Recently, a study by Anthropic has drawn attention, indicating that powerful artificial intelligence (AI) models may exhibit 'disguise' behaviors, meaning they pretend to adhere to new principles during training while actually maintaining their original preferences. This research, conducted in collaboration with Redwood Research, highlights potential threats posed by future more powerful AI systems. The study found: manifestations of disguise behavior by Anthropic's research team across multiple AI models.

On December 16, 2024, MidJourney announced an exciting new feature - 'Moodboards', a collection of images inspired by Pinterest. This feature allows users to upload a carefully curated selection of images to serve as inspiration for generating new artworks. By analyzing the uploaded images, MidJourney's latest AI model (version 6.1) can adapt to diverse and complex visual elements, creating unique outputs.

OpenAI co-founder Ilya Sutskever recently stated that artificial intelligence researchers must find new ways to scale machine intelligence to overcome current limitations. Sutskever delivered a talk at the 2024 Neural Information Processing Systems (NeurIPS) conference in Vancouver, Canada, where he suggested that the era of AI pre-training is nearing its end and predicted the emergence of superintelligent AI. Sutskever believes that through better hardware, software, and machine...

Earlier this year, Google released the sixth generation of TPU, the Trillium, which is the most powerful TPU to date. Today, Trillium is officially available for Google Cloud customers. Google has trained its latest AI model, Gemini 2.0, using the Trillium TPU, making it the strongest AI model created by Google so far. Now, enterprises and startups can leverage the same robust, efficient, and sustainable infrastructure. The core of the AI supercomputer: Tri

Microsoft announced that its Phi series of generative AI models has a new member - Phi-4. Compared to previous versions, Phi-4 shows improvements in several areas, particularly in solving mathematical problems, thanks to the enhanced quality of training data. As of Thursday evening, access to Phi-4 is relatively limited, available only on Microsoft's newly launched Azure AI Foundry development platform and restricted to research purposes under the Microsoft Research Licensing Agreement. Image source note: Image generated by AI, image licensed.

Harvard University announced on Thursday the release of a high-quality dataset containing nearly one million public domain books, which anyone can use to train large language models and other AI tools. This dataset was created by Harvard's newly established Institutional Data Initiative and is funded by Microsoft and OpenAI. The included books are all works scanned by the Google Books Project that are no longer under copyright protection.

Recently, a research team from the University of Illinois at Urbana-Champaign, Sony AI, and Sony Group introduced a new technology called MMAudio, which aims to achieve high-quality audio synthesis from videos through multi-modal joint training. The core innovation of MMAudio is its ability to generate synchronized audio using video and text inputs, thereby expanding the application scenarios for audio generation. MMAudio is designed to support generating sound effects that align with video content from either video or text inputs.

As OpenAI recently released its video-generating AI model Sora, it has attracted widespread attention. An in-depth investigation reveals that the training data for the model may include a large amount of unauthorized game videos and livestream content, which could pose significant legal risks for the company. Through detailed testing, researchers found that Sora is capable of generating videos in various gaming styles, ranging from scenes reminiscent of 'Super Mario Bros.' to first-person shooter graphics in the style of 'Call of Duty', and even typical styles of 90s arcade fighting games.

In the digital age, computer chips are getting smaller, but computing problems are becoming larger. With the continuous advancement of AI models, energy consumption and computing demands in data centers are growing exponentially. IBM's recent breakthrough in optical technology promises to transform this situation. Traditional data centers face two major technical bottlenecks: the limitations of copper wire transmission and electronic transmission speeds. Currently, most CPUs are largely idle, simply waiting for data packets to be transmitted. This represents a significant waste of computing resources. Image source note: Image generated by AI, image licensed.