Ramez Naam

66.7K posts

Ramez Naam banner
Ramez Naam

Ramez Naam

@ramez

Climate and clean energy investor. Author of 5 books. Energy & Environment co-chair @SingularityU. Trying to build a better world.

Seattle Katılım Mayıs 2007
9.5K Takip Edilen58.4K Takipçiler
Sabitlenmiş Tweet
Ramez Naam
Ramez Naam@ramez·
Despite this election, I remain an optimist about America and the world. Humanity will continue to produce new ideas and new innovations to improve our lives. Good people will continue to come together to improve the world. And the political tide will turn. We'll make it so.
English
38
19
290
134.1K
Ramez Naam
Ramez Naam@ramez·
@adeoressi I had a couple dozen fraudulent $20 Anthropic subscriptions billed to my Amex.
English
0
0
4
673
Adeo Ressi
Adeo Ressi@adeoressi·
According to AMEX, many members are disputing Anthropic charges... Hmmm
English
1
0
9
1.6K
Ramez Naam
Ramez Naam@ramez·
That's not what the document says. You're referring to Appendix A. But the opening sentence there is: "Our main recommendation is to begin negotiating something like Plan A as soon as possible." The rest of Appendix A are "less ambitious ideas that may still help". Of the fallback ideas listed there, I support the model spec sort of transparency and improving government capacity. I'm indifferent on export controls. I oppose the others listed given the current evidence.
English
0
0
0
20
David Manheim
David Manheim@davidmanheim·
@ramez @slatestarcodex The specifics of the immediate plan are transparency, enforced export controls, more R&D on verification, tracking compute, improving government AI capacity, and registering and limiting training of future models. It sounds like you only support the first 5, correct?
English
1
0
0
32
Ramez Naam
Ramez Naam@ramez·
Yes. We need epistemic humility. Before proposing a global surveillance state, concentrating all western AI compute in Mongolia so it can be easily destroyed, or whatnot, we should have more than lines on graphs. [And I love lines on graphs!]
Joshua Saxe@joshua_saxe

I'm perplexed at why people who I know are smarter than me, including the authors, buy the epistemic strategy used in the AI 2040 scenario, in which, tl;dr, the famous METR time horizon plot predicts eschatological doom. I don't believe the authors have sufficiently grappled with why the radiologists and programmers still have their jobs, and why baristas haven't been replaced by automatic espresso machines Or even why we haven't seen more small-scale scheming / reward hacking / misalignment damages due to the use of coding agents across Big Tech, in which we've already granted such agents significant autonomy. I think if these gaps were more deeply considered they'd be making a more nuanced and less extreme argument. In the meantime I'm concerned that resources, advocacy and governance are being influenced by their conceptual model.

English
4
9
74
14.6K
Ramez Naam
Ramez Naam@ramez·
I'm in favor of many sorts of AI safety precautions, including staggered rollouts of new models to cyber security organizations and maintainers of core digital infrastructure to reduce cyber risk. I'm not calling for zero safety work. I'm saying that we have nothing close to the evidence required to justify the fairly draconian measures in AI 2040, and the attendant risk that those measures themselves bring.
English
1
0
1
40
David Manheim
David Manheim@davidmanheim·
@ramez @slatestarcodex The argument from ignorance isn't much of an excuse here. There was strong theory explaining why we should have expected failures such as superhuman cybersecurity capability, and it's been empirically validated by the misuse cases we see in models. Again: x.com/davidmanheim/s…
David Manheim@davidmanheim

@ramez I think @DKokotajlo agrees that these are really bad tradeoffs to need to make, but as you say, they are convinced the risk of the "unproven threat" is significant. But not being willing to trade anything to prevent risks until it is too late seems like an insufficient response!

English
1
0
2
67
Ramez Naam
Ramez Naam@ramez·
@Dr_A_Stein Fair. Yet the Hooker Telescope wasn't intended to discover that the universe was expanding. The instrument came first. Then Hubble found something startling with it.
English
0
0
0
22
Adam Stein
Adam Stein@Dr_A_Stein·
@ramez It's inherently both. Sensors are tools. You're right that many great discoveries came from leveraging those tools. But tools are usually made for an initial purpose. The telescope was made to see farther, we didn't see farther because someone happened to invented the telescope
English
1
0
2
68
Ramez Naam
Ramez Naam@ramez·
I'm not the scientist that Dr. Young is, but throughout history, most of the largest scientific advancements came from new sensory and experimental instruments, not just thinking big thoughts. Telescope, microscope, spectroscopy, X-ray crystallography, PCR, gene sequencing, spectroscopy. Yes, there are some pure conceptual breakthroughs, like General Relativity and (sort of) antimatter. But mostly the instrument, sensor, or experiment comes first. AI will augment scientific discovery by pouring over data and perhaps by automating experimentation. But the sensors, data, and experiments are the bedrock of our understanding of reality.
Dr. Catharine Young@DrCatharineY

I’m still amazed by how many people think AI will eventually replace basic science. Or that we need less discovery because of it. Genuine breakthroughs require new observations. Without it AI will eventually run out of new knowledge to learn from.

English
10
5
72
17.1K
Ramez Naam
Ramez Naam@ramez·
@TeslaZoneUSA Tell that to the dishwasher or laundry machine. Or to the forklift.
English
0
0
0
18
TeslaZone
TeslaZone@TeslaZoneUSA·
@ramez Counterpoint - all physical work related tools and processes have been designed for use with human bodies - it will be easier and more cost effective to adapt robots to these systems than vice-versa.
English
1
0
0
11
Ramez Naam
Ramez Naam@ramez·
I agree with Martin here. What I've seen does not convince me that humanoid robots are going to change the world anytime soon. Robots are an incredible technology and the world will increasingly be suffused by them - but most will not be humanoid.
Martin Ford@MFordFuture

Terrific article by Stephen Witt in the @NewYorker about progress in humanoid robotics. I'd be happy to be wrong, but I fear the people making big investments in building humanoids (including @elonmusk) are making the same mistake that Zuckerberg (@finkd) made with his investments in VR and the metaverse: Someday it will be huge--but just not yet. However, non-humanoid robots in more controlled environments like warehouses and factories are continuing to become more capable and dexterous and will have a significant impact on jobs. Amazon, for example, has already declared its intention to scale the business via efficiency improvements (#AI and #robotics) rather than hiring large numbers of new workers. I cover all this in more depth in the new edition of my book "Rise of the Robots: Technology and the Threat of a Jobless Future." (Link to article in the reply) #RiseoftheRobots

English
6
2
35
6.6K
Ramez Naam
Ramez Naam@ramez·
Thanks for the thoughts, as always. I consider you and the AI Futures team good faith actors. I'm not upset with you. I simply see the rising authoritarianism of government around the world as a much greater threat than hypothetical runaway ASI. This is grounded in empirics. In regards to policy, the more evidence of a clear and likely harm, the more aggressive the policy I'd support. In the meantime, in a period where there is minimal evidence of risk, I do support a great many AI safety policies including many we're not implementing. These are what I classify at 'no regret' policies (pandemic preparedness, cyber hardening, 3rd party AI auditors) which have little to no infringement on rights and yet improve safety. Policies which infringe on rights and freedoms need much stronger evidence to justify them. Infringement on the 1st or 4th Amendment requires that we have very concrete evidence of imminent threats. That simply isn't present here. Thought experiments and expressions of p(doom) are not evidence of a real and imminent threat. It doesn't matter what any of our p(doom)s are. It's a matter of principle that we do not give governments (the truly most powerful entities on earth) additional power to restrict or invade the rights of their subjects expect in the most extraordinary of circumstances. That does not rule out AI safety. There is much more AI safety work that can be done without such intrusion into rights and without such increase in the architecture of authoritarianism. I'll try to respond at more length in the coming days.
English
1
0
3
299
Scott Alexander
Scott Alexander@slatestarcodex·
I tried to respond to an argument like this at blog.aifutures.org/i/168032563/5-… (if the internal link doesn't work, it's Part 5: "Can We Carve Out A Category Of Speculative Risk And Then Ignore That Category?") I think the reason this keeps coming up is that your position can't survive putting a probability estimate on the risk. If you say some normal probability like 5%, then it becomes obvious that it's worth putting a lot of effort into preparing. If you say some crazy low probability like 0.00001%, then it becomes obvious that you're the overconfident one and not us. So the only solution is to make some kind of argument that it's preemptively banned from discussion because it's "speculative", as if ignoring the possibility of black swans was some sort of principled rule of good decision-making. I think something might have escaped a conditional here. You seem upset as if we're instituting chip regulations now. But the chip regulations only happen after there's political will, which only happens if AI proves to be scary. What AIFP wants now is to "build a pause button", a low-risk endeavor which mostly involves designing technology to switch data centers from training to inference (we don't have to install the technology!) and starting negotiations with the Chinese where we set mutual red lines. Maybe the most useful thing to do here is: what's your red line? What would have to happen for you to support Plan A? What (short of the very catastrophe we're hoping to prevent) would move this out of the "speculative" category for you, and make it seem worthwhile to put more effort into preparing for?
English
2
1
40
817
Alex Imas
Alex Imas@alexolegimas·
We’ll then realize that it’s easier to exchange a token of value that allows us to prevent the problem of double coincidence of wants. Harder favors will earn more tokens than easier favors. We’ll start using these tokens so much that they’ll need to get standardized. We’ll form an organization that has a monopoly on making these tokens and trust that it doesn’t cheat. We’ll also need places to store these tokens; those places will lend tokens at interest.
English
6
7
91
7.6K
alz
alz@alz_zyd_·
When AI puts everyone out of a job, I guess all the unemployed people will start doing each other favors, like you cook me food, and I fix your toilet in return. I wonder what we should call all these favor exchanges
English
16
3
195
16.5K
Ramez Naam
Ramez Naam@ramez·
@curiouswavefn Very much agree. I use dishwashers as examples as well. Those are real robots.
English
0
0
1
54
Ash Jogalekar
Ash Jogalekar@curiouswavefn·
The example I give is the simple one of a dishwasher. It would be foolish to design a humanoid robot for dishwashing that looks and acts like a person doing dishes. For most jobs that can benefit from robots, we need to think like a robot in terms of design, not like a human(oid).
English
2
0
8
394
Ramez Naam retweetledi
Adi
Adi@adi_baradwaj·
There's a motte-and-bailey trick being played between two definitions of "superintelligence" "Big-S Superintelligence", which is definitionally omnipotent And "small-s superintelligence", which is the hypothesized end state of the current thread of AI research It's surprising how uncritically the AI safety community treats these dubious equivalences, especially since one of the most famous treatments of the motte-and-bailey fallacy comes from none other than Scott Alexander, who's quite prominent in the community
Timothy B. Lee@binarybits

I struggle with what to say about the new AI 2040: Plan A website. It all seems so implausible to me that I'm not sure where to start. There's an epistemic chasm between those who think superintelligence implies near-omnipotence and those (like me) who don't.

English
10
7
76
28.4K
Ramez Naam
Ramez Naam@ramez·
@annhafter New nuclear is quite expensive outside of China and S. Korea. Already built nuclear is cheap.
English
0
0
3
55
ann hafter
ann hafter@annhafter·
@ramez Its vastly more expensive than any other form of energy. Im not against nuclear,but "racing" to open an older technology, closed plant is hardly the direction this country should be going.
English
4
0
5
61
Andy Masley
Andy Masley@AndyMasley·
My simple test whenever I meet anyone in DC making strong claims about China is “Do you know who Deng Xiaoping is?” And the failure rate is way higher than you might think
English
122
107
2.7K
943.8K
Ramez Naam
Ramez Naam@ramez·
@Perrid13 @tenobrus The limiting factor on how many spousal murders there are is not whether or not you have access to an LLM to help you think through how to hide it.
English
0
0
0
42
Jacob
Jacob@Perrid13·
@ramez @tenobrus I mean neither is the demand that everyone else purposefully blind themselves to allow crime. Facilitate the damage and danger
English
1
0
0
39
Jacob
Jacob@Perrid13·
@ramez @tenobrus Libertarianism as a suicide pact. Nukes for everyone! What could go wrong.
English
1
0
2
59
Ramez Naam
Ramez Naam@ramez·
@Anthurium_xanth I believe that government should have the same power to restrict AI content as they do to restrict what can be published on the internet or in books. And that is nearly zero.
English
1
0
1
15

Keşfet

@adeoressi @slatestarcodex @Dr_A_Stein @TeslaZoneUSA @alexolegimas @alz_zyd_ @curiouswavefn @elonmusk