here)

[Andreas Kirsch 🇺🇦 @BlackHC]

@chontang @deanwball Notice how all scaling breakthroughs came from academia which has so much capital and compute available..?

You are comparing apples to oranges re what succeeds here

[Andreas Kirsch 🇺🇦 @BlackHC]

This doesn't answer the question. You need large upfront investments in research and infra to train models which is different from inference

The alternative is that the companies that sell the "shovels" will also finance that I suppose. So that could be Nvidia and maybe also Google (no idea re TPUs)?

Everything here is only my opinion and on my behalf ofc