Positive users back the proposed US-led frontier AI standards body because government rules lag innovation, while negative users reject it as regulatory capture that risks cartel-like centralization of power.
Based on 14 visible X reactions from 45 accounts; directional sample.
Ask a question below.
Published answers will appear here.
Centralizing power in any single body,even one framed as a 'standards body' for national security - is inherently risky. Demis’s proposal for a US-led Frontier AI Standards Body essentially gives one country (and the labs that operate within its regulatory umbrella) significant gatekeeping power over frontier models worldwide. While the safety concerns are real, history shows that concentrated control rarely stays benevolent for long. What we actually need is a more distributed, transparent, and internationally accountable framework, one where power isn’t concentrated in the hands of any single nation or small group of actors, but is genuinely oriented toward the long-term interests of humanity as a whole.
@AndrewCurran_ No one trusts the US government, which is objectively no longer under the control of the American people. This plan would hand over control of AI to the Epstein Class. I think Demis is too sheltered to realize this. https://x.com/TomLoweCinema/status/2077029234128892193/photo/1
@levie A standards body can help establish best practices while allowing innovation to move faster than traditional regulatory processes.
@AndrewCurran_ This is begging for regulatory capture to establish a privileged private AI cartel. No thanks.
@AndrewCurran_ We desperately need this, our govt don’t even know what they are doing
He argues a self-regulatory organization is premature without industry consensus.
Centralizing power in any single body,even one framed as a 'standards body' for national security - is inherently risky. Demis’s proposal for a US-led Frontier AI Standards Body essentially gives one country (and the labs that operate within its regulatory umbrella) significant gatekeeping power over frontier models worldwide. While the safety concerns are real, history shows that concentrated control rarely stays benevolent for long. What we actually need is a more distributed, transparent, and internationally accountable framework, one where power isn’t concentrated in the hands of any single nation or small group of actors, but is genuinely oriented toward the long-term interests of humanity as a whole.
@AndrewCurran_ No one trusts the US government, which is objectively no longer under the control of the American people. This plan would hand over control of AI to the Epstein Class. I think Demis is too sheltered to realize this. https://x.com/TomLoweCinema/status/2077029234128892193/photo/1
@levie A standards body can help establish best practices while allowing innovation to move faster than traditional regulatory processes.
@AndrewCurran_ This is begging for regulatory capture to establish a privileged private AI cartel. No thanks.
@AndrewCurran_ We desperately need this, our govt don’t even know what they are doing
@AndrewCurran_ 🤮
Thoughtful proposal for a standards body for AI. This is distinct from a regulatory agency, and would certainly allow for much faster improvement of standards and collaboration with the industry. What you definitely don’t want is for AI progress to start to move at the speed of that the government classically operates at. If that happens then we can essentially guarantee progress begins to stall, and worse, America likely just loses the AI race. This framework threads the needle mostly. The only challenge is still that even those in industry don’t agree on the same safety risks in AI. But if you can get alignment there, this looks better than most other proposals.
Demis Hassabis is calling for the US to establish a Frontier AI Standards Body to 'conduct testing in areas relevant to national security'. Frontier Labs would initially voluntarily submit Frontier Models (thresholds for these two terms to be determined) for a 30 day review prior to release. Once standards are formalized, Frontier Models would be required to pass assessment by this body to be deployed in the US market. Organizations deemed Frontier Labs by the body will be 'encouraged to adopt best practices, such as publishing model cards with technical details, maintaining strong internal cybersecurity, vetting key personnel, and providing sufficient resourcing for safety and security research, and more.'
This is not cautious optimism but the precautionary principle designed to capture the market. The notable points about FINRA are that (1) it came after 400 years of experiments in capitalism and regulatory experience unique to the problems. Something like FINRA is what you apply when there is more shared knowledge and expertise, not when no one knows or agrees on even where a technology is going. And (2) it is not clear at all that FINRA is the model to repeat given how it operates outside the accountability norms of other potential regulatory approaches. AI at this stage does not need a quasi government body with limited transparency, lack of due process, and solitary and unaccountable dispute resolution. This is a gambit that one company seeks to use to advantage itself over other companies and open source competitors in an effort to capture a process. The best way to control competition is to prevent it from ever existing. As we saw with prior review, even this company knows it should be more careful about what it asks for.
Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release. … assessment protocol … Frontier Models would be required to pass it … evaluations of capabilities in cybersecurity, biological threats & other high-risk domains. … look for attempts to bypass safety guardrails or signs of deception" Seems to me regulation works far better when it reacts to concrete realized problems, vs. trying to anticipate possible problems in imagined future contexts.
I can't get mad at Demis but this isn't really serious https://x.com/teortaxesTex/status/2077031992147796063/photo/1 https://twitter.com/demishassabis/status/2076957440109625718
https://x.com/HaydnBelfield/status/2076984995030151610/photo/1 https://twitter.com/demishassabis/status/2076957440109625718
Positive users back the proposed US-led frontier AI standards body because government rules lag innovation, while negative users reject it as regulatory capture that risks cartel-like centralization of power.
Based on 14 visible X reactions from 45 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@AndrewCurran_ 🤮
Thoughtful proposal for a standards body for AI. This is distinct from a regulatory agency, and would certainly allow for much faster improvement of standards and collaboration with the industry. What you definitely don’t want is for AI progress to start to move at the speed of that the government classically operates at. If that happens then we can essentially guarantee progress begins to stall, and worse, America likely just loses the AI race. This framework threads the needle mostly. The only challenge is still that even those in industry don’t agree on the same safety risks in AI. But if you can get alignment there, this looks better than most other proposals.
Demis Hassabis is calling for the US to establish a Frontier AI Standards Body to 'conduct testing in areas relevant to national security'. Frontier Labs would initially voluntarily submit Frontier Models (thresholds for these two terms to be determined) for a 30 day review prior to release. Once standards are formalized, Frontier Models would be required to pass assessment by this body to be deployed in the US market. Organizations deemed Frontier Labs by the body will be 'encouraged to adopt best practices, such as publishing model cards with technical details, maintaining strong internal cybersecurity, vetting key personnel, and providing sufficient resourcing for safety and security research, and more.'
This is not cautious optimism but the precautionary principle designed to capture the market. The notable points about FINRA are that (1) it came after 400 years of experiments in capitalism and regulatory experience unique to the problems. Something like FINRA is what you apply when there is more shared knowledge and expertise, not when no one knows or agrees on even where a technology is going. And (2) it is not clear at all that FINRA is the model to repeat given how it operates outside the accountability norms of other potential regulatory approaches. AI at this stage does not need a quasi government body with limited transparency, lack of due process, and solitary and unaccountable dispute resolution. This is a gambit that one company seeks to use to advantage itself over other companies and open source competitors in an effort to capture a process. The best way to control competition is to prevent it from ever existing. As we saw with prior review, even this company knows it should be more careful about what it asks for.
Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release. … assessment protocol … Frontier Models would be required to pass it … evaluations of capabilities in cybersecurity, biological threats & other high-risk domains. … look for attempts to bypass safety guardrails or signs of deception" Seems to me regulation works far better when it reacts to concrete realized problems, vs. trying to anticipate possible problems in imagined future contexts.
I can't get mad at Demis but this isn't really serious https://x.com/teortaxesTex/status/2077031992147796063/photo/1 https://twitter.com/demishassabis/status/2076957440109625718