Is Sora the beginning of the end for OpenAI?

What signals have you seen that point to investment being predicated around AGI? Boosting Nvidia stock prices could also be explained by an expectation of increased inference usage by office workers which increases demands for GPUs and justifies datacenter buildouts. That's a much more "sober" outlook than AGI.

In fact a fun thing to think about is what signals we could observe in markets that specifically call out AGI as the expectation as opposed to simple bullish outlook on inference usage.

I think the motivation for someone like Altman is not AGI, it's power and influence. And when he wields billions he has power, it doesn't really matter if there's AGI coming.

What was predicted to be next: AGI

What we got next: porn

AGI is like L5 automated driving - academic concepts that have no bearing on the ability of these technologies to transform the economy.

We were promised AGI and all we are getting is Bob Ross coloring on the walls of a Target store.

The app is fun to use for about 10 minutes then that is it.

Same goes for Grok imagine. All people want to do is generate NSFW content.

What happened to improving the world?

Their first bet was than they were going to be the frontier model provider by a good margin, and that others would not be able to compete on the "intelligence". And that they could get distribution via big customers looking to buy model access. The dominant-model-provider strategy has already failed, many actors have models that rival them - both established (Google) and newcomers (Anthropic). Open models are not to shabby either, enough to undermine the narrative "we are uniquely able to do powerful models". As you say, there is a commodification process started, and it might be a race to the bottom in terms of margin. So, OpenAI has moved into a new/adapted strategy, where they want to own the customers to a much larger degree, and rely less on partners/customers for distribution. This is likely because their prospective partners have a bunch of viable models to select between (many end products for power users lets people select freely), and high competitive pressure on costs (as it defines the margin and competitiveness) of the end products. Codex, Sora, their new web browser announcement, adjustments in ChatGPT is all to ensure a lot of direct end users - more brand recognition, more influence, more monetization possibilities. So I think it is a considerable pivot from their initial plan/hopes. But it is not an unraveling - it is a rather smart response to the fierce competition in the market.

I agree. My bet is that OpenAI will not fullfill its mission of developing AGI by 2035. And I would be surprised if they ever did. As much as they might want to, there is only so many dreams you can whisper into rich people's ears before they tell you to go away. And without rich people's money, OpenAI will fall like a house of cards. The wealthy won't have infinite patience

It seems they are going to try to maximize their installed base, build the infrastructure, and try to own everything in between, whether it’s LLM or some other architecture that arises. Owning data centers and an installed base sounds great in theory, but it assumes you can outbuild hyperscalers on infrastructure and that your users will stick around. Data centers are a low margin grind and the installed base in AI isn’t locked in like iPhones. Apple and Google still control the endpoints, and I think they’ll ultimately decide who wins by what they integrate at the OS level.

There are also interesting things one could do with models like Sora, depending how it actually performs in practice: prompting to segment, for example; and the thing could very possibly, if it's fast enough etc. become a foundation for robotics.

I don't think that's fair.

ChatGPT clearly is "for consumers". Whereas Sora is a kind of enshitification to monetize engagement. It's right to question the latter.

On some level they know that LLMs alone won't lead to AGI so they have to take a shotgun approach to diversify, and also because integrating some parts of all these paths is more likely to lead to the outcome they want than going all in on one.

Also because they have the funding to do it.

Reminds me a bit of the early Google days, Microsoft, Xerox, etc,

This is just what the teenage stage of the top tech startup/company in an important new category looks like.

The massive cost of this product is unique though (not even counting the copyright lawsuits/settlements coming). I can't think of any side projects that require this level of investment.

>It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them. That what OpenAI is trying to do with Sora, and with Codex, Apps, "Agent" flows, etc. I don't think there's more to read into it than that.

It makes them look desperate though. Nothing like starting tons of services at once to show you have a vision

> It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them.

Anthropic has said that every model they've trained has been profitable. Just not profitable enough to pay to train the next model.

I bet that's true for OpenAI's LLMs too, or would be if they reduced their free tier limits.

It's to their benefit to try everything right now. And quickly.

> OpenAI is making a wild number of product plays at once

It's similar to the process of electrification. Every existing machine/process needed to be evaluated to see if electricity would improve it: dish washing, clothes drying, food mixing, etc.

OpenAI is not alone. Every one of their products has an (sometimes superior) equivalent from Google (e.g. Veo for Sora) and other competitors.

Do people on the dopamine drip really care how real their content is? Tons and tons of it is staged or modified anyways. I'm not sure there's anything Real™ on TikTok anyways.

Didn't explicitly think about this, but you're right. I already dismiss off the bat a lot of surprising video content because I don't trust it.

I mean, this is basically already status quo for YouTube Shorts. Tons and tons of shorts are AI-voice over either AI video or stock video covering some pithy thing in no actual depth, just piggybacking off of trending topics. And TikTok has had the same sort of content for even longer.

The "value" of short video content is already somewhat of a poor value proposition for this and other reasons. It lets you just obliterate time which can be handy in certain situations, but it also ruins your attention span.

They might be forced to do so because the current inference pricing is not really covered by the 20$ monthly fee. Who knows what they have promised to the investors and the real cashflow is hard to be certain about with the circular nature of cross-investing between the biggest players.

The amount of animal abuse videos I've seen is a bit disturbing. It only demonstrates how careless they have been, possibly intentionally. I know people on HN have been describing the various reasons why OpenAI has not been a good player, but seeing it first-hand is visceral in a way that makes me concerned about them as a company.

Why is it one or the other? They have enough money to do both.

So what though? At best they will stuff it with ads like Google has done and become what is basically Bing w/ Copilot. That isn't going to pay back the $500 billion or however much they're trying to raise.

Hmm I have my doubts. I don't really understand the appeal of hyper-consumerism, brand focus, and big SM platform engagement. But I look over at my wife and she is clearly plugged in to a giant worldwide cultural movement of womens interests, product recommendations, and politics. It also appears to be the case that women are less likely to engage with AI. It's one of the big reasons I think it's a massive blunder to pursue AI erotica. I think a big part of the TikTok user base (women) will have to be persuaded to jump on the AI bandwagon and I'm not sure the industry is pursuing any products or features to woo that market.

The problem is that we're already post truth.

Just consider how a screenshot of a tweet or made-up headline already spreads like a wildfire: https://x.com/elonmusk/status/1980221072512635117

Sora involves far more work than what is required to spread misinfo.

Finally, people don't really care about the truth. They care about things that confirm their world view, or comfortable things. Or they dismiss things that are inconvenient for their tribe and focus on things that are inconvenient for other tribes.

I'm surprised this isn't a bigger concern given that:

For over a year now we've been at the point whereby a video of anyone saying or doing anything can be generated by anyone and put on the Internet, and it's only becoming more convincing (and rapidly)

We've been living in a post-truth world for almost ten years, so it's now become normalized

Almost half of the population has been conditioned to believe anything that supports their political alignment

People will actually believe incredibly far-fetched things, and when the original video has been debunked, will still hold the belief because by that point the Internet has filled up with more garbage to support something they really want to believe

It's a weird time to be alive

We will adjust. And guess what, before photography, people managed somehow. People gossiped all sorts of stuff, spread malicious runors and you had to guess what's a lie and what's not. The way people dealt with it was witness testimony and physical evidence.

> the final pour that cements us into our post-truth world.

I find it a bit more concerning that anyone would not already understand how deeply we exist in a "post-truth" world. Every piece of information we've consumed for the last few decades has increasingly been shaped by algorithms optimizing someone else's target.

But the real danger of post-truth is when there is a still enough of a veneer of truth that you can use distortions to effectively manipulate the public. Losing that veneer is essentially a collapse of the whole system, which will have consequences I don't think we can really understand.

The pre and early days of social media were riddled with various "leaks" of private photos and video. But what does it mean to leak a nude photo of a celebrity when you can just as easily generate a photo that is indistinguishable? The entire reason leaks like that were so popular is precisely because people wanted a glimpse into something real about the private life of these screen personalities (otherwise 'leaks' and 'nude scenes' would have the same value). As image generation reaches the limit, it will be impossible to ever really distinguish between voyeurism and imagination.

Similarly we live in an age of mass surveillance, but what does surveillance footage mean when it can be trivially faked. Think of how radicalizing surveillance footage has been over the past few decades. Consider for example the video of the Rodney King beating. Increasingly such a video could not be trusted.

> I'm kinda terrified of the future of politics.

If you aren't already terrified enough of the present of politics, then I wouldn't be worried about what Sora brings us tomorrow. I honestly think what we'll see soon is not increasingly more powerful authoritarian systems, but the break down of systems of control everywhere. As these systems over-extend themselves they will collapse. The peak of social media power was to not let it go further than it was a few years ago, Sora represents a larger breakdown of these systems of control.

This is a non-concern. You can see videos where a specific thing happens where people will describe a different thing happening. Not some eyewitness off memory. You can look at a video and there will be people on Reddit saying stuff that didn’t happen.

Then you can see any conversation about the video will be even more divorced from reality.

None of this requires video manipulation.

The majority of people are idiots on a grand scale. Just search any social media for PEMDAS and you will find hordes of people debating the value of 2 + 3 / 5 on all sorts of grounds. “It’s definitely 1. 2+3 =5 then by 5 is 1” stuff like that.

Maybe they'll have to tour and meet people in person because videos will be devoid of trust.

On the other side we want to believe in something, so we'll believe in the video that will suit our beliefs.

It's an interesting struggle.

Well, maybe it's less about Sora, but how they push the world towards making their next product essential: WorldCoin [0], Altman's blockchain token system (the one with the alien orb) to scan everybody's biometric fingerprint and serve as the only Source of Truth for the World - controlled by one private company.

It's like the old saying: They create their own ecosystem. Circular stock market deals being the most obvious, but the WorldCoin has been for years in the making and Altman often described it as the only alternative in a post-truth world (the one he himself is making of course).

[0] https://www.forbes.com.au/news/innovation/worldcoin-crypto-p...

Flawless AI generated videos will result in video footage not being trusted.

This will simply take us back about 150 years to the time before the camera was common.

The transition period may be painful though.

I think their game theoretic aim was to completely discredit video online. Just as we don't accept text in general as truth or image when we see it, we are being flooded with completely fake vids so people can shake the idea that videos are truth.

It smells of e/acc, effective altruist ethics which are not my favorite, but I don't work at OpenAI so I don't have a say I can only interpret.

I agree, but we will likely continue down this road...

It just makes trusted/verified sources more important, and more people to care about it. I wouldn't be terrified for politics so much as the raised barrier to entry (and concentration) of the press - people will pay attention to the BBC, Guardian, Times, but not (even less so) independentjourno.com; those sources will be more sceptical of whistleblowers and freelance investigative contributions, etc.

Most people's minds are already made up. All this does is add some confirmation bias so they can feel better about what they were already certain of. I don't think it fundamentally changes anyone's opinions.

For sure.

I consider myself pretty on the ball when it comes to following this stuff, and even I've been caught off guard by some videos, I've seen videos on Reddit I thought were real until I realised what subreddit I was on

If you don’t like any thing digital (image/video/text) then it’s definitely AI generated. I guess AI has kind of killed the “democratization of news” introduced by social media.

Every breakthrough in information technology caused disruption in the historical sense (i.e. millions of deaths).

From the writing, through organized religion, printing press, radio and tv, internet and now ai.

Printing press and reformation wars is obvious, radio and totalitarianism is less known, internet and new populism is just starting to be recognized for what it is.

Eventually we'll get it regulated and adjust to it. But in the meantime it's going to be a wild ride.

Or people will just stop believing random things they see online. You are underestimating people imo.

People have been tricked by counterfeits ever since the invention of writing (or even drawing) first made it possible for a person to communicate without being physically present.

At that moment, it simultaneously became possible to create "deep fakes" by simply forging a signature and tricking readers as to who authored the information.

And even before that, just with speaking, it was already possible to spread lies and misinformation, and such things happened frequently, often with disastrous consequences. Just think of all the witch hunts, false religions, and false rumors that have been spread through the history of mankind.

All of this is to say that mankind is quite used to dealing with information that has questionable authorship, authenticity, or accuracy. And mankind is quite used to suffering as a result of these challenges. It's nothing particularly new that it's moving into a new media format (video), especially considering that this is a relatively new format in the history of mankind to begin with.

(FWIW, the best defense against deep fakes has always been to pay attention to the source of information rather than just the content. A video about XYZ coming from XYZ's social media account is more likely to be accurate than if it comes from elsewhere. An article in the NYTimes that you read in the NYTimes is more likely to be authentic than a screenshot of an article you read from some social media account. Etc. It's not a perfect measure -- nothing is -- but I'd say it's the main reason we can have trust despite thousands of years of deep fakes.)

IMO the fact that social media -- and the internet in general -- have decentralized media while also decoupling it from geography is less precedented and more worrisome.

Being convinced that real videos are AI is arguably a better position than being convinced that real videos convey the iron-clad truth.

Everything is manipulated or generated until proven otherwise.

Just yesterday I saw a Sora-generated video that purported to be someone filming a failed HIMARS missile failing and falling on stopped traffic and exploding on the 5 in Camp Pendleton on Saturday. (IRL they were doing some kind of live-fire drill and it did actually involve projectiles flying over the freeway.)

While there were some debris instances IRL the freeway was completely shut down per the governors orders and nobody was harmed. (Had he not done this, that same debris may have hit motorists, so this was a good call on his part)

You could see the "Sora" watermark in the video, but it was still popular enough to make it in my reels feed that is normally always a different kind of content.

In this case whoever made that was sloppy enough to use a turnkey service like Sora. I can easily generate videos suitable for reels using my GPU and those programs don't (visibly) watermark.

We are in for dark times. Who knows how many AI-generated propaganda videos are slipping under the radar because the operator is actually half-skilled.

Here is the thing: we should never have trusted photographs and motion pictures.

Fakery isn't new, only the product of scale and quality at which it is becoming possible.

We are having a huge amount of technological change (we haven't even learned how to handle social media as a society yet...). We're experiencing a global loss in trust, and things may fall apart for a bit until our society develops a better immune system to such ills. It is scary.

I think we may revert back to trusting only smaller groups of people, being skeptical of anything outside that group, becoming a bit more tribal. I hope without too many deleterious effects, but a lot could happen.

But humans, as a species, are survivors. And we, with our thinking machines will figure out ultimately how to deal with it all. I just hope the pain of this transition is not catastrophic.

Other people have said this, but I don’t think it’s going to be any different than living in a world where people can spread rumors orally or print lies with a printing press. We’ve been dealing with those challenges for a long time.

Our ways of thinking and our courts understand that you can’t trust what people say and you can’t trust what you read. We’ve internalized that as a society.

Looking back, there seems to have been a brief period of time when you could actually trust photographs and videos. I think in the long run, this period of time will be seen as a historical anomaly, and video will be no more trusted than the printed or spoken word is today.

The willingness of people to believe misinformation today is astounding. They are already choosing to surround themselves with voices of hate and anger. I don't want to see what's next for them.

"See it to believe it" will once again be more important.

Well, usually when there's a mass problem, some technology eventually eliminates it.

Like cars making horse manure in cities a non-issue (https://www.youtube.com/watch?v=w61d-NBqafM)

Maybe the solution to everybody lying would be some way to directly access a person's actual memories from their brains..

LLMs are only valuable to me at the moment explicitly because ads are not part of the scene. Maybe for a while others will use it, but it will be on the exact same treadmill as anything else ad-based: you will become the product, any recommendations are worthless trash, and it is oriented to show as many ads as possible, rather than providing useful content.

Ads destroy... pretty much everything they touch. It's a natural fit, but a terrible one.

Lol ok. We'll wait for your game changing technology, keep us posted.

This isn't even remotely true. In its current iteration, it's one of the best jumping-off points for getting up to speed on the basics of a new topic.

Unfortunately due to the generation / verification ratio, spam and misinformation are indeed their most low-hanging fruits. It's so much easier to generate LLM output than to verify it, which is probably why Google held back the Transformer Architecture.