This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
The future of AI is likely decided this week with Sam Altman's Congressional testimony. What do you expect?
EDIT: the recording is here.
Frankly I've tried to do my inadequate part to steer this juggernaut and don't have the energy for an effortpost (and we're having a bit too many of AI ones recently), so just a few remarks:
AI Doom narrative keeps inceasing in intensity, in zero relation to any worrying change in AI «capabilities» (indeed, with things like Claude-100K Context and StarCoder we're steadily progressing towards more useful coding and paperwork assistants at the moment, and not doing much in way of AGI; recent results seem to be negative for the LLM shoggoth/summoned demon hypothesis, which is now being hysterically peddled by e.g. these guys). Not only does Yud appear on popular podcasts and Connor Leahy turns up on MSM, but there's an extremely, conspicuously bad and inarticulate effort by big tech to defend their case. E.g. Microsoft's economist proposes we wait for meaningful harm before deciding on regulations – this is actually very sensible if we treat AI as an ordinary technology exacerbating some extant harms and bringing some benefits, but it's an insane thing to say when the public's imagination has been captured by Yuddist story of deceptive genie, and «meaningful harm» translates to eschatological imagery. Yann LeCun is being obnoxious and seemingly ignorant of the way the wind blows, though he's beginning to see. In all seriousness, top companies had to have prepared PR teams for this scenario.
Anglo-American regulatory regime will probably be more lax than that in China or the Regulatory Superpower (Europeans are, as always, the worst with regard to enterpreneural freedom), but I fear it'll mandate adherence to some onerous checklist like this one (consider this as an extraordinary case of manufacturing consensus – some literally who's «AI policy» guys come up with possible measures, a tiny subset of the queried people, also in the same until-very-recently irrelevant line of work, responds and validates them all; bam, we can say «experts are unanimous»). Same logic as with diversity requirements for Oscars – big corporations will manage it, small players won't; sliding into an indirect «compute governance» regime will be easy after that. On the other hand, MSNBC gives an anti-incumbent spin; but I don't think the regulators will interpret it this way. And direct control of AGI by USG appointees is an even worse scenario.
The USG plays favourites; on the White House meeting where Kamala Harris entered her role of AI Czar, Meta representatives weren't invited, but Anthropic's ones were. Why? How has the safety-oriented Anthropic merited their place among the leading labs, especially in a way that the government can appreciate? I assume the same ceaseless lobbying and coordinating effort that's evident in the FHI pause letter and EU's inane regulations is also active here.
Marcus is an unfathomable figure to me, and an additional cause to suspect foul play. He's unsinkable. To those who've followed the scene at all (more so to Gwern) it is clear that he's an irrelevant impostor – constantly wrong, ridiculously unapologetic, and without a single technical or conceptual result in decades; his greatest AI achievement was selling his fruitless startup to Uber, which presumably worked only because of his already-established reputation as an «expert». Look at him boast: «well-known for his challenges to contemporary AI, anticipating many of the current limitations decades in advance». He's a small man with a big sensitive ego, and I think his ego will be used to perform a convincing grilling of the evil gay billionaire tech bro Altman. Americans love pro wrestling, after all.
Americans also love to do good business. Doomers are, in a sense, living on borrowed time. Bitter academics like Marcus, spiteful artists, scared old people, Yuddites – those are all nothing before the ever-growing legion of normies using GPT-4 to make themselves more productive. Even Congress staff got to play with ChatGPT before deliberating on this matter. Perhaps this helped them see the difference between AI and demons or nuclear weapons. One can hope.
Scott has published a minor note on Paul Ehrlich the other day. Ehrlich is one of the most evil men alive, in my opinion; certainly one of those who are despised far too little, indeed he remains a respectable «expert». He was a doomer of his age, and an advocate for psyops and top-down restrictions of people's capabilities; and Yud is such a doomer of our era, and his acolytes are even more extreme in their advocacy. Both have extracted an inordinate amount of social capital from their doomerism, and received no backlash. I hope the newest crop doesn't get so far with promoting their policies.
Yep, it's amazingly bad, especially LeCun.
I think it's because Anthropic has an AI governance team, led by Jack Clark, and Meta has been head-in-the-sand.
I know him and I agree with your assessment. Most hilarious is that he's been simultaneously warning about AI dangers, while pettily re-emphasizing that this is not real AGI, to maintain a veneer of continuity with his former life as a professional pooh-pooh-er.
Re: his startup that was sold to Uber - part of the pitch was that Gary and Zoubin Ghahramani had developed a new, secret, better alternative to deep learning called "X-prop". Astoundingly to me, this clearly bullshit pitched worked. I guess today we'd call this a "zero-interest-rate phenomenon". Of course X-prop, whatever it was, never ended up seeing the light of day.
Yep, we realize this. The economic incentives are only going to get stronger, no one who has used it is going to give up their GPT-4 without a fight. That's why we're focused on stopping the creation of GPT-5.
Can you clarify your reasons for joining the doomer camp? To be honest, I've been immensely disappointed by both sides of the debate, it feels like only randos on Twitter are clear-headed about it. Hinton with the *shocked* realization that his algorithms do work better than their biological counterparts, ridiculous naysayers who rely on shallow zingers and psychologizing, policy suits squawking over their sinecures, and the confused, exploited mob in the middle.
I think my perspective is clear enough. I don't care about stuff like muh spear-phishing and faked voices because it's just noise. I don't buy technical doom narratives out of Yuddist camp because they're bad Sci-Fi and crumble under scrutiny. I don't feel fear from the technically semi-literate ones («optimality», «instrumental convergence» stuff) because dangerously diverging designs yet capability-preserving designs are speculative and don't really make economic sense even at early steps. I don't worry about hacking, bioweapons and other serious problems based on amplification of human malice because they ought to be trivially surmountable with tool AIs of comparable level. In general, all AI doom plots just keep reinventing Bostrom's idea of a Singleton arising in a technically primitive world which is, of course, defenseless against it, and this isn't how this is playing out so far.
But above all, I have faith in the human will to power. We are not economic agents, we are apes. We rise to the top driven by the desire to see tiny apes below. This is both a blessing and a curse: the Tech Lords (or the government that expropriates their genies) are not willing to cede power to a glorified Microsoft Clippy, no matter how much better it gets at doing their own jobs. It's a curse because the top apes may not see much point, long-term, to preserving the proles in their current numbers and standing. But handing these elites the power to regulate proles out of this technology doesn't solve that issue! Distributing it widely does! Indeed, even the politicians in this hearing are appreciative of the empowerment effect that AI can provide, or at least pay lip service to it.
Do you just mean that GPT-5 would give OAI/MSFT too much of an edge? Or do you mean this level of capability in principle?
Thanks for asking. You're probably the person I see most eye-to-eye about this who disagrees with me.
I agree that regulating AI is a recipe for disaster, and centralized 1984 scenarios. Maybe I lack imagination about what sort of equilibrium we might reach under wide distribution, but my default outcome under competition is simply that I and my children eventually get marginalized by our own governments, then priced out of our habitats. I realize that that's also likely to happen under centralized control.
I think I might have linked this before, as a more detailed writeup of what I think competition will look like:
https://www.lesswrong.com/posts/LpM3EAakwYdS6aRKf/what-multipolar-failure-looks-like-and-robust-agent-agnostic
I'd love to think more about other ways this could go, though, and I'm uncertain enough that I could plausibly change sides.
This level of capability in principle, almost no matter who controls it.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link