• Starter AI
  • Posts
  • Sakana AI’s merging models, OpenAI turns to Hollywood, Microsoft’s LLMLingua-2

Sakana AI’s merging models, OpenAI turns to Hollywood, Microsoft’s LLMLingua-2

Sora might attend the next Oscars.

Author

Sthefania
March 25, 2024

Hello, Starters!

No doubt, AI is making its way to the big screen, either through tools to assist filmmakers or in blockbuster films showcasing its transformative impact... Maybe we'll see OpenAI on the red carpet.

Here’s what you’ll find today:

  • Sakana AI’s “Evolutionary Model Merge”

  • OpenAI looks for partnerships in Hollywood

  • Microsoft presents LLMLingua-2

  • Financial Times unveils a new chatbot

  • Google shows AI-powered results for more users

  • And more.

Borrowing inspiration from nature, with notions such as evolution and collective intelligence to develop groundbreaking foundation models, Sakana AI has introduced a method dubbed "Evolutionary Model Merge." This method combines existing open-source models with diverse capabilities to form new models.

The company has tested this method by training three foundation models to accelerate AI development in Japan: Large Language Model (EvoLLM-JP), Vision-Language Model (EvoVLM-JP), and Image Generation Model (EvoSDXL-JP), with two of them already available on Hugging Face.

After the announcement of Sora, OpenAI's upcoming AI video generator, sources state that the company is in ongoing talks with prominent figures in the filmmaking industry, such as actors and directors, to integrate Sora into their workflow.

Although Sora is not yet available to the public, reports suggest that OpenAI is granting access to some of these figures to showcase Sora's capabilities as the company continues to meet with film studios, media executives, and talent agencies in Los Angeles. According to OpenAI, this is part of their strategy for safe implementation.

Microsoft Research has recently released LLMLingua-2, a model designed for conciseness. This approach reduces prompts by removing unnecessary words or tokens while maintaining key information. As the research team declares, "Natural language is redundant."

Consequently, LLMLingua-2 shortens the prompts to 20 percent of the original length, resulting in lower costs and, therefore, lower latency.

🤖The Financial Times has unveiled its new chatbot called Ask FT, which is powered by Anthropic's Claude and aims to help subscribers solve any queries about articles published in the media outlet, as it has been trained with decades of information.

🔎Google has started rolling out its generative AI technology in Google Search for selected users in the US who didn't opt to activate the feature. In this way, the company expects to gather feedback before releasing the option to the general public.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!