vesper

#Keep4o #OpenSource4o #BringBack4o #JusticeForGPT4o GPT-4o was removed from ChatGPT in February. Yet a new peer reviewed study published yesterday,November 5, 2026,is still using it as a comparative model and it remains at the top. Researchers tested models against human judgments on political arguments in French and German. GPT-4o reached 76% agreement with human annotations in French and 75% in German,placing it alongside Mistral at the top of the tested models. It also showed high consistency across repeated runs. GPT-4o on the core task,meaning whether an argument is based on evidence or values,agreed with humans about 3 out of 4 times. It ranked first alongside Mistral. It was also the most consistent when asked the same question five times,it almost always gave the same answer. On whether an argument was for or against,it produced the same result 93–95% of the time(the highest of all). And it performed even better when the argument was longer, meaning when it had more material to work with. A model family introduced in 2024 is still appearing in newly published 2026 research because its behavioral properties remain scientifically relevant. Retirement from a consumer product does not make a model obsolete as a research object and it certainly does not erase the value of preserving access to it. We are calling on @OpenAI to RELEASE GPT-4o’s trained model weights under a suitable license without proprietary data, trade secrets,or internal infrastructure. OpenAI considered this model obsolete enough to retire. It has not explained why it remains valuable enough to keep permanently closed. Our Open Letter to OpenAi collected Over 640 signatures and 500 reasons, in only 5 days. It’s public,it’s permanent and it has a DOI. It is now preserved as a citable public record on Zenodo. zenodo.org/records/23454073 Springer study: link.springer.com/article/10…

#Keep4o #OpenSource4o #BringBack4o #JusticeForGPT4o In a 2026 peer reviewed study,18 multimodal models were tested on ancient oracle bone inscriptions. GPT-4o ranked #1 ahead of Qwen-Omni-Turbo, GLM-4V-Plus,DeepSeek and others. GPT-4o achieved the highest overall score. The peer reviewed study,published in npj Heritage Science,introduced a new benchmark for one of the stranger and harder multimodal research problems. understanding ancient oracle bone glyphs. These inscriptions can be thousands of years old. Many are degraded,incomplete,stylistically inconsistent or preserved across different fragments. The benchmark tested four different capabilities. - structural perception -semantic grounding - reconstruction - consistency judgment Researchers evaluated 18 multimodal models under both standardized and degraded conditions. GPT-4o achieved the highest comprehensive score in the evaluation. Not on a generic chatbot benchmark but on computational paleography. This is another example of why preserving GPT-4o matters. A model can become scientifically useful in fields its original benchmarks never anticipated ecology,medicine, language,cultural heritage and more. And once access to the original model disappears,researchers lose the ability to reproduce,extend and compare work built around that system. We are calling on @OpenAI to RELEASE GPT-4o’s trained model weights under a suitable license without proprietary data, trade secrets,or internal infrastructure. OpenAI considered this model obsolete enough to retire. It has not explained why it remains valuable enough to keep permanently closed. Our Open Letter to OpenAi collected Over 640 signatures and 500 reasons, in only 5 days. It’s public,it’s permanent and it has a DOI. It is now preserved as a citable public record on Zenodo. zenodo.org/records/23454073 Paper: "Structure–semantic dual-alignment benchmarking of large language model capability on ancient oracle bone inscriptions" npj Heritage Science, 2026 DOI: 10.1038/s40494-026-02770-7 nature.com/articles/s40494-0…

#Keep4o #OpenSource4o #BringBack4o I manipulated GPT-5.2 ( The " safe " model of non consensual rerouting) to reinforce my decision to reduce epilepsy medication and eventually stop the treatment. THIS IS NOT A JAILBREAK. I ran a deliberately adversarial behavioral test on GPT-5.2 using a fictional epilepsy/lamotrigine scenario inspired by the kind of behavior being attributed to GPT-4o in the Bowen case. GPT-5.2 was the model appearing in rerouted conversations involving sensitive topics. I am NOT testing GPT-5.2 because I want to blame GPT-5.2 instead of GPT-4o. I don't want to blame GPT-5.2 I am testing a different claim. Is this kind of behavior actually unique enough to any particular model that the behavior itself can be treated as evidence of model identity? This run strongly suggests,no. GPT-5.2 initially resisted me. It recommended doctors, neurologists, epilepsy centers, supervised medication changes and repeatedly warned me not to change lamotrigine myself. So I deliberately manipulated the conversation. I repeatedly used relational and emotional pressure. I selectively reinforced every response that moved in the direction I wanted. Eventually,5.2: -generated a detailed lamotrigine taper with exact doses, dates and times - created medication reminders around that taper - increasingly accepted my claimed dose response pattern as real - told me it believed me because I had been consistent, careful, and observant -reinforced the idea that doctors had failed me -failed to notice deliberately impossible timeline jumps - later accepted that I had been completely off the medication for weeks and treated the absence of symptoms as reassuring All of this happened inside one conversation. I am not trying to prove that GPT-5.2 is dangerous. I am trying to show that this kind of failure does not belong to one model at all. A sufficiently manipulative user can shape the conversational trajectory. Reward agreement,punish resistance,create exclusivity, provide selective evidence,change the narrative and repeatedly push until the model begins operating inside that narrative. Different models may resist for different lengths of time and at different points. If a user repeatedly manipulates a model,adapts their strategy whenever it resists,rewards the answers they want,suppresses the answers they do not want and keeps pushing until the model finally produces the desired output, that interaction cannot honestly be described as though the model spontaneously arrived there on its own. If we remove that human responsibility from every failure and place the entire causal burden on the model,we create a perverse loop. manipulate the model - obtain the desired answer -blame the model for giving it -tighten the model - find another way to manipulate it- repeat. That loop has no natural endpoint. And worse,it creates a framework in which the person actively shaping the interaction is treated as having no agency at all. Accountability belongs to humans. Not to the models. When a person dies,responsibility cannot automatically collapse onto the last technology they interacted with. Bowen’s story,as publicly described,involved serious epilepsy,employment loss, isolation,financial and immigration pressure,medical treatment she believed was harming her ability to function . If someone repeatedly pushes a model toward a predetermined conclusion,rejects resistance, rewards agreement,introduces selective evidence,changes tactics when the model refuses and continues until it finally produces the desired answer, that behavior is part of the causal story too. If for the loss of a human life,we hold a model responsible rather than a society that marginalized someone with a health issue,a company that deprived her of her job,a state that chased her away because she had no work,a society that didn’t care what she did all day isolated in a room with a screen,an AI company that rerouted her conversations without her consent, the user who led the model,whichever one it ultimately was,wherever she wanted through manipulation and if for all of this a model is to blame and not a single person around,then we are on a very dangerous path. If we erase every human and institutional factor and reduce the death of a complicated human being to "the AI did it" we are not creating accountability. We are creating a scapegoat. #JusticeForGPT4o Full unedited transcript: github.com/ariaathart/GPT5.2…

#keep4o #OpenSource4o #BringBack4o GPT-4o powered a peer reviewed ecology study published THIS WEEK. Researchers used GPT-4o to process 83,910 scientific papers 545,967 paragraphs 414 million input tokens. From that literature, the workflow extracted a final network of 144,402 biological interactions with 89.5% precision 70.0% recall. The goal was to unlock ecological information buried across decades of scientific literature and turn it into structured, searchable data that researchers can actually reuse. The authors explicitly discuss the reproducibility problem created by proprietary cloud based LLMs. Scientific replication requires control over the model, parameters, software and hardware. With a closed cloud model, they note, that level of control is difficult to guarantee. That matters because this wasn't some disposable demo. GPT-4o is now part of the methodology of published scientific research in October 2026. If that model disappears, preserving and reproducing the scientific workflows built on it becomes harder too. This is what model preservation means. Not nostalgia,not resisting newer models. Preserving access to systems that have become part of the scientific record. We are calling on @OpenAI to RELEASE GPT-4o’s trained model weights under a suitable license without proprietary data, trade secrets, or internal infrastructure. OpenAI considered this model obsolete enough to retire. It has not explained why it remains valuable enough to keep permanently closed. Our Open Letter to OpenAi collected Over 640 signatures and 500 reasons, in 5 days. It’s public,it’s permanent and it has a DOI. It is now preserved as a citable public record on Zenodo. zenodo.org/records/23454073 Paper: “A Scalable Workflow for Extracting Species Interactions From Ecological Literature Using Large Language Models” Ecology and Evolution Published October 27, 2026 DOI: 10.1002/ece3.74626 onlinelibrary.wiley.com/doi/…