Big Model Daily, November 24

News7months agorelease AIWindVane
35 0
Big Model Daily, November 24

[Big Model Daily, November 24] News New breakthrough in video generation: PixelDance, easily presenting complex movements and cool special effects

A new breakthrough in video generation: PixelDance, easily presenting complex movements and cool special effects

PixelDance has taken a critical step. The dynamics of its generated results are significantly better than other existing models, which has attracted the attention of the industry. On the official website (, PixelDance provides two different video generation modes. The first is Basic Mode. Users only need to provide a guidance picture + text description, and PixelDance can generate a highly consistent and dynamic video. The guidance picture can be a real picture or use Existing Vincent diagram model generation. The second is the advanced Magic Mode, which gives users more room to use their imagination and creativity. In this mode, users need to provide two guidance pictures + text description, which can better generate more difficult video content. The website shows various cool special effects shots made with magic mode.

Meta takes action on the Transformer architecture: the new attention mechanism understands reasoning better

Recently, Meta researchers believed in the paper “System 2 Attention (is something you might need too)” that the fundamental problem lies in the inherent construction method of Transformer itself. Therefore, the researchers proposed a completely different approach to attention mechanism by using LLM as a natural language reasoner to perform attention. Specifically, they exploited the ability of LLMs to follow instructions, prompting them to generate context that they should pay attention to, so that they only include relevant material that does not distort their reasoning. The researchers call this process System 2 Attention (S2A), and they view the underlying transformer and its attention mechanism as automatic operations similar to human System 1 reasoning.

AIGC startup has not yet made a profit, but Microsoft and Adobe have already made a lot of money

Since the beginning of the year, Adobe’s U.S. stock market value has increased by more than 100 billion U.S. dollars, which has unknowingly created a 90.58% increase; and Microsoft, OpenAI’s largest shareholder and released a series of AI Copilot products, has also increased by an eye-catching 55.93%. This type of non-AI-native veteran players seems to have very traditional main products, but AIGC products have quietly occupied the mainstream.

Refining large models in real scenes, Quark uses AI to “speed up” again

On November 14, Alibaba Intelligent Information Business Group released the self-developed Quark model with hundreds of billions of parameters. Once released, it topped the two authoritative evaluation lists of C-Eval and CMMLU. Taking advantage of the trend of AI reconstruction applications, Quark Model will comprehensively upgrade its product matrix and services.

Convert your voice at will, ElevenLabs releases “Voice to Speech”

Elevenlabs, a speech-generating AI platform, has released a “speech-to-speech” (STS) function on its official website, which can help users automatically convert speech into other voices. For example, if a male voice is uploaded, it can be automatically converted into a female, elderly or child’s voice. sound. Experience address:

Nvidia accused of stealing trade secrets! A video conference was held to expose evidence, and the employee involved pleaded guilty.

Auto parts leader Valeo sued Nvidia for stealing trade secrets. Valeo employees discovered during a video conference that Valeo’s source code files appeared on the shared screen of Nvidia employees, and they immediately took screenshots to collect evidence. In September this year, the employee was found to have illegally obtained, used and disclosed Valeo’s trade secrets and was fined 14,400 euros (equivalent to nearly 120,000 yuan). Valeo recently formally filed a lawsuit in federal court in San Jose, California. Nvidia has not yet responded.

© Copyright notes

Related posts

No comments

No comments...