site banner
PaperclipPerfector  30d ago (text post)   29021 thread views

Culture War Roundup for the week of August 19, 2024

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.

5
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Is it so over for OpenAI? I just signed up for a Grok subscription for $7/month. Apparently the reasoning performance of their new model is on the same tier as ChatGPT-4o and Claude Sonnet. Image generation is best-in-class (using Flux). And of course it's much less neutered than the competition.

The miracles that OpenAI accomplished in the last 3 years now seem rather... commonplace. There's a lot of competitors at nearly the same level. Facebook's open-sourced AI, Llama 3.1, will commoditize the space. While it's not really feasible to run these models on your home computer (yet) it will be easy for smaller companies to buy compute and then sell Chat-GPT similar services. It will be a race to the bottom now.

OpenAI is running at a gigantic loss. I'm sure they were planning to capture a monopoly and then raise prices. That seems less likely to work now as their product is undifferentiated.

And the irony of Elon controlling a Chat-GPT similar is just too delicious. OpenAI management stole the IP of the non-profit for their own financial gain. Now open source is routing around the damage and rebuilding it all from scratch. I bet Grok's total development costs are like 1% of OpenAI's.

I don't think its "over" so much as the hype has run it's course leaving only sober analysis in it's wake.

There are definite use cases for LLMs and Star Trek-style universal translators are a genuinely revolutionary "killer app" but LLMs are not "reasoning engines" nor were they ever a likely path to true AGI, and this was fairly clear early on to those in the know, but sober analysis along the lines of "this is signifigant but now where near the major breakthrough it has been portrayed as" doesn't drive social media engagement and attract VC dollars the way "I Created an AI Scientist" does, so naturally its the latter that got signal boosted.

I just signed up for a Grok subscription for $7/month

Which region are you in where X Premium+ is $7/month? In the US/UK/EU, it's (respectively) $/£/€16

United States.

OpenAI management stole the IP of the non-profit for their own financial gain.

What? How?

We discussed this at length last year but here's a short rundown.

In 2015, OpenAI was founded by Elon Musk, Sam Altman, and others. Elon was by far the largest financial supporter. OpenAI was a non-profit, dedicated to sharing its research openly (thus the Open in its name).

Today, though the non-profit fig leaf still exists, OpenAI exists as a closed, for-profit company, half of which is owned by Microsoft.

Last year, Sam Altman was fired by the non-profit board because he was not being honest with them. However, many employees had stock grants worth tens of millions due to the deal with Microsoft. With these grants being threatened, the employees pledged to leave en masse and work for Microsoft directly. The board caved, and now Sam Altman has de-facto complete control of the "non-profit" board.

For that reason, Elon is suing OpenAI for breach of contract as they perverted the mission of the original non-profit for their own financial gain: https://www.cnn.com/2024/03/01/tech/elon-musk-lawsuit-openai-sam-altman/index.html

It’s hilarious how these are the exact wrong people you want possessing decision-making capabilities regarding AI. Like, the moral test was placed in front of them, and they all failed it. They chose money over (1) honesty (2) their own pledged word (3) morality (4) the public Good.

I suspect that outside of a very small handful of genuine Yudkowsky-types, almost nobody who claims to be concerned about AI destroying the world is actually worried about AI destroying the world. They may say they are worried, but they are not actually worried deep down. The idea of AI destroying the world is abstract and distant, on the other hand getting tens of millions of dollars is very real and very visceral.

And for every one person who is genuinely worried about AI destroying the world, there are probably a hundred people who are worried about AI allowing Nazis to write bad no-no things online. Because Nazis writing bad no-no things online feels real and visceral and it pushes the deep buttons of ideology, whereas AI destroying the world sounds like a sci-fi fantasy for geeks.

almost nobody who claims to be concerned about AI destroying the world is actually worried about AI destroying the world.

Then the question would be why they would make such claims. I can see two reasons: (1) Signaling value. However, outside of the Less Wrong bubble, the signaling value of believing in p(doom)>0 is negative. Also, a significant fraction of partisans generally tend to believe the fears endorsed for signaling value: if some people are concerned that a Republican/Democrat will lead the US to fascism/communism, I think their fear may be genuine. Granted, they will not act rationally on their fears -- like emigrating to a safer country before the election. (2) Hyping AI. "Our toys are so powerful that our main concern is them taking over the world". This is certainly a thing, but personally, if I wanted to hype up the public about my LLM, I would go for Culture (post-scarcity), not Matrix (extinction).

As an anecdote, I happen to believe that p(doom) is a few percents. Bizarrely, despite me being a self-professed utilitarian, this does not affect my decision on where to be employed. I mean, given that alignment research is not totally saturated with grunt workers, and that there is a chance it could save mankind (perhaps lowering p(doom) by a third), it would be hard to find a more impactful occupation.

I think the reasons for my bizarre behavior (working conventional jobs) are as follows: (1) Status quo bias, social expectations. If half of my friends from uni went into alignment, this would certainly increase the odds for me as well. (2) Lack of a roadmap. Contrast with the LHC. When it was designed in the 1990s as a tool to discover the Higgs and SUSY, there was a plan. Ambitious, but doable, no big essential white spots marked "to be solved by technology yet to be discovered". Becoming a tiny cog in that machine, working on an interface for the cryo controls for the magnets or whatever would have been appealing to me. By contrast, AI alignment feels more like being kids on the beach who thinks there will be an incoming tide, and try to reinforce their sand castles so that they will withstand the water. It is possible that some genius kid will invent cement and solve the tide problem, but it is not something one can plan. Statistically speaking, I would likely end up in a team who tried to make the sand stickier by adding spit or melt the sand into lava over a campfire. The main reason our sand castles would survive would likely be that we are on the shores of a lake and the tide will end up rising only half a centimeter. This might be a psychological flaw of mine, but I prefer to make legible contributions with my work.

Of course, this means that you can say "by revealed preference, this means that quiet_NaN does not believe p(doom) to be in the percent range".

I know!! Gah if OpenAI really did secure the monopoly that would've been the darkest timeline. I definitely believe Altman and the rest of that cadre are incredibly corrupt, if not downright evil.

It's heartening to see how much genuine competition there is out there.

Why does everyone forget that there was a 3 year gap between GPT-3 and GPT-4?

Give it time. At least wait and see what GPT-5 brings before declaring their premature demise.

I didn't forget that, I think that it's a sign that returns are rapidly diminishing on their scaling approach and they are struggling to wring some new hotness out of what they've got.

I'm no fan of sama and the cabal he's built, but nonetheless I think it's still too early to write off any major company working on AI-related things right now. I'm not convinced all of the low-hanging fruit has already been picked wrt applications (even dumb stuff like "characterai but uncucked" alone is likely to be a smash hit), and besides most past/present developments were sufficiently arcane and/or undercover that you can't really predict where and what happens next - cf. Chinese LLMs being regarded as subpar until Deepseek, or Anthropic being safety fanatics with only a dumb soy model to their name until Claude 3(.5).

If Sora is anything to go by I think OpenAI still have some genuine aces up their sleeves, and while I don't believe they're capable of properly playing them to full effect, they at least have the (faded) first-mover advantage and Sam "Strawberry" Hypeman to exploit normies boost their chances.

I agree that it's hard to predict what happens next. But the next development could come from anywhere, even China. Two years ago this wasn't true. Back then, OpenAI was heads and shoulders above the competition.

My claim is not about AI in general but only that OpenAI is no longer special. Although the fact that everyone seems to have reached (but not exceeded) the level of Chat-GPT 4 might seem to indicate a plateau.

As regards Sora. In my mind, it was a neat demo but ultimately a dead end and a distraction. Where's the use case?

My claim is not about AI in general but only that OpenAI is no longer special.

That much is true, I agree.

But the next development could come from anywhere, even China. Two years ago this wasn't true. Back then, OpenAI was heads and shoulders above the competition.

I agree as well but I'll note the obvious rejoinder - the next development could indeed come from anywhere, even OpenAI. Sure, they get mogged left and right these days, whodathunk propping yourself up as a paragon/benchmark of LLM ability backfires when you drag your feet for so long that people actually start catching up. But this still doesn't negate their amassed expertise and, more realistically, unlimited money from daddy Microsoft; unless said money was the end goal (which to be fair there is nonzero evidence for, as you note downthread) they're in a very good position to throw money at shit and probe for opportunities to improve or innovate. Surely Altman can see the current strategy of resting on laurels is not futureproof right?

As regards Sora. In my mind, it was a neat demo but ultimately a dead end and a distraction. Where's the use case?

Fair enough but still counts as advancement imo, even though tech like that is guaranteed to be fenced off from plebs, no points for guessing what (nsfw?) the usual suspects try to make with "ai video" generators in this vein. I generally stopped looking for immediate use cases for LLMs, I think all current advancements (coding aside) mostly do not have immediate useful applications, until they suddenly will when multiple capabilities are combined at once into a general-purpose agentic assistant. Until one arrives, we cope.

Are there any other industries that AI has really had an effect on aside from coding and graphic design? Search maybe?

I use it daily for these tasks and I think most people like me do, but are there other industries like this?

Education? The kids are becoming lazier and more regarded than ever due to relying on LLMs to write all their assignments. Though I guess this doesn't destroy the industry itself, but it makes for teaching being even more frustrating than before.

Amazing that we now have a machine that will answer arbitrarily-worded questions about any commonly-taught topic, and the only education implications people talk about are fake assignment submittions.

Like no, the whole structure of the industry is now obsolete.

The more exciting implication is that future models may be able to act as personal tutors of a pretty high quality, with live audio input and output and visuals to boot. That sounds pretty good to me.

It's made educators no more obsolete than encyclopedias did.

Maybe the technology will improve, but AI is currently quite incapable of educating children despite what any cherrypicked demo might show you.

I have a friend who's a GMAT tutor and he's busier than ever, charging $300/hr. So even in a free market, no one wants to use AI as a teacher. For now at least...

As usual, the argument assumes no improvement in the models or any well-designed and marketed product to gain acceptance.

The unanswered question is whether kids who can't afford 300/hr are seeing benefits from LLM tutors.

I literally said: "Maybe the technology will improve".

So, no, I did not make that assumption.

Do you know how many kids who weren't able to afford tutoring are using it for tutoring?

Its a very strong claim that "AI is currently quite incapable of educating children."

Teaching has long been and will always be daycare, but the problem is that the carer must be able to discipline, and educators have lost that ability.

I would guess that many translation companies that previously used boutique translation models are now just using "common" LLMs, which has constituted yet another step in machine translation improving. Still a lot of translate-from-scratch work sent to me, though.

If we can get you to post quarterly reports on the state of the translation job market, that would be great. You're my personal barometer for when I should start taking AI seriously. My personal experience matches yours so far, and the idea that AI is some sort of a revolution in my field (software dev) is bizarre to me. So far it didn't deliver on anything more than a slightly better autocomplete. Some people use it as a replacement for Stack Overflow, which I still prefer to look up directly.

I've started retraining (returned to the university for a polsci degree), so hopefully I'll only be in the field for a few more years. (Partly preparation for if there's an unexpected leap in machine translation capacity that genuinely starts to eat into the job market, partly just that after ten years of freelancing and uncertainty a real job with a real salary and a real vacation has started to seem quite appealing).

Don't tell me I have to get someone to make a poli-sci AI so I can keep my barometer...

So far it didn't deliver on anything more than a slightly better autocomplete.

It is far better at that - for experienced developers doing something in area new to them

Some people use it as a replacement for Stack Overflow, which I still prefer to look up directly.

Often it is a superior replacement: you can get intern-quality work on any topic you want, very specific for your task. That sometimes is better than SO. And so far, on track of getting better and better.

I heard all that, tried it out for myself, and it just doesn't feel all that great to me. Maybe one day it will get better but it's just not all that useful in day to day tasks (which your own description kind of confirms).

you can get intern-quality work on any topic you want

That's terrible. I always seen internships essentially as charity work companies engage in to polish their PR, or, at best,extend their recruitment pool. You make it sound like the sale's pitch for Tesla's "full self driving (supervised)". It's awesome. Can take you anywhere. Almost no interventions... which results in you having to be ready to intervene at all times, or die driving into a truck.

Maybe one day it will get better, but I'd rather rely on Stefferi as my canary, than on the words of AI enthusiasts.

I wonder if we aren't observing some sort of split between students, hobbiests, web-devs and the like, and applications requiring genuine rigor. My own experience with AI generated code is largely summed up in this short here.

it's just not all that useful in day to day tasks

I guess it depends on how often you need to do stuff like "now I need to drop into badly written docs of a new setup and get simple program working".

I need it fairly often but I can imagine someone who needs it approximately never.

And I am quite surprised that it is useful even for that.

than on the words of AI enthusiasts

I am not an AI enthusiast

I've been hearing rumblings that it has made impact in the legal profession, big law firms are probably supplementing their paralegal staff, at least, and likely their attorneys.

We hear about the obvious ones that get caught fabricating caselaw wholesale, but I'd bet it's effective enough that most of the time no outsider notices.

https://www.floridabar.org/the-florida-bar-news/miami-dade-public-defender-is-using-artificial-intelligence-for-research-and-for-case-preparation/

I find it is excellent for small simple contracts. Want an NDA, a form for a leave of absence application, a privacy policy for a website, a small contract to sell a used car, a contract to hire the neighbors kid to work a bit during the summer as a tester or similar it works well. A will for a person who has simple finances or the paperwork that needs to be signed when checking into a hotel.

It is pretty great at coming up with domain/name combos for a business.

Chatgpt did a better job of explaining the swedish tax forms for me than the government website.

One thing I have realized after spending a few years in the startup scene is how legal expenses are a huge burden on a startup. Various contracts need to be drafted and they can cost thousands each. ChatGPT can draft these simple contracts.

I will use it for tasks where the scope of work is easily defined.

I've found it makes researching unfamiliar areas of law about 5x faster. It really seems to "understand" statutory interpretation and I've yet to catch it making a material mistake. It does a good job comparing and explaining two different statutes that are on point, especially when provided with current caselaw.

I have also used it to fine-tune an appellant brief. It is at least as helpful as having a particularly bright law student with access to massive knowledge reserves to bounce ideas off of.

A year or so back I suggested that any person who is currently in Law School should drop out. I think it is still good advice, although I understand the counter-arguments. It will not be long before these things are smarter and faster than any first-year associate.

What kind of law do you practice? I'm currently in litigation but I've done oil and gas law in the past and dabbled in bankruptcy and simple estate planning along the way, and I have a hard time thinking of any obvious uses for AI. It may make legal research easier, but I do legal research maybe a few times a year, and clients don't like paying for it so we usually only do it at their request, and they only seem to request it whenever I'm already pretty busy, so cutting my research billables by a couple hours wouldn't make much of a dent in the overall amount of work I have. The thing about most litigation is that few issues arise where there's any real fuzzy question that needs research. If you practice in one area the relevant appellate decisions are well-known and new ones are rare enough that it's news when they're handed down. This was even true when I was in oil and gas, and a relatively large number of decisions were being handed down during the boom, covering the three states I worked in.

Anyway, in litigation at least, I'm rarely ever doing the typical lawyer thing of applying the law to the facts and making an argument. What I spend most of my time doing is gathering facts and analyzing them so I can first make an argument to the client to get settlement authority in the amount I think I need and then making an argument to opposing counsel that they should accept what I'm offering them. The relevant information here is 1. The facts of the case at hand, and 2. The facts of other cases my firm has settled with Plaintiff's counsel. Any LLM would need access to hundreds of pages of depositions, thousands of pages of medical records, interrogatories, fact witness lists, expert reports, innumerable pages of discovery material, and other information each case generates. And then multiply this by every case the office has ever handled, and some that they didn't. Almost every case I handle involves discovery evidence and deposition testimony from prior cases that the Plaintiff is relying on as evidence. And I need it to digest the facts of all recent cases (at least the past 5 years, sometimes longer) to compare settlement amounts. In order to do this, a firm would need to be running their own AI servers, which would have to be training constantly. And that doesn't even get to the other problem, that AI can't take a deposition.

In oil and gas it's even worse since my job was in title, and title records are stashed in courthouses and often haven't been digitized. Some counties are getting better with digitizing land records but few counties have attempted to digitize historical probate records, and the ones that have don't have online access. I'm not aware of any county that has digitized historical court records. With the exception of Ohio, the counties that do have online access are fee-based, and I doubt many companies are willing to give AI the authority to charge credit cards. And once you do get the records, anything before about 1920 is going to be handwritten, often poorly, and anything before about 1970 is going to be typewritten in a way that OCR struggles with. Some online systems don't work off of a typical database, but simply have scanned index pages that require you to manually enter the book and page number you're looking for. These use indexing systems that computers have made obsolete, and it's an open question whether an AI could figure out how to use them absent specific instructions. But the ultimate question is whether or not the general AI's that exist now would even be able to understand what they're supposed to be doing. There's also the problem that even knowing if a particular instrument even applies to the parcel in question. In states that predate the US Land Survey System, property descriptions will often start with "Beginning at a white oak" or something similarly nonspecific, then run through survey calls. Sometimes the calls have inaccuracies that need to be untangled. Sometimes (particularly with old leases and ROWs) it will just state the owners of the adjoining property. Sometimes (pretty often, actually, a title chain will simply stop cold because it passed through an estate and the only record of the transfer is the probate record of the person who died, whose name you probably don't know. I could continue but you get the idea. Figuring out a title takes years of learning various techniques based on the resources available. And God help you if you work in West Virginia.

With bankruptcy and estate planning, while actual legal questions are more prevalent, the bigger issue is being able to advise clients about what they should be doing. The kind of people willing to half-ass estate planning are the kind of people who are going to get a basic will off of Legal Zoom for 80 bucks anyway and allow their heirs to deal with the consequences of the fact that their estate wasn't so simple after all. (Practically every client I did a will for told me their situation was "really simple" and this was almost never the case. One guy had property in another state. One couple had a blended family. One guy owned a fucking restricted business.) Bankruptcy is theoretically more straightforward, especially Chapter 7s, but bankruptcy clients need someone to tell them that things are going to be okay as much as they need legal advice. These people come into your office absolutely scared to death and want to hug you when they leave.

And then there's the thing that local courts have their own customs that can't easily be translated to LLMs. Does the PA Statute of Repose apply to equipment that's permanently affixed to a structure? In Cambria County it does, in Allegheny County it usually doesn't, and it's not something anyone is ever going to appeal. How will the bankruptcy trustee treat a particular situation? Depends on the trustee. These are things you can only know if you're a lawyer who practices in the jurisdiction, and there are no written opinions to guide the AI. I admit that it has some theoretical uses, but I wouldn't start telling people to drop out of law school just yet. I mean, there are plenty of reasons to not go to law school, but this isn't one of them.

I work in a specific heavily statutory / reg based area of the law.

I have asked it difficult questions about statutory interpretation and found that it missed a lot. So YMMV.

Just a note that you may be interested in my above reply.

If the model you're using allows you to upload information, it HAS helped to simply give it access to the corpus of laws that you're working with.

Digging deeper into regulatory law, beyond just the high level statutes where the rules and rulings may not have been part of the training data does seem like it would require heavier specialization.

God dang it's just so cool that we have so many smart and accomplished professional folks on here. Thanks for this insight, really interesting stuff.

Ironically I don't feel that accomplished because I consistently hang out in places where high level degrees and incredibly intelligent professionals is almost the baseline.

I have my moments, though.

Wouldn't it take the same amount of time to explain to ChatGPT what you want in the contract as it would to just write it yourself?

If you can dictate notes for the contract in about 5 minutes, you'll have a first draft from ChatGPT much more quickly than you could create one yourself, even using existing forms.

Teach it how to use you forms and it'd be even better.

ChatGPT can draft these simple contracts.

If I was gambling my career on a startup and handing over 30/40/50/60% equity to the venture capitalists I don’t know that I’d trust ChatGPT to let me know I was being screwed over [even more than expected].

I would probably get a better contract for something that big. But there are a million little agreements that have to be signed. I got a rental agreement for a 4 square meter space in an atic written by chatgpt. If I lose that small amount of money because someone actually wants to bring it to court I am still saving money.

I got a rental agreement for a 4 square meter space in an atic written by chatgpt.

I feel like I'm missing something here. I don't know much about law, but every rental agreement I've signed has been the locality's standard rental agreement. I don't think they were technically required to use it, but there was just no reason not to. Why aren't all of those simple boring contracts that are so trivial ChatGPT could do it just a standard contract that you fill in the blanks on like those rental contracts? How does the LLM help?

Tangentially relevant...

It's bizarre how much custom legal work gets down. It would seem that governments or standards bodies could provide "standard contracts" for things like wills, employment contracts, leases etc... Instead, everyone get their own bespoke contract, and most of them are badly written.

We already have a standard will. It's called intestacy law. In contracts cases, as long as there is some minimal reason to believe a contract exists, courts have no problem writing missing terms for you, and they're consistent enough to be predictable. You don't even need a price.

You'd be surprised (or not) how badly a layperson can fuck things up even using a basic form with instructions in plain english.

Standard contracts from reputable sources are available if you look for them. For example:

Sure, pretty much every industry where people write emails is being "disrupted". But Grok can write a bullshit email the same as ChatGPT or Claude or Gemini. It's all fake bullshit. Very little actual work is being done with LLMs.

Search is probably the killer app for AIs right now, but I wonder how long that lasts. Website owners are blocking AI bots, and for good reason.

https://blog.cloudflare.com/declaring-your-aindependence-block-ai-bots-scrapers-and-crawlers-with-a-single-click/

One day, AI will achieve the holy grail of being able to turn a PDF into a spreadsheet. But last time I tried it still didn't work.

A part of me thinks that the high end of AI is now being kept behind closed doors. Surely OpenAI has been doing something other than making GPT-4 smaller and lighter in the nearly two years since they've finished it.

At minimum they have the ability to produce lots of synthetic data to train future models with, now they have this newer and cheaper model. And they have far more compute than before.

Same with Anthropic. Where is Opus 3.5? Hidden from us mere mortals.

Wasn't this explicit when OpenAI announced Sora but gave no public access?

They surely have some impressive stuff that isn't even revealed yet.

To a large extent I would expect them to keep ChatGPT as the flagship product and any upgrades would just be added on as features rather than "separate" models.

I am wondering if they've lost first-mover advantage, since any other companies that were sniffing around the same research-space surely know where to direct efforts now, even if they haven't stolen IP directly.

My bet is on OpenAI having a couple sizeable rabbits to pull out of their hat, but they could be true witchcraft or mere illusions.

A part of me thinks that the high end of AI is now being kept behind closed doors. Surely OpenAI has been doing something other than making GPT-4 smaller and lighter in the nearly two years since they've finished it.

I'll admit that this thought has occurred to me. But given everything we know about human nature, do you think anyone could keep something of this magnitude under wraps?

The criminal mastermind always longs to confess his crimes so we can marvel at how clever he is.

Probably not indefinitely, but it is at least possible to keep something of that magnitude under wraps for a long time and effectively enough that the few occasional whistleblowers get dismissed as cranks or ignored until finally one day something comes out that is too big to be swept under the rug. Take the NSA domestic surveillance program, for example. It was successfully kept secret from the public for many years.

OpenAI are leaning hard into anthropomorphic design with their new voice features. These are unlike anything I've seen (or rather heard) on other platforms, and they're just rolling out a big update - Ethan Mollick had a good recent piece on it. While the idea of talking out loud to your AI may seem unnecessary, I've found it fits into a very useful niche in my workflow - e.g., getting an interactive lecture on long car journeys, workshopping a lecture while waiting for a train. And it's the one version of ChatGPT I've had success at pitching to older people in my life.

However, I suspect most of the revenue long-term from AI will come from B2B services. OpenAI are doing pretty well here - I don't have the exact figures, but chatting AI policy with some major (non-AI) firms, around half of them I've interacted with have service contracts with OpenAI (not Google, or even Microsoft) for an internal secure version of ChatGPT. However, these internal versions are typically worse than the off-the-shelf model, which is slowing employee adoption.

All in all I think it's way too early to write off OpenAI. OpenAI (and Anthropic) are definitely doing something a bit different by explicitly leaning into anthropomorphic design, as opposed to the Google/Microsoft/Apple model where LLMs should be utterly generic and boring and predictable.

They are still the top for now. No argument here.

But, a year ago, @sama said that AGI had been achieved internally and people thought he might be serious.

Now they are merely first among equals. As cool as the new features are, this is not AGI. I don't think they have any special sauce. There's probably nothing they can do that a startup with $100 million can't do. Hell, even China is making top-tier models now. OpenAI might end up as a successful company, they might not, but they are certainly not looking like the world beater that they did 18 months ago.

On the other side of coin, Elon is just crushing it on all levels now with Grok AI, Tesla FSD, Neuralink, and SpaceX. It's mind-boggling how he maintains all this despite apparently spending hours a day shitposting. GOAT.

But, a year ago, @sama said that AGI had been achieved internally and people thought he might be serious.

Some people still think he's serious. The "strawberry"/qstar meme train has been going for the better part of a year at this point I think, with the usual twitter personalities promising that it's just around the corner the whole time.

Elon is just crushing it on all levels now with Grok AI, Tesla FSD, Neuralink, and SpaceX.

Grok and SpaceX are the real deal. Neuralink seems promising but afaik there isn't much publicly demonstrated so far. Tesla FSD is multi year neverware best known for causing multi car pileups on US-101 thanks to oblivious drivers scrolling while driving.

You might need to update on FSD. The latest updates have been described as a phase change. It's not level 5 yet, but apparently they have a clear lead.

That sounds like the same marketing stuff that I've heard for previous versions. IMO - it is unlikely that we can get better than human driving performance (which is what will be necessary for the public to accept this) without advanced sensors (e.g. LIDAR). To my knowledge, despite the marketing name of FSD, Tesla requires that you remain attentive to the road, which is really a recipe for disaster.

But we'll see. I can believe they have a clear lead over other car manufacturers, because the rest of them are basically not touching this with a ten foot pole.

I'd dispute even first among equals. Claude Sonnet 3.5 is nearly always my go-to nowadays, falling back to GPT-4o when it's overloaded. (I will sometimes try the same prompt across all of Gemini/ChatGPT/Claude just to make sure they don't catch anything Claude would miss, and DeepSeek for code, but I've yet to be surprised.)

That said, I'd be surprised if Gemini and ChatGPT haven't caught up to/surpassed where Claude is today by end of year. But it's hard to imagine any of them building an enduring lead unless one has something up their sleeve.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats