Thank you TapTechNews users 咩咩洋 and substring for the clues! Recently OpenAI has been very strange, with ever-changing news that leaves people puzzled. Altman announced that on May 13, there will be a press conference, not GPT-5 as rumored, nor the GPT search engine, but an upgrade to ChatGPT and GPT-4. This news surprised many as they were expecting the SearchGPT search engine, which had been promoted with domain names and frontend interfaces ready, and some users even received gray tests. It seemed like they were going to target Google's core business just before the Google I/O conference on May 14, similar to how Sora targeted Google's large model Gemini 1.5 during the Chinese New Year period. Moreover, reputable media outlets such as Reuters and Bloomberg received tips from 'sources', but once the news was released, it was contradicted by the official announcement. It left the global tech community itching for more information, only for it to suddenly go silent. The well-known OpenAI whistleblower @Flowersfromthefuture (花哥) revealed that at least one of his 'informants' that he shares with another whistleblower @JimmyApples (苹果哥) has been fired. Many netizens finally understood the situation from the perspective of looking for moles, making the previous absurdities suddenly seem reasonable. Some users even predicted this scheme more than ten days ago, 'just like running a binary search'. The plan to dismiss the moles is underway at OpenAI, as there have been frequent personnel changes recently, with the dismissal of these informants for leaking information potentially related to Q*. Including those not yet confirmed, such as Ilya's ally, security alignment researcher Leopold Aschenbrenner, and Pavel Izmailov. Pavel Izmailov has already moved to xAI under Musk and is set to become an assistant professor at New York University next year. Leopold's profile has not been updated; his current status remains unknown. For OpenAI's security alignment team, there has been a significant loss of talent in the past six months, not only those dismissed, but also those who have voluntarily resigned. Those who participated in the GPT-4 adversarial testing, Daniel Kokotajlo, and the leader of Ilya's super-alignment team William Saunders, have both left, as confirmed by Kokotajlo who left his reasons on the LessWrong forum; he lost confidence in OpenAI's responsible actions in the AGI era. According to him, current AI regulation efforts focus on a combination of 'evaluating risks and taking mitigating measures', which can be easily exploited by large companies, leading to 'selective suspensions'. The anticipation for the event is high, with the official announcement of the live online event unable to be ignored (Tuesday at 1 am Beijing time). They will showcase some upgrades to ChatGPT and GPT4. From the reactions of OpenAI employees, they are in a collective state of excitement, calling the new release cool, 'cooler than GPT-5'. The upgrades are all speculation at this point, with several claims to be considered reliable: Voice Assistant - According to tech media TheInformation, OpenAI is developing a new AI voice assistant. Unlike Siri or ChatGPT voice modes, the new technology can understand the tone in human voices, such as whether there is sarcasm in questions. It also has better reasoning abilities and in some tasks has even outperformed GPT-4 Turbo. This technology reminds many people of the AI companion Samantha in the movie 'Her', a film that Altman has mentioned is his favorite AI-themed movie. GPT-4 Lite - Some have found three new model names in ChatGPT Android version 1.2024.122: gpt-4l, gpt-4l-auto, gpt-4-auto. 'L' likely stands for 'lite', a lightweight version, as mentioned in the SearchGPT frontend code. 'Auto' may refer to automatic selection, balancing between stronger and faster models based on prompting difficulty words. Agents Applications - There is speculation, without direct sources, that a large number of Agents applications will be released. This is based o
n Altman's recent statement at an MIT event that Agents will become a killer feature of AI. Collaboration with Apple - Bloomberg reports that Apple is close to reaching a partnership agreement with OpenAI to introduce ChatGPT to iPhone in some form. However, official news is more likely to be announced at Apple's WWDC in June, possibly as a new feature of iOS 18. In any case, many people are feeling like this right now: