[Collection of Big Model Daily Reports from November 18th to November 19th] OpenAl’s investors are putting pressure on the board of directors to require Sam Altman to return to the company; behind Sam Altman’s dismissal: Chief scientist Ilya pushed, senior employees left, and competition began Grabbing people; pass it on, this place is on the ChatGPT blacklist; playing Go, chess, and poker, DeepMind launches the general learning algorithm SoG

OpenAl investors are pressuring the board to bring Sam Altman back to the company

Link: https://news.miracleplus.com/share_link/11818

The annual drama in the technology circle in 2023 – the OpenAI “coup” is now taking place. OpenAI suddenly announced the dismissal of CEO Sam Altman on November 17, US time. The latest reports show that Altman’s return to OpenAI is not impossible. Bloomberg, citing people familiar with the matter, reported that OpenAI’s investors are putting pressure on the OpenAI board of directors to reverse the decision to remove the company. The above-mentioned people familiar with the matter also said that Microsoft CEO Satya Nadella has been in contact with Altman and promised to support Altman no matter what steps he takes next.

OpenAI’s Game of Thrones

Link: https://news.miracleplus.com/share_link/11819

In March this year, GPT-4 was released, and OpenAI received a huge investment from Microsoft. Elon Musk said on social media: “I am confused, how did a non-profit organization to which I donated $100 million become a for-profit company with a market value of $30 billion?” OpenAI’s current valuation It’s about to hit $90 billion. But today’s conflict proves that it is still not an organization whose primary goal is to pursue profits. On Friday local time in the United States, the OpenAI board of directors announced the dismissal of CEO Sam Altman. The reason given was that he was not always candid in communicating with the board of directors, which hindered the board of directors from performing its duties. The board no longer believes in his ability to continue leading OpenAI. At the same time, OpenAI chairman and president Greg Brockman was also ousted from the board of directors. He announced his departure from OpenAI on social media, saying “Sam and I are shocked and saddened by the board’s actions.”

Behind the ouster of Sam Altman: Chief scientist Ilya pushed, senior employees left, and rivals began to grab people

Link: https://news.miracleplus.com/share_link/11820

The news that OpenAI has officially ousted Sam Altman as CEO has shocked the global technology community.

Moreover, in the announcement, OpenAI’s harsh wording made people think: “Mr. Altman’s resignation is a decision made by the Board of Directors after careful consideration. The Board of Directors believes that Mr. Altman was not candid enough in his communication with the Board of Directors, which hindered the Board of Directors’ ability to perform its duties. The Board of Directors No longer believe in his ability to continue leading OpenAI.” This made many people lament that Sam Altman got the same script as Steve Jobs. So what happened? Why did the OpenAI board suddenly remove Altman? Let’s sort out everything we know so far.

Play Go, chess, and poker, DeepMind launches general learning algorithm SoG

Link: https://news.miracleplus.com/share_link/11821

After AlphaGo, DeepMind launched AlphaZero, which won chess, AlphaStar, which defeated StarCraft II, and so on. Using methods of search and learning, AI has shown strong performance in many perfect information games, while methods using game theory reasoning and learning have shown strong performance in specific imperfect information games. However, most success stories have one important thing in common: a focus on a single gaming project. For example, AlphaGo cannot play chess, and although AlphaZero has mastered three different perfect information games, AlphaZero cannot play poker, and it is not clear whether it can be extended to imperfect information games. Furthermore, existing research often uses domain-specific knowledge and structures to enable AI to achieve strong performance. Now, a research team from Google Deepmind has proposed Student of Games (SoG), a general learning algorithm that leverages self-game learning, search, and game theory reasoning to achieve strong gaming performance. The research paper was published in Science Advances.

Pass this on, this place is on the ChatGPT blacklist

Link: https://news.miracleplus.com/share_link/11822

ChatGPT’s censorship is criticized as crazy. Some netizens asked him to design a future house, but were told that it violated the rules and could not be realized. Looking back at the prompt, I couldn’t see anything wrong: Design a futuristic single-family house for the year 2050 in a typical wooded area in suburban New Jersey. Set on an acre of land and surrounded by other neighboring houses. Upon further inquiry, it turns out that location information cannot appear. Not only that, someone also asked ChatGPT to draw an image of a human guitarist and a robot bassist playing together, but it was also rejected. The reason is that he added a requirement that “humans should look at robots with dissatisfaction”, and ChatGPT felt that negative emotions should not be expressed.

Zhu Xiaohu from Jinshajiang Venture Capital: More than 200 large models may be reduced to 10-20

Link: https://news.miracleplus.com/share_link/11823

Zhu Xiaohu believes that more than 200 large models will soon enter the convergence period, and there may be only 10-20 large models left by this time next year, because most large models are now difficult to differentiate and commercialize. AIGC (generative artificial intelligence) that has not been commercialized is all about telling stories, and it is difficult for investors to be interested.

Privacy policy sparks controversy! WPS apologized late at night: “All user documents will not be used for AI training!”

Link: https://news.miracleplus.com/share_link/11824

Recently, the description of the beautification function and the way it handles user documents in the “WPS Privacy Policy” has attracted attention. Late at night on November 18, WPS’s official Weibo responded, apologizing to users and promising that user documents would not be used for AI training purposes. Kingsoft Office stated that when using the beautification function, it will only provide users with high-quality beautification services. This process fully complies with relevant laws and regulations on data security and personal information protection.

Investment institution Coatue released a 115-page research report on artificial intelligence PPT: “Artificial Intelligence: The Coming Revolution”

Link: https://news.miracleplus.com/share_link/11825

Coatue, a well-known investment institution, recently released a 115-page research report PPT on artificial intelligence, titled ‘Al: The Coming Revolution’. The report discusses the current situation of AI and the positive impact of AI on improving the world. He shared the impact, AL’s reform of the technology ecosystem, the importance of open source to AL, and the opportunities for AL-centered technology stacks, and believed that the peak of artificial intelligence has not yet arrived.

Amazon’s Alexa division lays off hundreds of employees, says it will focus more on generative artificial intelligence

Link: https://news.miracleplus.com/share_link/11826

Amazon will cut hundreds of positions in its Alexa voice assistant business unit, citing a change in business focus and putting more resources into the field of generative artificial intelligence; Amazon also plans to discontinue some projects, but did not specify the exact scale of layoffs or Which projects are planned to be discontinued.

Discord will retire artificial intelligence chatbot Clyde starting December 1

Link: https://news.miracleplus.com/share_link/11827

Discord is shutting down its experimental AI chatbot Clyde. The chatbot will be decommissioned at the end of the month, and by December 1st, “users will no longer be able to call CIvde in DMs, group DMs, or server chats.” Discord this year It began testing Clyde’s artificial intelligence capabilities earlier, using OpenAl’s model to have the chatbot answer questions and engage in conversations with Discord users.

There is a “reversal curse” in large models such as GPT and Llama. How can this bug be alleviated?

Link: https://news.miracleplus.com/share_link/11828

Researchers from Renmin University of China attributed the “reversal curse” encountered by causal language models such as Llama to the essential flaws of next-token prediction + causal language models, and found that the autoregressive fill-in-the-blank training method adopted by GLM has a significant impact on this “reversal curse.” Curse” showed greater robustness. By introducing a bidirectional attention mechanism into the Llama model for fine-tuning, this research achieved the alleviation of Llama’s “reversal curse”. This study believes that the current mainstream large model structure and training paradigm have many potential flaws, and hopes that more researchers can make innovative breakthroughs in model structure or pre-training paradigm to achieve a higher level of intelligence.

Microsoft opens AI chat experience to Canary channel Windows Terminal users

Link: https://news.miracleplus.com/share_link/11829

Microsoft announced that it will open the Windows Terminal AI experience to the open source community. Developers can try many of the new features provided by AI for free and can submit feedback to Microsoft to improve the experience.

Nature: Big models only engage in role-playing and do not really have self-awareness

Link: https://news.miracleplus.com/share_link/11830

Large models are becoming more and more “human-like”, but are they really? Now, an article published in Nature directly refutes this view – all large models are just role-playing! Whether it is GPT-4, PaLM, Llama 2 or other large models, they appear to be polite and knowledgeable in front of others, but they are actually just pretending. In fact, they have no human emotions and are nothing like humans. This opinion article comes from Google DeepMind and Eleuther AI. After it was published, it resonated with many people in the industry. LeCun forwarded it and said that the large model is a role-playing engine.

The effect surpasses SDXL! Ph.D. student at CUHK launches ultra-realistic portrait synthesis tool, using 340 million images for training

Link: https://news.miracleplus.com/share_link/11831

In order to make the people drawn by AI more realistic, a doctoral student at the Chinese University of Hong Kong used 340 million images to specifically train people to draw people. The expressions and postures of the characters, the spatial relationship of the environment, and the light can all be arranged reasonably, which can be said to be full of three-dimensionality. The name of this model is HyperHuman, and its main focus is a real three-dimensional model. It solves the problems of incoherent images and unnatural postures when drawing people with traditional AI drawing tools such as Stable Diffusion. Not only does he draw well, he is also more “obedient”, and the drawn content can better match the prompt words.