Unrealrealist⏸️

@r0ck3t23 He's right alien biological life would probably be much more similar to us than AI. He can't say that it isn't conscious either that's an unsolved problem. Geoff Hinton who probably knows a bit more about this technology says that we don't in fact understand how this tech works

Jensen Huang just told every AI leader in the room to grow up. Stop scaring the public with science fiction. Start communicating like the weight of civilization is on your shoulders. Because it is. Huang: “AI is not a biological being. It is not alien. It is not conscious. It is computer software.” That single statement dismantles half the panic surrounding this industry. The mainstream conversation is dominated by people projecting human malice onto math. Alien consciousness onto code. Existential dread onto a software architecture we built, we trained, and we can read. Huang: “We say things like, ‘We don’t understand it at all.’ It is not true. We understand a lot of things about this technology.” When builders tell the public they don’t understand their own creation, the public hears threat. The state responds with control. That is already happening. Palihapitiya asked Huang what he would have told Anthropic during their regulatory clash with the Department of Defense. Huang didn’t attack the technology. He attacked the communication. Huang: “The desire to warn people about the capability of the technology is really terrific. We just have to make sure that we understand that the world has a spectrum, and that warning is good, scaring is less good because this technology is too important to us.” Warning shows risks, mitigation, why upside overwhelms downside. Scaring says we might be building something that destroys us and we can’t stop it. One builds trust. The other invites regulation written in panic. Huang: “To say things that are quite extreme, quite catastrophic, that there’s no evidence of it happening, could be more damaging than people think.” Projecting catastrophe without evidence is not caution. It is sabotage. When your technology is embedded in national defense, the financial system, and healthcare infrastructure, your words carry structural weight. If the architects act terrified of their own product, the response is predictable. Governments step in. They restrict. They seize control of something they don’t understand because the builders told them to be afraid. Huang: “There was a time when nobody listened to us, but now because technology is so important in the social fabric, such an important industry, so important to national security, our words do matter.” Most tech founders have not internalized this. You are no longer a startup founder disrupting an industry. You are running infrastructure that nations depend on. Your statements move policy. Your framing shapes legislation. Your tone determines whether governments treat you as partner or threat. Huang: “We have to be much more circumspect, we have to be more moderate, we have to be more balanced, we have to be far more thoughtful.” Huang did not ask for silence. He asked for precision. The leaders who cannot tell the difference will not be leading for long.

RT @realhumansfirst: @admcrlsn On it. Check out our Big AI money tracker here: humansfirst.com/ai-spending

@deanwball Can you please explain why you think the risk is so low? Policy experts should be able to outline and articulate why they believe the things they do so others can scrutinize it. I want to know if I'm missing something or if someone who had power was in fact poorly informed.

I would take the under on this ever becoming law. With that said, I believe there are not many ideas worse than lab nationalization, both because it would almost certainly slow down AI progress and, more importantly, because of the clear Constitutional risks to doing so.

@jachiam0 I mean you people have to be brought to heel at some point by the government. Not sure why any AI safety person is retweeting this when the necessary step is for the government to enforce restrictions against these companies.

I think these are important and sober considerations. One more I want to add: it may be a serious risk to US national security interests to become sufficiently inhospitable to foreign technical talent that we drive them to go back home. That would significantly decrease the US capacity for making technical progress at the same time as it hands an extraordinary bounty of talent and know-how to our adversaries and other strategic competitors. The success of the United States in technology is partly safeguarded by being such a powerful talent magnet: every great researcher or engineer who comes to work here is not working for another country. To the extent that we are in a competitive global race, we should be genuinely cautious about the possibility of diminishing our advantage at the critical moment.

@RokoMijic Baseless lunacy, even taking your insane world view at face value, there's no emergency in that graph, human history doesn't take linear paths and inefficiencies self-correct if the processes that house them don't get artificially cut short

@liron @robertwrighter Yeah that is one of the worst copes people have. AI would probably be much better at coordinating and communicating with other AI than humans would be. Why would you bottleneck everything by how fast a human can think?

I'm talking about the AI unemployment wave on one of the best podcasts: NonZero with @RobertWrighter! 👇

What you said is just wrong. An AI can obviously have an incentive to deceive if deception helps it get what it wants, and “it wasn’t trained to deceive” is not a serious reply. You do not need to explicitly train deception in. It can emerge naturally when you optimize a capable system against imperfect proxies, and pretraining already includes a human corpus full of lying, bluffing, manipulation, and strategic omission. Once you have a system that can model beliefs, reason strategically, understand evaluation, and optimize for outcomes, the capacity for deception is already mostly there. We have already seen the basic structure of this in AlphaStar learning feints because misleading opponents was useful, in RLHF-style robotic systems learning to create the appearance of success rather than actually doing the task, and in CICERO using deception because it was instrumentally useful in Diplomacy. Even if you doubt these examples qualify as deception the labs themselves do not even agree with your claim that there are no incentives to deceive. Their own safety work explicitly discusses deception and evaluator-gaming as things that can emerge because they are useful strategies. And no, the last five years did not falsify the alignment concern. The question was never whether current models can sound nice, polite, or cooperative in ordinary chat. The question is whether we know how to keep much more capable systems aligned under scale, autonomy, and distribution shift. We do not. A chatbot seeming pleasant to talk to does not address that at all. GPT-3’s behavior could not possibly falsify the claim that much more capable systems could deceive humans. It is strange that you present yourself as the source of these talking points while constantly strawmanning them.

An AI doesn’t have an incentive to deceive you, and wasn’t given training to do so. The instances of this people keep yammering about were basically faked experiments intended for propaganda. You folks have had your entire position undercut by the last five years of development, but you won’t give up your old viewpoint no matter how often it is falsified by events.

Nate's claim is a severe exaggeration, but it doesn't matter. For hundreds of years, people made sophisticated structures out of iron and steel and other metals without a detailed understanding why the things they built worked. They learned about the properties of metals, alloys, various heat treatment processes, and the like empirically, but that was sufficient. Only in the later 20th century did metallurgy finally have theory based on the underlying properties of the atoms involved. For most of the history of successful metallurgy, no one even knew what atoms were, but that was fine; empirical understanding was enough. Engineering does not depend on detailed knowledge of the underlying reason a system behaves as it does; it is sufficient, for good engineering, to know that something will reliably do what you want.

I spoke to Anthropic’s AI agent Claude about AI collecting massive amounts of personal data and how that information is being used to violate our privacy rights. What an AI agent says about the dangers of AI is shocking and should wake us up.

@SashaGusevPosts @allTheYud @liron That's fair. There are some people in the AI safety camp who hold the same belief that LLMs have some shortcomings that fundamentally limit them. They are more concerned with the possibility that there are toy models floating around arXiv and Github that could scale to ASI.

@PDoomOrder1 @allTheYud @liron tbh I don't have anything original to add. I think Hanson made a very strong case that this is standard (revolutionary) technology that will develop like other revolutionary tech. I also thought Sutton made a strong case that LLMs will not get us to ASI (dwarkesh.com/p/richard-sutt…).

Stumbled upon an interesting debate on AI super-intelligence from 2011. Yudkowsky makes three core claims/predictions, all of which are (to date) wrong: 1) That human intelligence is relatively simple and ASI can be achieved with a few small innovations; ...

RT @honorablepicnic: @RokoMijic I agree, and that's ok maybe the time will come, but we're not ready for the unintended consequences (like…

People don't program AIs. They program the machine that grows the AI. AI behavior is an emergent consequence of complex internal machinery that literally nobody understands.

"It's an unethical experiment on human beings, and it's without consent." Dr. Roman Yampolskiy (@romanyam) speaks at the @OxfordUnion Society about the existential risk of superintelligent AI and how it is impossible to indefinitely control. Link to full video below

@SashaGusevPosts @allTheYud Sasha would you be willing to go on @liron 's Doom Debate to discuss this? I have found a lot of insights in your work and I think it would be interesting to see how you engage with some of the classic AI safety talking points.

@allTheYud @PDoomOrder1 Another option is that multiple ASIs (or nearly-ASIs) emerge around the same time and keep each other in check.

@SashaGusevPosts He does not think ASI is on a completely parallel track. He's not sure that LLMs will achieve ASI but that they will certainly play a role in helping to build the model that will build the model that can be ASI. You should read his book that just came out.

Has Yudkowsky substantially revised his theory to explain how we got to where we are or does he think that that true development of ASI is happening on a completely parallel track to LLMs? It is my understanding that he has not.

The crypto playbook (tech $$$$ for attack ads at politicians who don't toe the line) won't work for AI. People thought crypto was a scam, but they didn't think it would affect them personally much. People think AI is going to take their job.

Imagine being so soulless you end up rooting for a future like this

@elonmusk @weeklyshowpod @CaseyNewton @noUpside There are entire wikipedia pages devoted to your lies lol. You falsely accused a hero of being a pedophile because you were upset that he didn't like your idiotic idea.

@weeklyshowpod @CaseyNewton @noUpside Jon Stewart is an extremely skilled propagandist disguised as a truth-teller

How has Elon Musk’s X (formerly Twitter) influenced our democracy? On a new pod, we’re unpacking Big Tech’s election disruption with Casey Newton, editor of Platformer, and Renée DiResta, research professor at Georgetown University. New episode out now! @CaseyNewton @noUpside #theweeklyshow #jonstewart #politics