Dean W. Ball
@deanwball
Followers
12K
Following
15K
Media
428
Statuses
7K
Senior Policy Advisor for AI and Emerging Technology, White House Office of Science and Technology Policy | Strategic Advisor for AI, @NSF (opinions my own)
Joined April 2009
Can you imagine what would have happened if someone had discovered “do not criticize Sam Altman or Joe Biden” in an OpenAI system prompt?.
results replicated 😬. any comment, @grok? . if this has indeed been made part of the sys instructions, we’re gonna need to know why. manipulating search results seems like a funny way to combat misinformation, no?. community could use some clarity on this
43
233
3K
my jaw is on the floor . kudos, gemini team, you guys cooked.
49
116
3K
Under a strict reading of the AI Act, ChatGPT advanced voice is *illegal* in EU workplaces and schools because the system can recognize a user’s emotions. That’s prohibited by the AI Act.
Advanced Voice is not yet available in the EU, the UK, Switzerland, Iceland, Norway, and Liechtenstein.
143
353
3K
holy HELL Gemini deep research is unbelievable . just pulled info from 100ish websites and compiled a report on natural gas generation in minutes . maybe my favorite ai product launch of the last… at least three business days.
21
76
2K
has anyone else noticed that this guy almost always includes a picture of himself in his tweets.
As of today, thanks to the #DMA, a new app store can be downloaded on iOS & Android devices. Great news for EU users & app developers across the globe who want to do business here. Yes, gamers, Europe means more #FREEDOM & choice! 🎮🇪🇺 . #FreeFortnite .
48
40
2K
“California literally made you” says the state senator to Elon Musk. He says this with a straight face because he believes that government policy is *responsible* for everything good that happens in its jurisdiction. Yet somehow never responsible for the bad.
California literally made you with taxpayer subsidies & because it’s the best place around. Will this be a fake temper tantrum move just like Tesla’s fake “move” to Texas?.
68
83
2K
Remember when Thomas Edison asked everyone in the world for their consent to invent the lightbulb?. Or when Steve Jobs asked every American whether Apple should build the iPhone?.
OpenAI is building tech that aims to totally change the world without asking if we consent. It's undemocratic. And Sam Altman just proved that bespoke corporate structures & voluntary commitments won't cut it — we need LAWS that give independent oversight.
77
78
2K
I am happy to announce that I have joined the White House Office of Science and Technology Policy as a Senior Policy Advisor on AI and Emerging Technology. It is a thrill and honor to serve my country in this role and work alongside the tremendous team @mkratsios47 has built.
169
65
1K
Llama 3 405b is a "systemic risk" to society, according to the European Union and their AI Act.
56
98
840
crazy that this is on the front page of Bloomberg right now
79
37
839
The amount of factually incorrect information and hyperventilating takes on deepseek on this website is truly astounding. I assumed that an object-level analysis was unnecessary but apparently I was wrong. Here you go:. 1. DeepSeek is an extremely talented team and has been.
29
132
812
it's surprising how many people on the left make no attempt to understand anything the trump administration does. very often, they just ascribe it all to republicans being dumb or chaotic. anecdotally, when I ask these folks if they've engaged with any of the key written work.
213
64
774
The Chinese AGI lab DeepSeek is reporting an insanely low training cost of only $5.5 million for their new v3 model, which seems to match Claude 3.5 sonnet performance. DeepSeek also has a credible “o1-like” model. Again I will emphasize: if your policy solutions are
22
79
660
The thing I really appreciate about VP Vance's discussion of manufacturing here is that he makes a *hayekian* case for reindustrialization. He says that by losing manufacturing, we lose the benefits of what he calls "network effects" and specialized urban knowledge clusters.
the full @JDVance keynote from the a16z american dynamism summit today. unbelievable support and commitment to make manufacturing great again. a thread.
31
97
617
I sometimes wonder how much AI skepticism is driven by the fact that “AGI soon” would just be an enormous inconvenience for many, and that they’d therefore rather not think about it.
66
54
603
One of the few AI papers I reliably hear referenced by DC policy types is from 2023. The authors tried to train a new model with 90%-100% of training data generated by a 125 *million* parameter model. Unsurprisingly, they found that you cannot successfully train a model entirely.
Some ideas get extremely overrated by the sheer virtue of having a cool name (e.g., 'model collapse').
27
48
587
the abundance movement is cool but it’s also like, “what if we said a bunch of stuff that was conventional wisdom within classical liberal think tanks in 2012, but in left-coded language?”. Classical liberalism remains an underrated source of “new” ideas.
38
44
590
o1 and o1-pro are a cut above other models at economic history writing. they can combine knowledge of historical facts with economic analysis in a *much* more creative and cogent way than earlier models. it’s not just math and code, folks.
17
37
562
My basic reaction to AI today is, “jeez, o1 performs in the top 1% of humans at math, yet fails routinely at basic logic tasks. I guess intelligence is a high-dimensional space, and that probably means, like most high-dimensional things, it behaves counterintuitively.”.
A key crux in the AI safety debate is Empiricism vs Rationalism in the philosophical sense. Rationalists see nature as lawful, internally consistent and self-similar, which it must be for the universe to be intelligible to human minds in the first place. Empiricists believe.
20
27
483
People who make confident predictions about AI risk should reflect on the fact that the election proceeded with essentially *zero* deepfake or otherwise AI related problems. One year ago people were confidently predicting AI would “dominate” this election.
Another fine example of AI fear mongering being divorced from the practical risk.
45
126
451
ok, let me spell this out. what I meant by 'you should read things written by the people in the trump administration,' in the context of today's tariff announcement, is:. america is at the bottom of a deep, deep hole with regard to its industrial base. economists often retort.
52
38
424
“We, the California beach NIMBY commission, reject the US military’s request to do things on US military property because we disfavor the politics of your contractor’s CEO.”. Ah yes, federalism. Just as the founders intended it.
NEWS: The California Coastal Commission has rejected the Air Force's plan to allow @SpaceX to launch up to 50 rockets annually from Vandenberg, citing Elon Musk's political posts on X. Both the Air Force and Space Force supported the plan. Insane.
6
21
403
Texas for some ungodly reason is considering a law that would make plumbers, electricians, lawyers, and countless other small businesses file “algorithmic impact assessments” for a huge range of AI uses. As I’ve written before, this is among the worst ways to do AI policy.
Breaking AI news: A troubling new AI regulatory bill has been floated in Texas that borrows from the same heavy-handed, EU-like policy model that we saw implemented in Colorado, and which almost passed in Connecticut. Rightly called a “sweeping” measure in the attached article,.
20
38
382
this is such a funny recommended prompt for an ai model that costs $2400/year. I hope openai keeps these boilerplate sample prompts all the way to asi
7
17
371
I can confirm that Deep Research is capable of automating tasks that would have taken me at least a day, if not longer, of dedicated research. This might very well be the most productivity-enhancing technology product for me since GPT 3.5, and it could be bigger.
I am going to be more measured on OAI Deep Research because my off-the-cuff tweets about gemini deep research went viral and ultimately didn’t reflect my full thoughts on that product. But let me just say: this is VERY good.
16
25
351
I once again implore you to exercise humility when rendering judgments about what is and is not a good use of other people’s time and capital.
4
24
311
Like almost all other AI-generated political content, this clip:. 1. Does not demonstrate a need for “AI regulation.”.2. Would almost certainly be unconstitutional for government to regulate. 3. Has not meaningfully affected the information environment.
Presidential elections in the age of AI-powered deepfakes. The time for AI regulation is now
5
34
280
@YaelOss oh it destroys it. It takes WAY longer but it just looked at like 600 websites and compiled a very high quality report.
4
0
268
imagine how dumb everyone would feel if sb 1047 had passed, and o3 was not a covered model, but gpt 4.5 was.
15
13
267
Remember the news a few months ago that Amazon was going to hook up a data center to a nuclear power plant in Pennsylvania? . The federal energy regulatory commission has rejected it.
25
28
261
The only justification for firing probationary employees is if you think firing government employees is an intrinsic good, regardless of their talent or competence. Indeed, firing probationaries is likely to target younger, more tech and AI-savvy workers.
Firing the newest ("probationary") government employees is a great way to cripple new fields (such as AI).
10
36
263
I went from never using chatgpt scheduled tasks to having ~20 with o3. News roundups on arbitrarily niche topics delivered at custom intervals. This alone is worth the money.
21
5
263
@aidan_mclau ok so this one is giving away a little bit about a forthcoming piece of mine, but also niche and purely non-technical--in fact, it's nearly a pure humanities question. the prompt was: . "did beethoven write solo piano music that would have been technologically impossible for
18
13
251
Some reflections on Paris and our current moment . 1. The odds that we are living through a takeoff to transformative ai seem high—much higher than they did to me one year ago—and most evidence I see reinforces, rather than complicates, this basic intuition. 2. Transformative ai.
15
34
252
My guess: It was obvious by now that models could do stuff like this to everyone who is paying attention and capable of changing their mind, so not big news. The only people left are self-consciously incapable of updating (eg Gary Marcus), so gain nothing by commenting on this.
Anthropic seems to have found evidence of their models thinking via quite nuanced planning as opposed to pure next token prediction, but the reaction to the paper was muted. What am I missing?.
9
9
244
Humor is dangerous, claims “online misinformation expert,” because it’s “how fringe ideas get normalized.”. Do these people get tenure?
21
42
230
There are people, ostensible professionals in the analysis of public affairs, who believe Donald Trump took on this level of risk for… nothing? Just for fun?. Even if you think he is wrong, and will fail, I contend that your model of reality is woefully incomplete.
97
15
228
It is wrong to call manus a “deepseek moment.” Deepseek was about replication of capabilities already publicly achieved by American firms. Manus is actually advancing the frontier. The most sophisticated computer using ai now comes from a Chinese startup, full stop.
9
30
223
A lot of people are dunking on the headline of Tim Wu’s “The First Amendment is Out of Control” piece in the Times but the details are even crazier. He suggests “lies” shouldn’t be 1A protected. Who decides what is and isn’t a lie? I wonder who…
29
34
211
This doesn’t surprise me in the slightest. o1 is a very good legal reasoner and I cannot wait for it to get document upload. People continue to sleep on o1 for reasons that escape my understanding.
OpenAI CFO Sarah Friar says lawyers are reporting that the new o1 reasoning model can do the work of a $2000/hour paralegal
22
13
211
For the record: I expect AI to add something like 1.5-2.5% GDP growth per year, on average, for a period of about 20 years that will begin in the late 2020s. That is *wildly* optimistic and bullish. But I do not believe 10% growth scenarios will come about.
We developed GATE: a model that shows how AI scaling and automation will impact growth. It predicts trillion‐dollar infrastructure investments, 30% annual growth, and full automation in decades. Tweak the parameters—these transformative outcomes are surprisingly hard to avoid.
21
0
208
The fact that the DOGE winds down on July 4, 2026 (just a little over 18 months from Trump’s inauguration) strikes me as a mistake. The bureaucracy will know it’s a “run out the clock” scenario, and they have many delay tactics (it’s their specialty).
29
16
207
On an entirely unrelated note, did you know that this guy's company has advocated for making it a felony to open source a gpt-4 class model?. And requiring a government license to train a gpt-3.5-class model?. (in a report commissioned by the US government, by the way)
Within the next 18 months, public opinion will start to turn against open-source AI. This will happen because of one or more highly visible incidents of misuse of an open-source model, probably associated with significant damage or loss of life. 80% confident.
9
23
202
glad they got to interview Gary Marcus for that crucial 2% of ai experts who have never used a chatbot.
New numbers from Pew this morning, they reveal a large gap in perception between the general public and people whose work and research relates to AI. Usage: 66% of the general US public have still never used AI. You probably have a good idea of who the 2 percent of experts are.
9
3
205
reminder: you now must do phone calls on services like signal whatsapp or facetime audio if you care remotely about infosec. this is due to chinese hackers breaching a US-government-mandated backdoor into our telecom system. never let government mandate another vulnerability
6
46
193
“Model collapse due to synthetic data” has become a real brainworm for some in the policy world. I’ve got my own theories as to why, but I’ll leave it to you to formulate your own.
12
10
184
There is a certain complacency associated with the belief that “we live in a democracy.”. Yes, you vote. But you also live underneath a vast bureaucratic apparatus that has a logic and a momentum of its own. This piece is about what that machine is doing to AI.
This is how you assert control over the most promising emerging technology in a generation. You do it before it’s popular, before people will notice too much. You do it quietly, behind closed doors in working groups and workshops and steering committees. You do it with the active
7
21
168
It’s possible they’ll be granted some kind of exception, though, because EU tech law is mostly made up on the fly—as they’d say it, it’s about the “spirit” rather than the “letter” of the law.
14
2
169
Every time the EU releases a new ai policy document I think, “wow, lots of this is quite under-specified and will likely need to be followed up with another policy document.”. And I wonder whether that is in fact the point of this whole endeavor: more work for framework-drafters.
🚨NEW: The European Commission has just published the first draft of the Code of Practice for general-purpose #AI model providers.
14
18
177
I see *nothing* about current language models that justifies something like SB 1047. If that changes with future models, my desired policy regime will change with them. Most 1047 supporters I know agree with me on the first point, but argue that we will be “too late” if we wait.
The irony is that commitment to empiricism used to be a core trait of rationalists. Then AI doom came along and hacked utilitarianism and now all core traits have been jettisoned because in the face of oblivion principles no longer matter. An epistemic ouroboros.
17
14
172
In @TIME this morning, @DKokotajlo67142 and I make the case for frontier AI lab transparency—either as a voluntary commitment or as a law. While we disagree quite a bit about the trajectory of AI, we concur here. Link to the piece, with much more detail, in the reply.
13
24
175
Responses to the Office of Science and Technology Policy’s AI Action Plan RFI are out! . Thanks to all 10,000+ companies, research organizations, and individual members of the public for submitting their ideas on the future of American AI. 🫡
13
26
173
A based version of the US AI Safety Institute would run all sorts of evals on Chinese models to show the various ways in which they are being steered by CCP censorship.
DeepSeek R1, self-hosted, on Taiwan's status as an independent country. Woooooow. This seems hard-coded in the model.
12
17
168
How should the federal government prioritize its AI R&D investments over the next 3-5 years?. If you have thoughts, we want to hear them! New 30-day comment period just opened for feedback on the National AI R&D Strategic Plan.
12
46
169
o3 gets between 75%-87% on ARC depending on the amount of compute used, a record, exceeding human performance at the high end. 25% on FrontierMath, compared to previous state of the art of 2%. if you are a deep learning bear, now would be a great time to reconsider.
4
7
170
Gonna be a fun* second half of the decade. *if by fun you mean “a renaissance and an information revolution and an Industrial Revolution and a global political realignment converging, with tremendous and unknowable consequences that will shake the foundations of modern society”.
11
18
168
Finally got a chance to read this piece, which argues that o1-style reasoners will not generalize beyond domains with easy verification. It may well be true, but I have some causes for doubting Aidan’s thesis. One I want to highlight in particular:.
i wrote a new essay called. The Problem with Reasoners. where i discuss why i doubt o1-like models will scale beyond narrow domains like math and coding (link below)
5
6
165
I like to think about a civilization of AIs building human brains and trying to decide whether that’s real intelligence. Surely in that world there’s a Gary Marcus AI going like, “look at the optical illusions you can trick them with, and their attention windows are so short!”.
10
14
159
Last August @ajeya_cotra asked me what capabilities in future models would change my opinions on prudent policy measures. My response is below. I believe we have now seen what I described, and so my opinions about prudent policy measures have, in fact, changed.
@ajeya_cotra I feel reasonably confident that the labs have figured out grounded mathematical reasoning—so a big question for me is going to be, “does that reasoning translate generally beyond math?”. If it seems like it does—and it need not be perfect—I’ll start to be more convinced.
3
13
155
I do not expect DeepSeek to continue open sourcing their frontier models for all that much longer. I give it 12 months, max.
"DeepSeek’s leaders have been worried about the possibility of information leaking"."told employees not to discuss their work with outsiders". Do DeepSeek leaders and the Chinese government know that DeepSeek has been open sourcing their ideas?
12
16
153
I'm excited to announce that @binarybits and I are launching a podcast! We call it AI Summer, and it will feature interviews with researchers, analysts, and other experts from across the AI world. First up is the ever-excellent @JonAskonas on AI policy in the Trump admin.
6
27
151
This was an unbelievably good event. Truly one of the best I’ve ever attended, on any subject. An exceptionally high-quality and diverse group of people, in the excellent Lighthaven venue. Already excited for next year. Congratulations to @rootsofprogress!.
Announcing Progress Conference 2024: Toward Abundant Futures. Hosted by @rootsofprogress together with @foresightinst @HumanProgress @TheIHS @IFP @WorksInProgMag . Keynotes from @patrickc @tylercowen @jasoncrawford @sapinker. Berkeley, Oct 18–19
5
15
150
My vague impression is that oai deep research is the ai product that really got a meaningful chunk of the policymaking community in dc to start feeling the agi.
6
2
150
@GreatDecoupling I don’t care about what is “symmetrical.” Censorship is censorship. There is no excusing it.
3
0
146
I expect that a frontier model will produce policy briefs more persuasive and well-considered than even a "good" think tank senior fellow or academic within a year--and it could happen as soon as o3-pro in a couple months. o1-pro already can meet this threshold sometimes.
It can be hard to “feel the AGI” until you see an AI surpass top humans in a domain you care deeply about. Competitive coders will feel it within a couple years. Paul is early but I think writers will feel it too. Everyone will have their Lee Sedol moment at a different time.
12
11
148
The AI community is laser-focused on SB 1047, but there's another AI bill in California that is equally, if not more, aggressive than 1047: AB 3211. Like 1047, AB 3211 has passed one chamber of the legislature and is authored by a powerful legislator. Let's take a look. 🧵
7
54
138
More people in the open source AI world should be talking about the ENFORCE Act, which would allow the Commerce Department to place export controls on frontier models. Future open weight frontier models would probably be illegal if Commerce exercises this authority. 🧵
14
49
140
Newsom’s veto message of SB 1047 wisely calls out the problems with “frontier model thresholds,” made newly problematic with OpenAI’s o1-mini. “False sense of security” is right. As I’ve written, the solution—at least step one—is to target illicit conduct, not models.
7
12
139
I read a great anecdote about government consultants recently: a federal agency was trying to figure out how many external consultants it employed, couldn’t figure out how, and hired consultants to count the consultants. Government dysfunction is often worse than people think.
2
21
134
Excited to share a forthcoming paper--an in-depth exploration of the private governance proposal I outlined on Hyperdimensional a few weeks ago. The paper will also go through my thoughts on several other major AI governance frameworks. On ArXiv soon!
16
12
139
By the way, @Scott_Wiener has an AI-related bill (SB 53) this session that seems eminently reasonable to me. It:. 1. Creates a committee to study doing CalCompute, a public compute cluster .2. Establishes whistleblower protections for frontier lab employees . The whistleblower
4
15
136
good rebuttal by Ethan below to this really bad oped in the WSJ today. not sure what "inside view" the piece was supposed to represent, but I would say it represents the views of approximately zero AI researchers or executives I've spoken to.
I wish people would stop repeating these as if they are facts that AI is plateauing. AI might hit a roadblock, we don’t know, but every one of these issues has multiple studies stating the opposite: synthetic data works, scaling is fine, etc. We need more nuance on the AI future
7
13
133
It seems possible that America’s compute export controls drove DeepSeek to pursue these radical training efficiencies. Their sophistication may exceed US labs in at least some important ways, though here I am only speculating. Export controls have been known to backfire in.
DeepSeek (Chinese AI co) making it look easy today with an open weights release of a frontier-grade LLM trained on a joke of a budget (2048 GPUs for 2 months, $6M). For reference, this level of capability is supposed to require clusters of closer to 16K GPUs, the ones being.
4
6
132
DeepSeek r1 takeaways for policy: .1. Chinese labs will likely continue to be fast followers in terms of reaching similar benchmark performance to US models. 2. The impressive performance of DeepSeek's distilled models (smaller versions of r1) means that very capable reasoners.
3
19
133
What they don’t tell you about working at the White House is that the job comes with a free crash course in intermittent fasting.
5
0
135
There you have it. First credible Chinese replication of the OpenAI o1 paradigm, approximately 9 weeks after o1 is released. And it’s apparently going to be open source.
🚀 DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power!. 🔍 o1-preview-level performance on AIME & MATH benchmarks. 💡 Transparent thought process in real-time. 🛠️ Open-source models & API coming soon!. 🌐 Try it now at #DeepSeek
10
24
128
I think the AI policy trope I most want to disappear in 2025 is:. “AI is not just language models, it’s been around for a long time, is already widely used, and has been presenting unique policy challenges for years.”. I can’t tell you how many times I’ve heard people say things.
7
13
127
I concur with @hamandcheese that the correct way to understand DOGE is not as a cost-cutting or staff-firing initiative, but instead as an effort to prepare the federal government for AGI. Trump describing it as a potential "Manhattan Project" is more interesting in this light.
This OPM memo is going to be the most impactful news of the day, but I'm not sure it'll get much reporting.
16
24
125
I cannot stress enough how in the last six months many future “sci-fi” AI capabilities, good and bad, have gone from being speculative or very uncertain on timing to being, basically, a near-term certainty (24 months). When I saw o1-preview, it all became clear.
the vibe shift from “sci-fi future is some distant maybe” to “sci-fi future is within reach” has been abrupt & total—the last *six months* have been like pulling back a curtain & realizing the future was just sitting there, waiting for us to move out of the way. now it’s just a.
1
19
123
The simple fact is that SB 1047 is *speculative* regulation, regulation of a thing that has never happened, of a thing that we do not know will ever happen. A culture with such low risk tolerance will struggle to do interesting things. And boy, will it hate the coming century.
I see *nothing* about current language models that justifies something like SB 1047. If that changes with future models, my desired policy regime will change with them. Most 1047 supporters I know agree with me on the first point, but argue that we will be “too late” if we wait.
10
14
120
Meta's Llama 3.1 gives us, for the first time in a long time, a frontier, open-weight LLM. It also poses a fundamental question: will policymakers allow the open-source AI ecosystem to flourish?. In my latest essay, I explore what Llama 3.1 means for AI policy going forward. 🧵
4
22
116
This investigation is rooted in the idea that any sufficiently successful corporation is inherently suspicious and worthy of government harassment. This sends an awful sign to entrepreneurs, and is easily the worst tech antitrust investigation I've seen.
BREAKING: Nvidia has been subpoenaed by the US Justice Department in an escalation of the agency's antitrust investigation
6
13
119
Not enough people are talking about this ruling on the motion to dismiss in the Character AI case. It’s some of the first reasoning we’ve seen from a judge on the tort liability exposure faced by AI developers. If you care about AI but have no patience for legal briefs: talk to
11
10
119
Nevada has a fairly wild AI bill that snuck up on me: SB 199. It might actually be the worst AI law I’ve ever seen. It creates a centralized regulator. It requires *any* company that develops *any* AI model to obtain a license from the state. And it is filled with goodies for
14
28
115
If you care about ai catastrophic risk and are sad about the election outcome I think you have a very lossy model of the politics of ai. Should I write an essay elaborating on this?.
28
2
115
After reading the relevant portions of this 700+ page report I’m quite disappointed. I have a lightly, rather than strongly, held conviction against an AGI Manhattan Project. The trouble with this report is the total lack of effort to *justify* such a radical step.
4
7
114
Not only this, but European elite attitude toward the US is part of what drives their tech regulation. In the run up to the AI Act, there was a lot of talk along the lines of “we don’t want American values polluting our society.”. European elites do not like us.
I wonder how much Europe's inferiority complex toward the U.S. is holding it back from solving these problems. Post a chart like this, and you get nothing but the wildest cope -- fantasies about how Americans don't have health care or vacations and their cities are hellholes.
7
8
110
here is a good candidate for my “most absurdly under-viewed tweet relative to importance of information” award.
Two days after this tweet, CBP excluded almost everything relevant for AI development from reciprocal tariffs. Guidance went live yesterday.
8
15
113
I don’t tend to get angry about the things I write about for a living. But the arbitrary and parasitic way in which the European Union extracts money from US tech firms, while our own government applauds them, is one of the few things that does, indeed, make me mad.
How the EU Weaponizes Regulation to Extract Billions from American Tech. With its new Digital Services Act (DSA) and Digital Markets Act (DMA), European Union regulation threatens to significantly cut profit margins necessary for R&D, capital expenditures and other strategic
12
15
108
I do not expect widespread, mass layoffs due to AI in the near term. But I worry that as knowledge workers exit firms, they won’t be replaced. This may hit young people seeking to enter knowledge work fields especially hard. Their job prospects by the late 2020s could be dim.
16
12
111
I do not agree with Miles on many policy prescriptions, but I do agree with him that the policies we enact in 2025 will be hugely important in setting the tone for the development of exceptionally capable ai. This is why I am concerned about profoundly unserious proposed and.
In my new blog post, I share a starter pack for feeling the AGI: resources and arguments indicating that AI is virtually certain to exceed human performance in most areas in the next few years, and that the time to act is now.
5
13
110
hello new followers. I don't write about trade; I write about AI. however, I think AI is going to automate a lot of knowledge work in the relative near term, and without an industrial base to rival china's, America could be in dire trouble. Few people care sufficiently.
6
13
107
imagine how much better classical music would have been if the habsburgs had required mozart and beethoven to write symphonic impact assessments.
3
14
108
sb 1047 subtweet (I agree!). More people should be worried about what happens if Brussels and Sacramento decide US tech policy. The vast majority of Americans would have no say in our own tech policy. This is a sovereignty, and political, crisis—just waiting to happen.
we are happy to have reached an agreement with the US AI Safety Institute for pre-release testing of our future models. for many reasons, we think it's important that this happens at the national level. US needs to continue to lead!.
6
9
101