Daniel Privitera

Making AI go well | Founder @kira_center_ai

[Daniel Privitera @privitera_]

Should we choose AI progress or AI safety? Address present-day impacts of AI or potential future risks?

In our op-ed for @TIME , Yoshua Bengio and I argue that these are false dilemmas.

And we propose a “Beneficial AI Roadmap” (BAIR).

1/n

[Daniel Privitera @privitera_]

New report by @MonikaSchnitzer , my @kira_center_ai colleague @philip_fox_, and me: Germany already has very little AI compute, things are on track to get dramatically worse, and urgent and decisive action is needed to reverse this trend.

Germany currently has less than 2% of global AI compute. And my impression is that in Berlin it has not yet fully sunk in how bad things are about to get if nothing changes. One or two potential new German “AI Gigafactories” are the most ambitious things being discussed, but one such Gigafactory would still be at least 50x smaller than the biggest clusters announced internationally. Germany's projected compute capacity (existing + announced) does not get anywhere near to closing the expected gap in demand in coming years.

So what should the Merz government aim for instead? We outline 3 possible approaches. The appropriateness of each approach depends on how seriously the government really takes AI and how ambitious it really wants to be, beyond talking points and buzzwords.
- Low ambition: aim for enough compute to deploy AI in selected sectors
- Medium ambition: aim for enough compute to deploy AI across society
- High ambition: aim for enough compute to deploy AI across society and develop frontier models

Even the low ambition approach could not be realized, by far, with the currently announced clusters. And all three approaches will require proportionally decisive actions beyond just building datacenters.

So urgent action is needed. We make 6 recommendations that make sense irrespective of the ambition level with which the Merz government wants to approach AI:
- Expand AI compute capacity according to strategic priorities
- Create political enablers, e.g. centralized steering
- Secure cheap, reliable energy
- Accelerate planning/permitting
- Ramp up data center security
- Strengthen the domestic AI ecosystem

Feedback is very welcome – what do you agree/disagree with?

[KIRA Center @kira_center_ai]

New KIRA Report by @philip_fox_ @MonikaSchnitzer (LMU) & @privitera_
Germany is lagging behind on AI infrastructure: Its share of global AI compute (<2%) will shrink even further if current trends continue. But with decisive action, the Merz government could reverse this trend.🧵

[Anton Leicht @anton_d_leicht]

Very excited to announce I've joined @CarnegieEndow as a Visiting Scholar with the Tech & IA team!

This means I get to take my work on AI & political economy full-time: I'll keep writing ~weekly blog posts on fast AI progress, slow institutions and thorny incentives. (1/3)

[Daniel Privitera @privitera_]

Most frontier AI companies, including @AnthropicAI , @Google/@GoogleDeepMind, @Microsoft, @MistralAI , @OpenAI , and @xai have signed the Safety & Security Chapter (which I co-authored alongside 8 great co-chairs) of the Code of Practice. I think this is great news for a number of reasons:

- It allows these companies to comply with the EU AI Act in a streamlined way, provided that they actually adhere to the Chapter in full.

- It gives the regulator (the Commission) greater visibility into how these companies will comply with the Act. I hope this transparency can contribute to increased mutual trust and a productive relationship between signatories and the Commission.

- By signing and committing to the Chapter, these companies also give the public more visibility into how they will approach risk management for cyber, CBRN, and other large-scale risks. This is valuable, and it will be interesting to see whether frontier AI companies that don’t sign the Safety & Security Chapter will offer a similar level of transparency to the public.

The objective here was to create a practical tool for streamlined compliance with the AI Act that is of real use to frontier AI companies, while also being effective at meeting the AI Act’s “systemic risk” obligations. I think we achieved this.

This would not have been possible without the thoughtful engagement from companies, civil society, and academics throughout the process – and without the Commission taking the somewhat bold decision to outsource this work to external experts. Thank you to everyone involved!

This is only a starting point, of course. Much more needs to happen for the EU's regulatory regime for frontier AI to go well in practice. My co-chairs and I shared some thoughts on this here: https://t.co/1jISQB4rux

[Yoshua Bengio @Yoshua_Bengio]

I’ve been thrilled to see the support for the Safety & Security Chapter of the Code of Practice. Most frontier AI companies have now signed on to it: @AnthropicAI, @Google, @MistralAI, @OpenAI, @xAI
Why this is important: 🧵
1/6

[Yoshua Bengio @Yoshua_Bengio]

I’ve been thrilled to see the support for the Safety & Security Chapter of the Code of Practice. Most frontier AI companies have now signed on to it: @AnthropicAI, @Google, @MistralAI, @OpenAI, @xAI
Why this is important: 🧵
1/6

[Markus Anderljung @Manderljung]

The EU's Code of Practice for General-Purpose AI is out. As one of the co-chairs who drafted the Safety & Security Chapter, focused on frontier AI, I'm proud of what we've put together. It’s a lean but effective framework for frontier AI companies to comply with the AI Act.

[Yoshua Bengio @Yoshua_Bengio]

The Code of Practice is out. I co-wrote the Safety & Security Chapter, which is an implementation tool to help frontier AI companies comply with the EU AI Act in a lean but effective way. I am proud of the result!
1/3

[Daniel Privitera @privitera_]

🗞️ @MonikaSchnitzer und ich haben für die @faznet einen Gastbeitrag über AGI geschrieben: KI, die praktisch jede kognitive Aufgabe gleich gut oder besser löst als ein Mensch.

Wir sind besorgt darüber, dass in der deutschen Politik keine ernsthafte Auseinandersetzung mit AGI stattfindet. Die meistzitierten KI-Forscher der Welt, die führenden Köpfe aus der Industrie und immer mehr Politiker in anderen Ländern: Sie halten AGI schon in den kommenden 4 Jahren für möglich und bereiten sich auf dieses Szenario vor.

In Deutschland passiert hingegen: nichts. AGI ist wie der Elefant im Raum, der entweder ignoriert oder gar nicht erst gesehen wird – während man in D.C., Peking, San Francisco und London die Möglichkeit sehr ernst nimmt, dass AGI bald erreicht wird, und sich für dieses Szenario vorbereitet.

Im Beitrag beschreiben wir, was die Kernelemente einer ernstzunehmenden AGI-Strategie der Bundesregierung sein sollten: Rechenzentren ausbauen, kluges Ausspielen der deutschen KI-Trümpfe, und die staatliche Handlungsfähigkeit erhöhen.

[FAZ Wirtschaft @FAZ_Wirtschaft]

Wenn Intelligenz in unvorstellbarem Ausmaß verfügbar, kopierbar und skalierbar ist, wird sich das Arbeiten und Wirtschaft dramatisch ändern. Wir müssen besser vorbereitet sein, schreiben @MonikaSchnitzer und @privitera_ in einem Beitrag für @faznet: https://t.co/TpqeWqEfgK

[Daniel Privitera @privitera_]

@ShayneRedford @Yoshua_Bengio @DIGoldfarb Thanks for your fantastic work @ShayneRedford !!

[Daniel Privitera @privitera_]

@gcolbourn tbd!

[Daniel Privitera @privitera_]

The 1st International AI Safety Report is out today! Being the Lead Writer and collaborating with 100 leading AI experts (including Nobel laureates, Turing Award winners, etc) has been an honor.

I look forward to it being read and discussed by policymakers and governments around the world. If you have thoughts on how to improve the report next time, please share them with us!

[Yoshua Bengio @Yoshua_Bengio]

Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU.

It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵

Link to full Report: https://t.co/k9ggxL7i66

1/16

[Stanford HAI @StanfordHAI]

The UK gov't just released the first International AI Safety Report, offering key insights on general-purpose AI to guide informed policymaking. HAI leading scholars @YejinChoinka, @percyliang, @MarietjeSchaake, and @RishiBommasani contributed to the report, which will shape discussions at the upcoming AI Action Summit in France. https://t.co/4IuReMR6sw

[Arvind Narayanan @random_walker]

I appreciate that the authors of the report solicited and incorporated feedback from experts with a diverse range of views on AI Safety, including me. The report is stronger for it.

[Daniel Privitera @privitera_]

@DIGoldfarb Thanks so much for your invaluable work on this Danielle!!

[Daniel Privitera @privitera_]

@ChinasaTOkolo Thank you for your fantastic work Chinasa! It’s been a pleasure working with you.

[François Chollet @fchollet]

If you're curious about current expert assessments of AI risk, this report is an excellent reference. It offers thorough content that is not overly editorialized.

[Daniel Privitera @privitera_]

@TobinSouth Thanks for your fantastic work Tobin!

[Sayash Kapoor @sayashk]

More than 60 countries held elections this year. Many researchers and journalists claimed AI misinformation would destabilize democracies. What impact did AI really have?

We analyzed every instance of political AI use this year collected by WIRED. New essay w/@random_walker: 🧵

[Markus Anderljung @Manderljung]

The first draft of the EU's General-Purpose AI Code of Practice was just published.

I was one of the vice-chairs involved in drafting it.

As we've got many more drafts to go until May 2025, I'd be keen on folks' input on how it can be improved.

[Yoshua Bengio @Yoshua_Bengio]

Today, the AI Office of the European Commission published the first draft of the Code of Practice for GPAI. In @Euractiv, my fellow Co-Chair @nuriaoliver and I share principles which guide the drafting of the Code and why we believe this consultative process is very important.

[Alexander Zacherl @dbltnk]

🧵We just published the first draft of the EU AI Act's Code of Practice. https://t.co/hWIfTrSmYK