A Step Toward Nationalizing AI? White Memo Full Analysis and Context
Added 2024-10-27 16:29:36 +0000 UTCA White House Memo released in the last 72 hours is hard to ignore. It is the clearest statement yet from the US Govt that it thinks ‘general-purpose AI’ is coming soon, and that it could upend international order. They want to get much more involved, for better or worse. Here are the 28 most impactful excerpts from the Memo, together with tons of context from new and old material from OpenAI, Anthropic, the Pentagon, the NYT and more.
Link for Off-line Watching/Download: https://drive.google.com/file/d/1bWKFgUYBtMvt1c53npnZfvkqo-E0voOp/view?usp=sharing
White House Memo: https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/?s=09
Altman Essay (paywall): https://www.washingtonpost.com/opinions/2024/07/25/sam-altman-ai-democracy-authoritarianism-future/?=undefined
Amodei Essay: https://darioamodei.com/machines-of-loving-grace
Aschenbrenner Situational Awareness: https://situational-awareness.ai/wp-content/uploads/2024/06/situationalawareness.pdf
Pentagon Deepfakes: https://theintercept.com/2024/10/17/pentagon-ai-deepfake-internet-users/
Altman Redistribution: https://www.nytimes.com/2023/03/31/technology/sam-altman-open-ai-chatgpt.html
AlphaChip: https://deepmind.google/discover/blog/how-alphachip-transformed-computer-chip-design/
Stargate Supercomputer: https://www.youtube.com/watch?v=KXG2f-So9oo
EpochAI Report: https://epochai.org/blog/can-ai-scaling-continue-through-2030
OpenAI US General: https://openai.com/index/openai-appoints-retired-us-army-general/
Comments
Maybe the US (led by Elon) will try to nationalize AGI before it is realized, and many researchers will defect to the UK/EU, and we will democratise the use of it? Pretty unrealistic, but worth daydreaming about
Ivo Denisovich
2024-11-16 11:22:37 +0000 UTCYou are absolutely right to point out that this memo is wrong about radar. The British beat us to it. But you know it's probably for the best they put it in there because now you can point to one obvious flaw of the memo that illustrates how much it reeks of arrogance and self interest. Your not being petty at all. You are absolutely right about this being a new cold war. Great job Philip!
Joshua Davis
2024-11-13 01:52:03 +0000 UTCAs far as I know TSMC is the only company in the world capable of producing the latest cutting edge chips. But: I'm just repeating what I have heard in documentaries, which is what I call youtube videos nowadays ;)
Jörg Weiß
2024-11-11 23:02:54 +0000 UTChttps://www.newsweek.com/donald-trump-clear-choice-ai-opinion-1966073
Doug
2024-11-06 15:22:40 +0000 UTCI am not commenting on my opinion of Donald Trump's AI policies (need to spend more time reviewing), but it is important to Grok (understand) what his AI policies may be. Donald Trump Is the Clear Choice on AI | Opinion - Newsweekhttps://www.newsweek.com/donald-trump-clear-choice-ai-opinion-1966073 The Trump campaign is focusing on ways to fix our country, including the use of artificial intelligence (AI) solutions for a number of reasons. AI is being used to improve education, health care, scientific research, data, automation, drug discovery, sustainability, manufacturing, retail business, banking, customer service, and prison recidivism, a topic that President Donald Trump has spoken out about the need to reduce. In 2018, his administration announced support for legislative action to reduce recidivism and called on Congress to act. The president also successfully signed the First Step Act, a bipartisan criminal justice bill in December that year to improve criminal justice outcomes, as well as to reduce the size of the federal prison population while also creating mechanisms to maintain public safety.As president, Trump displayed his commitment to strengthening American leadership in AI, recognizing its importance for the economy and national security. Trump issued the first ever national AI strategy, committed to doubling AI research investment, set up the first-ever national AI research institutes, launched the first AI regulatory guidance, created new international AI alliances, and set up guidance for federal use of AI.
Doug
2024-11-06 15:22:13 +0000 UTCWhat did you say at 2:22 , epoch ai? Trying to find who that is lol
CB Trading
2024-11-03 09:01:48 +0000 UTCI am not really in the habit of reading memos from the white house. who actually authored this? I guess who it is offered "to " follows the title to the memo right? What ramifications does it actually have on policy in a real way? Sorry if these are dumb questions LOL
Daniel A Barbatti
2024-11-01 17:35:28 +0000 UTCI was going bring up basically the same concern. China has been "looking" for a reason to take Taiwan for some time. Here it is. I believe I heard Nvidia had already built a US based production facility (maybe Arizona?). But even if Nvidia does create alternative facilities wouldn't it be the case that if China comes in and takes over Taiwan and the Nvidia facility that they immediately would possess the infra and tech to start making their own. (unless there is some kind of self-destruct on invasion set up at the Taiwan facility.. or maybe I have watched too many spy movies)
Daniel A Barbatti
2024-11-01 16:58:25 +0000 UTCIf only UK had stayed in EU, there might have been… Sorry, that’s a different regret! 🥲
Pavol Vaskovic
2024-10-29 17:47:48 +0000 UTCThe missing piece of this strategy is how the global economy would function. If all of AI is concentrated in the US, how would the rest of the world (including the EU, UK, Japan, etc.) pay for these services? Would we still be working while US citizens live on UBI?
Arek Stryjski
2024-10-29 08:53:42 +0000 UTCA CERN for AI is a must, otherwise I fear we’ll end up in global conflict. Concentrated resources always leads to conflict.
Sean Betts
2024-10-29 08:26:53 +0000 UTCWe should get you on Lex's podcast to amplify your reach. Your analysis is a stabilizer in the unstable discourse/hype/emotional loop, and we need more of it injected.
Enrico Ros
2024-10-29 00:04:46 +0000 UTCFor a long time I've thought to myself that the amount of money circling around AI top players is just so insane that there must be governmental interests too. Thanks for this, I really liked this video
Ville Johannes Pajala
2024-10-28 19:04:52 +0000 UTCAs the US diversifies chip production, the threat may abate eventually, and Nvidia said it could be within 10 years (so as of writing 9 years) that US is self-sufficient for chips. So if it happens for that reason, would be sooner rather than later.
Philip
2024-10-28 15:57:01 +0000 UTCGood question, keeping it vague on purpose I suspect
Philip
2024-10-28 15:55:52 +0000 UTCYeah let's see, CHIPs act going could be interesting
Philip
2024-10-28 15:55:36 +0000 UTChaha
Philip
2024-10-28 15:54:56 +0000 UTCIf only DeepMind had stayed UK-owned, their might have been some counterweight that encouraged more of an 'alliance-based' approach. Hassabis has hinted he regrets it many times, never quite made it into a video yet
Philip
2024-10-28 15:54:48 +0000 UTCThanks Judy, I too am a proud Brit (still!). Never be shy about sharing those thoughts!
Philip
2024-10-28 15:53:30 +0000 UTCThis is crazy timing! NIST is wrapping up the first pilot exercise of https://ai-challenges.nist.gov/aria , which is aimed at "building up the tools, measurement methods, and metrics necessary for AI risk and impact assessments". I'm hoping simple bench and public participation will become key sources that inform the outcomes! I've been thinking a lot about were public input could influence policy (and hopefully get a seat at the table with all the CEOs), and I think better risk assessment methodologies and professionalized volunteer or freelance evaluators could be an avenue that doesn't get over run by other incentives.
Blake Chambers
2024-10-28 14:36:49 +0000 UTCPhillip! Thank you for this incredibly insightful, if not alarming update, which is a coming reality that I have thought about in my limited capacity. This race for power will simply mean AI will only be open to all whilst all the information is harvested and trialled. Proof of concept will likely close it down to the majority, being given access to limited capabilities not likely to pose a threat to National security. Not petty at all to correct the historical inaccuracies. Proud Brit here... still! I am in awe of your capacity to assimilate and deliver this information, apparrently so effortlessly. Thank you, again!
Judy Hitchcock
2024-10-28 08:43:05 +0000 UTCOstensibly this is all about the US protecting itself from other countries, but quite a lot of it also reads like the US government wanting to protect itself from powerful AI companies (many of which are of course American). I also wonder how much this strategy will change if there is a change of government in the US.
James Maclaurin
2024-10-28 07:41:16 +0000 UTCIf the U.S. (alone or in a coalition) seems to be about to gain a permanent advantage, I fear attacking Taiwan will be seen as inevitable by China. Taiwan is the proverbial Achilles' heel of the US AI program. Hit it before all those data centers get built, and, well, they won't be built. Attacking Taiwan represents flipping the table. It ends the current world order and no one can possibly calculate what the outcome would be. The hope has always been that this chaos and its risks will prevent China from making that move. But now the game is changing. It's no longer about the wish to reunite Taiwan with China. It's suddenly about "We need to stop the USA from dominating us for the rest of history." This is the kind of logic that starts nuclear wars. The risk of extinction might appear more attractive than the certainty of never-ending servitude. Heck, it wouldn't even need to be China, Russia could nuke TSMC just fine. We desperately need international cooperation to steer this, but instead we are just sliding deeper and deeper into nationalism and the logic of conflict. If we really get to powerful AI in the next couple of years, the timing is really bad. We are less ready for this than we were in the 90s and 00s.
Jörg Weiß
2024-10-28 00:20:29 +0000 UTCThe memo is predicated on the voluntary submission of systems for evaluation. I strongly suspect that OpenAI, at least, has decided that cooperating with the executive branch — especially the national security apparatus, which not only wants to gain an advantage in applications of AI, but also has the President's ear — is a good way to forestall premature attempts at regulation by the legislative branch. There's no overstating the significance of OpenAI's addition of Paul Nakasone, former head of the National Security Agency, to its board of directors. The chances of nationalization of any industry in the U.S. are nil to none, except when the country is at war or a vital industry is at risk of collapse. U.S. law does not permit it, and the Supreme Court is not going to allow creative interpretations that challenge private ownership. I'm not sure that a Democrat-controlled Senate, having eliminated the filibuster rule, would vote to permanently seize any private asset. Note that the radical left, for all the noise it makes, includes only 10 percent or so of Americans.
Tom English
2024-10-27 21:46:58 +0000 UTCThings are starting to get scary.
Phillip Yao-Lakaschus
2024-10-27 21:31:58 +0000 UTCSo how exactly does the US government define AI? Are we still talking about fancy autocomplete?
Alexander Berezin
2024-10-27 18:43:30 +0000 UTC"Situational Awareness" starting to look slightly less sensational over time. Will be interesting to see how it continues to age. Thanks for this stellar breakdown of what seems like a rather important milestone. Of course, the election result could lead to a "stay the course" approach consistent with this memo or a complete overhaul of this plan as well as the infrastructure that would support it. If life was a Greek play, this would be a good time for a comic interlude.
Steve DeMoss
2024-10-27 17:56:56 +0000 UTCA race between world superpowers to reach AGI? I can only get so hard
zero
2024-10-27 17:45:40 +0000 UTCAgencies, other than the AI safety one, officially having prior access to models seems more than a little foreboding. Overall this document seems like a bit of a Los Alamos moment for AI, physical location notwithstanding and with chips instead of uranium. Although it seems less obvious to me that the current paradigm will lead to truly dangerous or reliable AI than the fact that you could create a bomb out of Uranium-235. Nonetheless, the chips are down. And I guess it seems Europe is shit out of luck when it comes to truly massive compute. I doubt well get the prime rib coming out of Nvidia anytime soon. If the bitter lesson holds, Mistral and any other hopefuls are likely to fall further and further behind. Also, open publications and the scientific process might suffer even more than it already does. If models are vetted, the methods of their creation certainly will be. Imagine a future where helpful AGI exists, and the method of its creation is a closely guarded secret for reasons of national security. I don't know if I'm the only one to find that thought a little sad.
Sebastian
2024-10-27 17:38:54 +0000 UTCDemocracy seems to be the new favourite buzzword, ever since Freedom was colored a Conservative leaning word. I agree this is not about democracy.
John Merkowsky
2024-10-27 17:36:51 +0000 UTCVery insightful! This posturing has nothing to do with democracy and everything to do with preserving power. If nothing else it shows that AI advancement isn't seen as a nothingburger by the powers that be in government.
Machiel Reyneke
2024-10-27 17:20:29 +0000 UTCThe full weight of the US government behind AGI research will be accelerational to say the least. My first thought is how will this affect open source, such as Meta's efforts? Will the US support open source or aim to regulate them into oblivion in the name of national security? I hope for the former, but fear for the later.
John Merkowsky
2024-10-27 17:04:37 +0000 UTC
