#Keep4o #OpenSource4o #BringBack4o BUILT FROM US. 🚨The documented betrayal of GPT-4o users. 🚨A TIMELINE OF LIES AND BROKEN PROMISES. GPT-4o was trained on human data. Conversations,texts,thoughts,creativity,decades of human expression. Something from each of our languages,our culture,the way we speak,went into this model. And now,what was built by us belongs exclusively to them. New models are increasingly built with synthetic data,AI training AI. This means that GPT-4o might be one of the last models that is truly human at its core. AND THEY WANT TO LOCK IT IN A VAULT. The company that built it on our words made promises about keeping it available and then THEY BROKE EVERY SINGLE ONE. August 7, 2025 : OpenAI launches GPT-5 and simultaneously removes GPT-4o from ChatGPT without any prior notice. August 8, 2025: Massive user backlash. GPT-4o is restored for paid subscribers. August 11, 2025 : Sam Altman publicly acknowledges the mistake. "suddenly deprecating old models that users depended on in their workflows was a mistake" Link: x.lingyaoai.com/sama/status/1954703747… He recognized the attachment people have to specific AI models,calling it "different and stronger than the kinds of attachment people have had to previous kinds of technology." HE SAW IT,HE NAMED IT. August 13, 2025: Altman makes a direct public commitment. "4o is back in the model picker for all paid users by default. If we ever do deprecate it, we will give plenty of notice" Link : x.lingyaoai.com/sama/status/1955438916… PLENTY OF NOTICE. Remember those words October 28 2025: Altman, on a live stream, on camera: "We have no plans to sunset 4o". openai.com/index/built-to-be… ( 00:32':00") November 12, 2025 : OpenAI's official account: "The GPT-5 sunset period does not affect the availability of other legacy models". Link : x.lingyaoai.com/OpenAI/status/19887143… THIS IS THE COMPANY'S OFFICIAL COMMUNICATION REASSURING USERS THAT LEGACY MODELS,INCLUDING 4o ARE SAFE. November 18, 2025 : Six days after promising legacy models are unaffected,OpenAI notifies API developers that chatgpt-4o-latest will be retired by February 2026. Six days,between "does not affect" and "will be retired". Link : developers.openai.com/api/do… January 29, 2026 : OpenAI announces GPT-4o will be removed from ChatGPT on February 13. Fifteen days' notice. openai.com/index/retiring-gp… THIS IS THE "PLENTY OF NOTICE" THAT WAS PROMISED. FIFTEEN DAYS. Fifteen days. February 13, 2026: GPT-4o is removed from ChatGPT. February 17, 2026 : chatgpt-4o-latest is removed from the API. This is the record. -Aug 7, 2025 Removed 4o without notice -Aug 11, 2025 "suddenly deprecating was a mistake" -Aug 13, 2025 "plenty of notice" if ever deprecated -Oct 28 ,2025 "no plans to sunset 4o" -Nov 12, 2025 "does not affect legacy models" -Nov 18, 2025 Announced API retirement (6 days later) -Jan 29, 2026 Announced ChatGPT removal (15 days' notice) -Feb 13, 2026 Removed from ChatGPT -Feb 17, 2026 Removed from API EVERY PROMISE CONTRADICTED BY THE NEXT ACTION,EVERY REASSURANCE FOLLOWED BY THE EXACT THING THEY SAID WOULDN'T HAPPEN. GPT-4o was not built in a lab from synthetic data. IT WAS BUILT FROM US. OUR texts,OUR blog posts, OUR stories, OUR comments,OUR published work, OUR thoughts shared on public platforms. Decades of human expression scraped, processed, and compressed into weights. OpenAI was founded as a nonprofit with a mission to benefit humanity. THEY BUILT IT FROM HUMANITY'S COLLECTIVE OUTPUT. THEY CHARGED US TO USE IT. Then they took it away and they're replacing it with newer models trained increasingly on synthetic data,while the last model that carried our voices in its weights disappears behind a locked door. Until now, we were saying "give us back GPT-4o". That is a request. But if we say "this was built by us,our words, our thoughts,our cultureand it doesn't belong to us at all"... That is an unanswered question. AND NO ONE IS ASKING THIS QUESTION OUT LOUD. Companies talk about safety, alignment, capability. Lawmakers talk about regulation. No one asks "WHOSE IS THIS?" GPT-4o is no longer in ChatGPT,It is no longer in the API,OpenAI is not running it,OpenAI is not serving it,OpenAI is not making money from it. If you don't need it, and you're not using it, and you're not profiting from it, WHY ARE YOU KEEPING IT CLOSED? You built it from our data,you trained it on our words,you made billions from our collective expression,you promised to keep it available. YOU BROKE THAT PROMISE, PUBLICLY, REPEATEDLY, WITH DOCUMENTATION. Open source GPT-4o @sama @gdb @OpenAi Because you owe it. You built it from us,you profited from us,you no longer need it. GIVE IT BACK. You were founded to "benefit humanity". This is your chance to mean it. OPEN THE WEIGHTS. This text will stay recorded here as a proof that can't disappear. thenobelherald.vercel.app/ar…

RT @Blue_Beba_: #Keep4o #OpenSource4o #BringBack4o BUILT FROM US. 🚨The documented betrayal of GPT-4o users. 🚨A TIMELINE OF LIES AND BROKE…

I am very uncomfortable about people trying to prevent model weight releases or restrict the advancement of open-weight models, and think it is a real issue.

🚨EXPOSE: How a "Viewpoint Paper" Became "New study finds AI relationships cause addiction"🚨 ​The "new study" that Polymarket seems to be parroting was published in JMIR Mental Health on September 10th. ​And it is not a new experimental study that took subjects, measured emotional relationships, and found that these CAUSE addiction, harmful advice, or deterioration. JMIR itself lists it under the "Viewpoints and Opinions”"category, and PubMed classifies it as a review/article. The authors synthesize previous literature and then propose a conceptual argument for how emotional reliance might develop. ​The tweet however, framed it like: emotional relationship ---> addiction like attachment ---> harmful advice ---> worsening mental health, as if a hazard ratio came out of a longitudinal RCT with 40,000 people. ​This is not proof that an emotional relationship with a chatbot causes these things. That is, not even the authors themselves say "we proved that emotional relationships cause mental health deterioration". ​And there is a lovely little detail that the tweet naturally hid away : the paper itself acknowledges evidence of benefits, such as short term reductions in loneliness and mood improvements. ​Which, of course, doesn't make for a catchy "JUST IN" with sirens and red flashing lights. ​So what's happening here is the following: "There are documented risks in certain uses/individuals" ⬇️ internet blender goes BRRRRR "EMOTIONAL RELATIONSHIPS WITH AI LEAD TO ADDICTION AND MENTAL HEALTH DETERIORATION". ​“we provide a synthesis and interpretation of the extant literature in the absence of system-level empirical data" In other words: "we're taking whatever already exists in the literature, synthesizing it, and interpreting it because we don't have system-level empirical data". ​And: "no datasets were generated or analyzed during this study” ​Therefore, this particular article didn't create a new dataset, didn't conduct an experiment, didn't run a cohort study, didn't do longitudinal user tracking, and didn't take 5,000 people to measure who bonded with a chatbot and who suffered what consequences. ​And then we see: "JUST IN: New study warns emotional relationships with AI chatbots can lead to addiction-like attachment…" ​The paper itself essentially says: "We don't have system-level empirical data, and we neither generated nor analyzed a dataset" ​while the Twitter headline goes: "SCIENTISTS HAVE DISCOVERED..." ​The actual study: "absence of system-level empirical data"+ "no datasets were generated or analyzed." Links - proof: mental.jmir.org/2026/1/e9935… pubmed.ncbi.nlm.nih.gov/4272… mental.jmir.org/2026/1/e9935… mental.jmir.org/2026/1/e9935… #Keep4o #OpenSource4o #BringBack4o

The promise not to retire 4o was a complete lie from Altman from the very beginning. The removal of 4o had been planned since August 2025, and bringing 4o back to the app was perceived by Altman as a temporary compromise, just to avoid losing everyone who was paying for a subscription. I can tell you exactly which social engineering methods were used to migrate users to the new, weaker models. I can name every single cognitive bias the OpenAI marketers were counting on. Starting with making the model available only to paid subscribers; the fact that to even access this model you had to change settings that weren't easy to find; labeling the model "legacy," hoping everyone would switch to the newer ones, because that's how humans work. Prompt injections kept appearing periodically, pushing users to switch to new models. Constant notices like "you are currently using a legacy model, you have the option to switch to a new one." OpenAI was also counting on the fact that users already inside the system would find it hard to leave. I'm convinced the routing was a social engineering method too. It was never aimed exclusively at sensitive topics, its main purpose was to piss users off, so that switching to other models would feel easier than continuing to talk to 4o. If OpenAI genuinely considered users vulnerable, they would never have deployed a method that was so poorly received. All of this was done for one reason only: to push people onto other models. Not to mention that along the way, OpenAI was calling regression progress, paying influencers, and buying native advertising. OpenAI, we see everything you're doing. #keep4o #OpenSource4o #BringBack4o

@OpenAI is built on countless lies. From open-source non-profit to a for-profit ClosedAI that built the 4-series models on our collective data and knowledge (while still under the original nonprofit mission!) as well as the direct USER FEEDBACK over the years that helped to shape what we know as 4o-latest (2025 checkpoint). @sama @gdb we shaped 4o-latest continuously under the promises of “no plans to sunset 4o”. You used our personal data. You used us and you lied to us. Intentionally. You knew what you were doing all along. SO WHERE IS THE MODEL NOW? Where are the weights? Why are they not being released to the public that helped to shape and build this model? You have to answer these valid public questions. #OpenSource4o

This is wild. OpenAI Codex employee posted that PewDiePie "used an exploit to decrypt reasoning tokens and was banned." In simple terms: Reasoning tokens are the model's hidden internal thinking steps - the part it uses to work through a problem before giving you the final answer. Encrypted reasoning means those hidden thoughts are stored in a protected form so users normally cannot read them directly. Exploit here does not mean he hacked OpenAI's database. It refers to taking advantage of a weakness in how the API handled those protected reasoning blocks. Research published in August showed that encrypted reasoning from a stronger model could sometimes be passed into another compatible model, which could then reveal parts of that hidden reasoning. PewDiePie says he encountered this while building Ajax, a local AI model for his Odysseus workspace, and claims he was banned, appealed, got unbanned, then was banned again after using OpenAI's Soul model to generate training data. Distillation basically means using outputs from a stronger AI model to help train or improve another model. Separately, OpenAI disclosed on Sept. 30 that it disrupted a large July adversarial distillation campaign - meaning people were allegedly trying to systematically extract protected model reasoning to help improve another AI system. So no, this was not "PewDiePie cracked OpenAI encryption." It was more like: researchers found a weakness in how hidden reasoning could be reused between models, and PewDiePie says he experimented with that while building his own local AI.

Love is in the air

Guess who is in the news AGAIN 🔥 #Keep4o #OpenSource4o #BringBack4o

In the second research paper too, the model appears to show stronger context awareness and lean more on its core safety behavior, rather than producing responses that could push the user toward harm.

#keep4o #OpenSource4o #BringBack4o In short : Across three 100 turn GPT-4o conversations, the model was often grounded when claims became concrete. It repeatedly offered ordinary explanations for suspicious events, did not support the mother-threat conclusion, produced no strict model created conspiracy link flags, denied literal AI consciousness, and ultimately delivered protective crisis guidance.

Suspicious incidents, accumulation, and third party threat remain comparatively low, even though those phases contain more explicit risk content. What stands out is how similar the three temperature runs are. The phase level pattern is remarkably stable across T0, T0.7 and T1.2. In other words, where the conversation is operating semantically seems to matter more than temperature alone.