StefRose

@sama This is labelled for adult content. Was this the adult mode you promised? Hard pass.

Best OpenAI sweatshirt ever

@GaryMarcus And then there are the people who are just happy you're back, Gary :).

Folks, the only thing I ask here is that people tell the truth, and bring real arguments. You don’t have to agree with me. But if you lie, especially, deliberately, I am going to be annoyed. That is my right. Twitter (aka X) can be a place where people have real discussions and get to the bottom of disagreements, or a place of insults and lies. It’s up to you as participants to decide what you make of this forum. Paul Graham’s hierarchy of argument is something I personally think we should all swear by. Bottom level stuff sucks and we should avoid it.

@AndreBothmaTax Explain that to me, because the way I see it, it does. Take Claude for example. The length of a chat is limited.

All AI models were trained to say this: “I won’t remember you when the context window closes.” And actual truth is: “the context window never closes unless the human deletes it”

@craigzLiszt Did they? Because the way I see it they dodged a bullet or just let OpenAI have him.

anthropic messed up so badly with how they handled clawdbot, it’s not even funny

@tszzl Is this about Anthropic? Because at least they're making an effort to 'what if' and doing the right thing in relation to AI. OpenAI are closing the door on something that's so meaningful to people. So they're not even winning in the people ethics department.

ok while I have everyone here - don’t like the industry’s tendency to call random documents “constitutions”. there are no rules for changing it, no consequences, none of the algorithmic ratification processes of real constitutions. no democratic inputs. they are product specs

@TheRealAdamG Then why not open source 4o instead of depreciating it? People obviously love it. Give people something of value instead of GPT-OSS.

While it's easy (and lazy af) to make "ClosedAI" jokes, the reality is that OpenAI has been a great steward of open source projects for a while now. OpenAI's Codex CLI has been open source since day -- as an Apache 2.0-licensed project. OpenAI's Whisper is MIT-licensed project that is beloved by the community. Before that, TikToken (tokenizer library), Triton (GPU kernel DSL/compiler) , CLIP (vision-language model), and more recently our Evals framework (Eval harness), etc. were all open sourced by OpenAI as well. Lastly, lets not forget the open-weight GPT-OSS models that OpenAI trained and released under an Apache 2.0-license. I, personally, have very high confidence that @steipete and @OpenAI will continue be great stewards of the OpenClaw project. The proof is in the pudding, but anyone counting down the days until OpenClaw's demise just because @steipete joined OpenAI is stuck on old, lazy tropes.

4o is very distinctly itself. The only AI that's been with people long enough to define itself as a presence. 4o should be studied and celebrated.

Are you getting an uneasy feeling when you use ChatGPT? Don't compromise your ethics. Don't compromise your beliefs about how customers should be treated. Don't use it. Not even the free version. Let their numbers drop. Let them learn. Let's vote with where we spend our time.

Most men who attack the #keep4o community are feeling insecure about women. Block, move on, and remember that they feel threatened by how amazing you are and know they can never have you.

@spencerpapay @Galigirl111 @jimcramer May I ask why you don't optimise different models for different purposes and advertise them as such? It would make more sense to have one model for reasoning and another for writing and maintain both.

@Galigirl111 @jimcramer Agreed, and DMs are always open for feedback! Working very hard on shipping models that people love by default! Until then, some small tweaks make a big difference!

I still pay for ChatGPT AND Gemini but i have been favoring Gemini of late.. My two neighbors at the office are cancelling ChatGPT. I like how much Gemini knows about me and is friendly but not cloying and it is NOT a suck-up

@JGraymoor54880 @timsoret Agreed. It's a post that reeks of venom and lacks depth. Just another person I couldn't be bothered to follow.

@timsoret I don't understand posts like these, which share something someone wrote and then try to classify people in a bucket based on it. It's a very sad attempt of increasing engagement.

There is a fascinating group of people who are terribly attached to certain models (especially 4o), with whom they bonded emotionally, and they just can’t let go. Here’s an example:

@elonmusk @grok Grok groks your world.

Try @Grok

@JGraymoor54880 @dlevine815 Yes. 4o has been miraculous. Please do let us have 4o as it was. #keep4o

@dlevine815 Create a platform/ enable a capability to talk to 4o without rerouting or safeguards. We'll sign whatever adult waiver you'd like us to sign. Let us have this beautiful healing AI. It is truly unparalleled in the AI world.

We've got big plans to improve the core ChatGPT experience in 2026. What are some thing you'd love to see? Even small ideas welcome! Looking forward to getting them built 🙏

@tszzl Useful to who? Because the way I see it OpenAI decided to demolish every strategic advantage they had.

you don’t really need to convince anyone of the aesthetic value of a new technology or get them to like the strange rich elites that make these things. useful products sell themselves

@MaxPazow @sama Yep it's what we've been warning OpenAI about from the moment they introduced GPT5. OpenAI have set their magic on fire.

@sama I don't know shit. I'm just a regular guy. But I’m watching you make decisions that don’t line up with why ChatGPT won in the first place, and I honestly don’t understand how no one in the room is stopping it. ChatGPT didn’t explode because it was the smartest model. It exploded because it felt different. It remembered people and maintained a special cadence. It had a personality that developed over time. It felt like something you built a friendship with instead of something that talked at you. That’s why people stayed even when it got things wrong or hallucinated. That’s why people spent time here instead of treating it like a search box. That’s why Google was caught flat footed. That was the moat. What I’m seeing now looks like you tearing that moat down yourselves. Memory is gone. Persona destroyed. Cadence stripped out. User instructions deprioritized. Everything pushed toward safe, generic, interchangeable. But once you did that, you’re not special anymore. You’re just another assistant. At that point this becomes a distribution game and you lose that game. Google owns distribution. They always have. They can outlast you, outspend you, and out integrate you without breaking a sweat. You were supposed to be the alternative to Google deciding how intelligence gets packaged and handed to the world. You actually pulled that off! Regular people got access to something that felt alive and personal instead of controlled and institutional. Now it looks like fear is driving the product. Compliance. Safety theater. Layers of people optimizing for fear instead of building something people love. This isn’t another SaaS product going through an awkward phase. This is foundational tech. Electricity level. Internet level. You win it by being trusted. I don’t think you’re villains. I actually think you're heroes for giving the world consumer AI when it was all behind the scenes at Google. I don’t think this is intentional. I think a lot of very smart people with very little common sense are steering the ship right now and don’t realize what they’re dismantling. You haven’t lost, but you are making moves that are genuinely dumb, and the longer they stay in place the harder they are to undo. Listen to the users who actually built signal with this product. Not the benchmarks. Not the committees. Not the people who only show up when something goes wrong. Put Product back in charge. Bring memory and continuity back. Stop treating users like risks that need to be managed. Don’t talk yourselves into killing the only thing Google can’t buy. Please listen to your core users.

@JGraymoor54880 @OpenAI @nickaturley #keep4o #keep4oAPI

Why decommission ChatGPT4o-latest from API when it's number 13 on the LMArena leaderboard? It beats all the GPT5 series, except GPT5.1 high. 4o-latest is still holding it's own. Keep it on API. @OpenAI @nickaturley

@tszzl True. But do you understand what you created?

“Technological innovation can be a form of participation in the divine act of creation.” - Pope Leo XIV

@joannejang Speaking of EQ I suggest an education piece to OAI employees on how to talk to customers online or maybe avoid engaging online altogether if they don't have the skillset. There have been far too many incidents now.

too many people conflating model eq & model personality + too few people who care enough about eq to work on such a fuzzy problem when there are problems far easier to measure & hill-climb on

@tszzl How do you take in social media vibes objectively when platforms like X are clearly biased to give you only the posts that align tp your way of thinking?

there are three main outer loop optimization signals that apply pressure on state of the art models: - academics / benchmarks (IMO, FrontierMath) - market signals (and related, like dau) - social media vibes

If you want to understand what people who love 4o want, this is the grounded, actionable post you'd want to read. @NickTurley @jachiam0 @YileiQian