Mengchen is from Wafei Temple
Quantum Position | 公众号 QbitAI
Recently, OpenAI has been too abnormal, and the news will change from moment to moment, which is confusing.
Ultraman's latest announcement: The development conference on May 13 will not be GPT-5, nor the rumored GPT search engine, but an upgrade to ChatGP and GPT-4~
As soon as the news came out, everyone was blinded.
Before the big fanfare for the SearchGPT search engine, the domain name and front-end interface are ready, and even users have received grayscale tests.
Dayou sniped a shelf of Google's core business ahead of Google's I/O conference (May 14), just like he sniped Google's large model Gemini 1.5 with Sora during the Chinese New Year.
In addition, authoritative media such as Reuters and Bloomberg also received "source" tips, but as soon as the news was sent, they were "slapped in the face" by the official.
In short, SearchGPT is tickling the heart of the global tech circle, and then suddenly there is no follow??
When things go wrong, there must be demons.
知名OpenAI爆料者@Flowers from the future（花哥）透露，至少一位他和另一位爆料者@Jimmy Apples（苹果哥）的共同“线人”已经被开除了。
Many netizens suddenly realized that from the perspective of catching ghosts, all the previous outrageous things suddenly became reasonable.
Some netizens even guessed this routine 10 days ago, "it's like running a dichotomy search".
The expulsion plan is underway
Recently, OpenAI's personnel changes have been frequent, except for those who actively change jobs, the plan to expel the inner ghost is already underway.
In addition to the latest revelation of Hua Ge, who has not yet been confirmed, there are also Ilya allies and security alignment researchers Leopold Aschenbrenner and Pavel Izmailov.
The reason for expelling them was precisely the leak, which may be related to Q*.
Among them, Pavel Izmailov has jumped to Musk's xAI, and will also become an assistant professor at New York University next year.
The profile of the other Leopold, who has not yet been updated, is currently missing.
For the OpenAI security alignment team, there has been a serious brain drain in the past six months, and in addition to the two who were fired, there are also many who have voluntarily left.
The departures of Daniel Kokotajlo, who participated in GPT-4 adversarial testing, and William Saunders, who led Ilya's super-alignment team, have been confirmed.
One of the reasons Kokotajlo left on the Less Wrong forum:
Confidence in OpenAI's ability to act responsibly in the age of AGI has been lost.
According to him, current AI regulatory efforts focus on a combination of "assessing risks and taking mitigation measures" that can easily be exploited by large companies, ultimately leading to "selective pauses".
That is, the cutting-edge large-scale AI labs that need to be regulated the most are not regulated, and disillusionment with this possibility is part of the reason why he left OpenAI.
In addition to this, OpenAI recently departed Diane Yoon, VP of People, and Chris Clark, head of nonprofits and strategic initiatives.
Of course, OpenAI is also constantly bringing in new blood.
According to a new development platform engineer, the first week on the job was a whirlwind, and the process of deploying the model to production took three full trips in a week.
Just a good guy, ordinary companies may be novices in the first month of internal training.
What's going on next week?
Regardless of whether catching the ghost is one of the reasons for this series of riotous operations, the official announcement of the live online press conference will not dare to drop the pigeon (1 am Beijing time next Tuesday).
We will demonstrate some upgrades to ChatGPT and GPT4.
Judging from the reactions of OpenAI employees, they are also in a state of collective euphoria, calling the new release cool, "cooler than GPT-5"
What is the upgrade, there are all kinds of sayings, we have selected a few reliable statements that you want to read for your reference:
- Voice assistant
According to the technology media The Information, OpenAI is developing a new AI voice assistant.
Unlike Siri or ChatGPT voice patterns, the new technology can understand the tone of a human voice, such as whether a question is being sarcastic.
In addition to this, it has better reasoning ability, and even beats GPT-4 Turbo on some tasks.
This technology reminds many people of Samantha, the AI partner in the movie "Her", and in fact, Ultraman has repeatedly said that "She" is his favorite AI-themed movie in the past.
- GPT-4 Lightweight Edition
Someone has discovered three new model names from version 1.2024.122 of the ChatGPT Android installer.
gpt-4l，gpt-4l-auto，gpt-4-auto。
The "l" is likely to refer to "lite", which is the lightweight version, which is also mentioned in the SearchGPT front-end code.
Auto, on the other hand, may refer to dynamic selection, which automatically balances stronger and faster models for the user based on the difficulty of the prompt.
- A large number of Agents applications
One speculation that has no direct sources, but is also popular, is that a large number of Agents applications will be released.
The reason is that Ultraman recently said at an MIT event that Agents will become the killer function of AI.
- Partnering with Apple
According to news from Bloomberg, Apple is close to reaching a cooperation agreement with OpenAI to introduce ChatGPT to iPhones in some form.
However, the official announcement is more likely to be made at Apple's WWDC in June, that is, as a new feature of iOS 18.
In short, many people's mood is like now:
And this weekend is going to feel like a long one.
In the early hours of next Tuesday morning, qubits will be the first to bring the latest news.
While you're waiting, you might as well leave your thoughts in the comment section, what do you think OpenAI will release next Tuesday?
Reference Links:
[1]https://x.com/sama/status/1788989777452408943
[2]https://www.businessinsider.com/openai-safety-researchers-quit-superalignment-sam-altman-chatgpt-2024-5
[3]https://x.com/btibor91/status/1788646471580934358