• Starter AI
  • Posts
  • Altman is (officially) back, Microsoft’s NaturalSpeech 3, Huawei’s new model

Altman is (officially) back, Microsoft’s NaturalSpeech 3, Huawei’s new model

OpenAI’s board is on track.

Hello, Starters!

The wear and tear of the weekend may catch up with us this Monday, but we have the best selection of AI insights to help us make it through.

Here’s what you’ll find today:

  • Sam Altman has been reinstated

  • Microsoft presents NaturalSpeech 3

  • Huawei’s new text-to-image model

  • xAI will open-source Grok

  • Inflection AI launches Inflection-2.5

  • And more.

After months of a rollercoaster that kept the AI landscape in shock, Sam Altman has been reinstated to OpenAI's board of directors after an investigation commissioned by the company's board found no sign of wrongdoing. Both board members and employees believe in Altman's vision and trust him and Greg Brockman as the right leaders.

Additionally, OpenAI has announced the integration of three new board members, all women: Sue Desmond-Hellmann, Nicole Seligman, and Fidji Simo. This move aims to bring more representation and diversity to the company.

NaturalSpeech 3 is the result of collaboration between Microsoft Research Asia, Azure Speech, and partner universities. It is capable of cloning voices and emotions through the use of a technique that divides speech into different units, such as content, prosody, timbre, and acoustic details.

The breakdown of speech into these sub-areas allows NaturalSpeech 3 to achieve more detailed and controlled results that truly resemble the human voice. This differs from previous text-to-speech models that failed in terms of naturalness.

A group of researchers from Huawei's Noah's Ark Lab presented PixArt-Σ (Sigma), a groundbreaking text-to-image model inspired by previous work from the company. This recent iteration boasts improved image quality, prompt accuracy, and efficiency in handling training data.

The team harnessed multiple methods that enable PixArt-Σ to better understand images and increased the token length to approximately 300 words. This adjustment allows the model to achieve better matches between the prompt and the results.

💥Elon Musk has announced through a post on X that xAI is planning to open-source its chatbot "Grok" this week. This move doesn't come as a surprise, as the CEO has consistently expressed interest in open-source AI.

🤖Inflection AI unveiled Inflection-2.5, a revamped version of its in-house model that powers its famous chatbot, Pi. According to the company, Inflection-2.5 approaches GPT-4's performance by using only 40% of computing resources for training.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!