444 comments

[ 2.8 ms ] story [ 351 ms ] thread
With all the other social networks trying to keep their data private because they all want to try their own AIs, it makes sense that OpenAI would want to have its own social network that wouldn't charge them for the data. I still doubt they actually launch it.
I've always thought that the social networks like X and BlueSky are sort of like the distributed consciousness of society. It is what society, as a whole / in aggregate, is currently thinking about and knowing its ebbs and flows and what it responds to are important if you want to have up to date AI.

So yeah, AI integrated with a popular social network is valuable.

Social networks tend to reflect the character of their founders. Do you really want to see what Sam Altman can do?
> Social networks tend to reflect the character of their founders.

I would say "owners" rather than "founders", but I agree with you. I think Sam Altman's couldn't be worse than Elon Musk's X, no?

Both are founders of a so-called non-profit and are suing each other. Their legal arguments are public at this point. By reading them, one may understand that it's hard to choose between 'yes' and 'no' as an answer. Maybe, we could request and take into account the opinion of what they 'created' that might outlast them and their conflict, namely AI.
I don't use X neither. Looks like it won't be around for much longer anyway, except as American Pravda (even though "Truth" Social already exists).
>One idea behind the OpenAI social prototype, we’ve heard, is to have AI help people share better content. “The Grok integration with X has made everyone jealous,” says someone working at another big AI lab. “Especially how people create viral tweets by getting it to say something stupid.”

This would be a decent PR stunt, but would such a platform offer anything of value?

It might be more valuable to set AI to the task of making the most human social platform out there. Right now, Facebook, TikTok, Reddit, etc. are all rife with bots, spam, and generative AI junk. Finding good content in this sea of noise is becoming increasingly difficult. A social media platform that uses AI to filter out spam, bots, and other AI with the goal of making human content easy to access might really catch on. Set a thief to catch thieves.

Who are we kidding. It's going to be Will Smith eating spaghetti all the way down.

No, nothing of value. If you ever want to lose faith in the future of humanity search "@grok" on Twitter and look at all the interactions people have with it. Just total infantilism, people needing tl;drs spoon-fed to them, needing summarization and one-word answers because they don't want to read, arguing with it or whining to Musk if they don't get the answer they want to confirm what they already believe.
the worst is like a dozen people in the replies to a post asking Grok the exact same obvious follow-up question. Somehow, having access to an LLM has completely annihilated these commenters' ability to scroll down 50 pixels.
I bookmarked this example where it is confidently incorrect about a movie frame/screenshot:

https://x.com/Pee159604/status/1909445730697462080

Your example doesn't appear to contain a reply from grok, only a question.
It does, you just can't see it without logging in because Twitter is shit now.

https://xcancel.com/Pee159604/status/1909445730697462080

I was logged in but it wasn't showing, but it's showing now.

> confidently incorrect

I disagree. Grok had a crack and got it wrong. LLMs get things wrong sometimes.

Besides, it said "likely from Species", which is guesswork. The original post is garbage. "Chimp the fuck out"... I don't even know what that means, so Grok didn't have much to go on by analysing the "green text".

(comment deleted)
> people needing tl;drs spoon-fed to them, needing summarization and one-word answers because they don't want to read

It's bad that this need exists. However, introducing this feature did not create the need. And if this need exists, fulfilling it is still better, because otherwise these kind of people wouldn't get this information at all.

This is worse because the AI slop is full of hallucinations which they will now confidently parrot. No way in hell does this type of person verify or even think critically about what the LLMs tell them. No information is better than bad information. Less information while practicing the ability to critically use it is better than bad information in excess.
Do you have examples of recent models hallucinating when asked to summarize a text?
All decent people I know have deleted their Twitter accounts - the kind of people you now see on twitter in the mentions are... not good people.
> needing summarization

Before we get too excited with disparaging those seeking summaries, it's common for people of all levels to want summary information. It doesn't mean they want everything summarized or are bad people.

I'm not particularly interested in "tariffs, what are they good for, what's the history and examples good or bad"... so I asked for a summary from grok. It gave me a decent summary. Concise and structured. I asked a few follow-ups, then went on with my life knowing a little more than nothing about tariffs. A win for summarized information.

"@gork explain this tweet"
You also can get Grok to fact check bullshit by tagging @grok and asking it a question about a post. Unfortunately this is not realtime as it can sometimes take up to an hour to respond, but I've found it to be pretty level headed in its responses. I use this feature often.
True. I see that too. It's a good addition to community notes. It can correctly evaluate "partially true" posts and those lacking details, so it's great at spotting cherry-picked information.
People sort of expect it to 100% subscribe to the right wing dogma now, but Elon apparently wasn't joking about it being "truth seeking". It seems pretty impartial to me, on some topics even "woke".
Why would AI be any better at filtering out spam than developers have so far been with ML?

The only way to avoid spam is to actually make a social network for humans, and the only way to do so is to verify each account belongs to a single human. The only way I've found that this can be done is by using passports[0].

0 - https://onlyhumanhub.com

How do you handle binationals who might not have the same details (or even name) on each of their passports?
You can always get around identification requirements, for example by purchasing a fake passport in this case. The idea is to increase the cost/friction of doing so as much as possible.

A fake ID is a lot harder to get your hands on than a new email, burner phone, etc.

1 passport = one human

Yes, this does mean that dual nationals can have two separate human accounts. But it’s still better than an infinite number of accounts, which is the case for social networks right now.

So you have to just trust them to permanently delete the data after verifying you?
That's interesting. Is there a social network where you can only connect with people you meet in real life?
(Stretching a definition of social network.)

Not strictly but Debian, where member inclusion is done through an in person chain of trust process so you have clusters of people who know each other offline as a basis.

Also, most WhatsApp contacts have been exchanged IRL, I presume.

I've never been comfortable with this idea that people should use their real identity online. Sure they can if they choose to, but IMO it absolutely shouldn't be required or expected.

The idea that I would give a copy of my passport to a social media company just to sign up, and that the social media company has access to verify the validity of the passport with the issuing government, just feels very wrong to me.

I agree. That’s why onlyhumanhub doesn’t expect you to share your name. The passport verification is there to ensure you are a unique human, but the name of that human is not stored.

I’m perfectly happy talking to someone without knowing their real name. I just want to be more confident that they’re a unique human, and not just another sock puppet account run by some Russian agent (or evil corporation) trying to change people’s beliefs at scale.

But how do you actually trust them?

There's almost no time investment in building onlyhumanhub. It's only a few months old (based on the copyright), have effectively a text-only homepage, and account creation which I assume allows you to upload photos of your passport and link your existing social media profiles.

There are so many ways that could go wrong, from this being a phishing attack to this being a well intended project that happens to create a database linking passport IDs to all of a person's social media accounts.

The idea that they may eventually offer a social media platform that doesn't require public use of your real identity is all well and good, but they're still a honeypot for doxing.

Well, I built the site.

I agree that trust is a problem. I try to be as transparent as possible around how your passport data is used and what is stored in the database. Far more than what ordinary banks/trading apps say when they ask you for a passport.

Hah, well sorry for my confusion there! I didn't realize it was yours so that definitely clears up why you'd trust it.

While I have you here I am curious what's evolved in validating passports? Is it as simple as a unified API run by some service to validate, or an API per country?

An interesting use for AI right now would be using it as a gatekeeping filter, selecting social media for quality based on customisable definitions of quality.

Using it as a filter instead of a generator would provide information about which content has real social value, which content doesn't, and what the many dimensions of "value" are.

The current maximalist "Use AI to generate as much as possible" trend is the opposite of social intelligence.

It's a nice idea in principle, but would probably immediately become a way by the admins to promote some views and discourage others with the excuse of some opinions being of lower quality.
That's what moderation is and is perfectly fine. Dang does that here on HN and for good reason.
It's not moderation, for one thing it never will be used with moderation.
I think that's right. Twitter without ads, showing you content you _do_ want to see using some embeddings magic, with decent blocking mechanisms, and not being run as a personal mouthpiece by the world's most unpopular man ... certainly not the worst idea.
> This would be a decent PR stunt, but would such a platform offer anything of value?

Like all those start-ups that are on the 'mission' to save the world with an app. Not sure if it is PR for users or VCs.

Sam's last social media project included users verifying their humanity, so there is hope that something like that slips into the new platform.
A 4chan but images can be prompt generated? Makes sense. Everything's going back to early 2000s, it seems.
I would try to make a platform like Deviantart or Tumblr except OpenAI pays you to make good content that the AI is trained on.
Nice in theory but don’t know how practical it is to actually do.

How do you define “good”? Theres obvious examples at the extremes but a chasm of ambiguity between them.

How do you compute value? If an AI takes 200 million images to train, wait let me write that out to get a better sense of the number:

200,000,000

Then what is the value of 1 image to it? Is it worth the 3 hours of human labour time put into creating it? Is it worth 1 hour of human labour time? Even at minimum wage? No, right?

You really think an OpenAI-sponsored social network is going to attract people who create and share original content?
How do you stop people gaming this by feeding it the output of other AIs?

(not to mention defining "good")

This is just part of the ongoing feud between Sama and Musk.
(comment deleted)
They should use their resources to make OpenAI good at coding.
Sam got a jawline lift, anyone noticed?
Did he? Flipping back and forth between old vs. new photos of him, his facial structure seems roughly the same.
Yes, I've been cataloging the mewing and lookmaxxing progress of hundreds of public figures
It'd be cool to see Google+ resurrected with OpenAI branding. Google+ was actually a pretty well designed social network
Not well designed enough to live, though.
Not well designed to live under Google*

Tumblr is still alive. LiveJournal is still alive. Newgrounds is still alive and Flash doesn't even exist anymore.

It doesn't matter how well-designed it is if people aren't there. Social graph lock-in is the single biggest issue with any contender.
that would be cool, google+ was very unique and i was kinda sad google killed it off
what did you like about it?
I liked the UX. I liked Circles. There were other nice options that I can't remember but I thought Google+ was a big improvement over Facebook.
I don't believe it was well designed, it felt clunky to use, concepts weren't intuitive enough to understand after a few uses.

I tried to use it for a few months after release, always got frustrated to the point I didn't feel like reaching out to friends to be part of it.

The absurd annoyance of its marketing, pushing it into every nook and cranny of Google's products was the nail in the coffin. I'm starting to feel as annoyed by the push with Gemini, it just keeps popping up at annoying times when I want to do my work.

I think a social network is not necessarily a timeline-based product, but an LLM-native/enabled group chat can probably be a very interesting product. Remember, ChatGPT itself is already a chat.
What's a "LLM-native/enabled group chat"?
Telegram and slack bots are probably the best example so far. Bot gets added to a chat and can respond when mentioned in the group chat.
Gotcha, the NLP-enabled version of the good old IRC weatherbot.

For a moment I had a funnier mental image of a chat app with an input field that treats every input as a prompt, and everyone's chatting through the veil of an LLM verbosity filter.

There might be something chat RPG-like there worth trying though ...

Yes, this. That's my bet if OpenAI follows through with social features.

Extend ChatGPT to allow multiple people / friends to interact with the bot and each other. Would be interesting UX challenge if they're able to pull it off. I frequently share chats from other platforms, but typically those platforms don't allow actual collaboration and instead clone the chat for the people I shared.

I am building this with a team currently and we are launching in a couple days.

Would love an alpha tester or two if anyone wants to test it.

My email/twitter is in my profile, shoot me a message and I will be in touch.

Yeah, the dream is the AI facilitating "organic" human connection
What else are they going to spend billions on to turn a profit?
I don't know, but a weight bench goes under $200 and Sam needs some chest gains fast.
Aren't they unprofitable enough already?
Is this just a data play? Need more data. Start a social network. Own said data.
I think its more likely that they're desperate to find a profitable business model.
Seems telling that an org had arguably the leading AI, as the planet knows it at least, and still can't exist without putting ads in front of eyes. So much for the hype.
Honestly I wonder if it’s because Altman loves X and is threatened by Grok
Is making yet another twitter clone really the way to build a path towards super-intelligence? A worthy use of the organization's talent?
Collecting millions of people’s thoughts and interactions with each other IS probably on the path to better LLMs at least.
I'd love for my agents to be created in the image of humanity's best side, its interactions on social media.

Perhaps then we can all let LLMs take care of tweeting outrage for us, and go outside to find each other rolling around on the grass.

Another twitter clone will help the decline of human intelligence, the dumber humans are the smarter the Ai appears.
Imagine that, a social network where all of the participants are bots.
My guess ... it's probably less of a "social network" and more of a "they are trying to build a destination (portal) where users go to daily".

E.g. old days of Yahoo (portal)

They just want the next wave of Ghibli meme clicks to go to them, really.

This will be built on the existing thread+share infra ChatGPT already has, and just allow profiles to cross-post into conversations, with UI and features more geared toward remixing each other's images.

That was my thought: a meme-sharing platform.
The answer seems more obvious to me. They dont even care if its competitive or scales too much. xAI has a crazy data advantage firehousing Twitter, llama FB/IG and CGPT just has, well, the internet.

Id hope they have some clever scheme to acquire users, but ultimately they want the data/

I actually would love this. I hate having to go to another website to share some thoughts I had using tools in a platform.

I miss the days when experiences would actually choose to integrate other platforms into their experiences, yes I was sort of a fan of the FB/Google share button and Twitter side feed (not the tracking bits though).

I wasn't a fan of LLM and the whole chat experience a few years ago, I'm a very mild convert now with the latest models and I'm getting some nominal benefit, so I would love to have some kind of shared chat session to brain storm, e.g. on a platform better than Figma.

The one integration of AI that I think is actually neat is Teams + AI Note taking. It's still a hit or miss a lot of the time, but it at least saves and notes something important 30% of the time.

Collaboration enhancements would be a wonderful outcome in place of AGI.

Sounds like they are thinking about instagram, which originated as a phone app to apply filters to a camera and share with friends (like texting or emailing them or sending them a link to a hosted page), and evolved into a social network. Their new image generation feature has enough people organically sharing content that they probably are thinking about hosting that content on pages, then adding permissions + follow features to all of their existing users' accounts.

honestly it's not a terrible idea. it may be a distraction from their core purpose, but it's probably something they can test and learn from within a ~90 day cycle.

AI bots already make up a significant percentage of users on most social networks. Might as well just take the mask off completely--soon, we'll all be having conversations (arguments, most likely) with 'users' with no real human anywhere near them.
I've been saying for a while that the next innovation beyond TikTok, Instagram, and YouTube is to get rid of human creators entirely. Just have a 100% AI-generated slop-feed tailor made for the user.

There's already a ton of AI slop on those platforms, so we're like half way there, but what I mean is eliminating the entire idea of humans submitting content. Just never-ending hypnotic slop guided by engagement maximizing algorithms.

An idea which sounds horrifying but would probably be pretty popular: a Facebook like feed where all of your “friends” are bots and give you instant gratification, praise, and support no matter what you post. Solves the network effect because it scales from zero.
This is Altman increasing the mass of the investment black hole that OpenAI is.
Sam Altman is retaliating against Musk for Grok and Musk's lawsuit against OpenAI, trying to ride the wave of anti-Musk political heat, and figure out a way to pull in more training data due to copyright troubles.

If they launch, expect a big splash with many claiming it is the X-killer (i.e. the same people that claimed the same of Mastadon, Threads, and Bluesky), especially around here at HN, and then nobody will talk about it anymore after a few months.

Here's how to kill Twitter and Bluesky AND Mastodon:

1: use an LLM to extract the text from memes and relatable comics.

2: use an LLM to extract the transcriptions of videos.

3: use an LLM to censor all political speech.

OpenAI, I believe in you. You can do it. Save the Internet.

If you can clean my FYP of current events I'll join your social media before you can ask a GPT how to get more users.

Just use an LLM to verify only humans are on the social network and no bots and you win
> 3: use an LLM to censor all political speech.

And who gets to decide what is political? Are human rights political? Is a trans person merely existing political? Is calling for genocide political?

The LLM decides it. That's what the AI is for.

There is a lot of stuff on the Internet, so I think the AI can just censor 80% of it and we're still going to have enough to have a social media.

Really saying "please, let the robot run humanity", huh.
Let the robot run ONE SITE. You still have Twitter, Mastodon, Bluesky, Facebook, Reddit, etc., if you want a non-stop stream of political content on your feed.
LLMs are guessing machines, they don’t “decide” anything. It would be decided by the people programming it and putting in alignment guardrails.
wonder if the LLM would censor this post
If it works the way I want it absolutely should. Mentioning politics is political content.
Not-exactly-devil's-advocate: you're trying to sort content by quality. That's elitist. Also, those those filtered contents are worth more. You can't have only premium contents.

Someone should do it anyway and make it dominant anyway ASAP.

Feels like a natural next step, honestly. If they already have users generating tons of content via ChatGPT, hosting it natively and adding light social features might just be a way to keep people engaged and coming back. Not sure if it's meant to compete with Twitter/Instagram, or just quietly become another daily habit for users
This would be a natural step if it were 2010. In 2025, it sounds like a lack of imagination to me.
Perhaps, but currently OpenAI is stuck sharecropping with existing social networks – producing the content but not deriving the value. It is hard to move grander visions forward when you don't own the land.