• Starter AI
  • Posts
  • Etched presents Sohu, Amazon’s upcoming Metis, OpenAI delays “Voice Mode”

Etched presents Sohu, Amazon’s upcoming Metis, OpenAI delays “Voice Mode”

A new chatbot is coming!

Hello, Starters!

Sometimes AI developments happen so quickly that we cannot keep track of all of them; other times, great features that we've been expecting for a while don't come as quickly! Every step is a game of patience!

Here’s what you’ll find today:

  • Etched is building the next AI chip

  • Amazon will finally have a chatbot

  • OpenAI pushes back “Voice Mode”

  • AI model detects 13 cancer types

  • Andrej Karpathy shares “Storyteller AI LLM”

  • And more.

Etched is a newly founded startup from California that has a big plan in mind. They're developing next-generation AI chips that are set to dethrone Nvidia's reign. The chip-making industry is a tough one, and many big tech companies are also making efforts to come up with their own. However, Etched is confident its arrival is happening at the "right place and right time."

The chip, called Sohu, is destined for AI models using a “transformer” architecture, like OpenAI's ChatGPT. According to the startup's CEO and Co-Founder, Gavin Uberti, as AI keeps advancing the need for specialised chips will grow, shifting away from Nvidia's GPUs, as they're not as cost-effective.

Reports reveal that we're not far from seeing Amazon's presence in the chatbot arena. The company has been developing an AI chatbot dubbed "Metis," which may be a strong contender to ChatGPT. Metis will be powered by their Olympus model, currently used internally.

The chatbot is set to be accessible through the web, and it will do what most chatbots do: answer queries, generate images, and provide sources for its responses. Metis is an ambitious project for Amazon, and it is expected to launch this September, alongside the company's Alexa event, which will also reveal the assistant's new AI-powered capabilities.

After an eventful month, OpenAI has announced some delays on its roadmap. The upcoming Voice Mode, which turned heads at their latest event, was initially planned to be available for Plus users this week, but now the company is telling customers to wait a bit more, as they want to do a few tweaks to its reliability and safety before finally going public.

According to OpenAI, they aim to improve the model's ability to “detect and refuse certain content,” as well as get the infrastructure ready to meet demand, while keeping the model able to give real-time responses. There's still no exact timeline for its launch.

🏥EMethyINET, an AI model developed by researchers from the University of Cambridge and trained with over 6,000 tissue samples from the Cancer Genome Atlas that includes 13 different types of the illness, has achieved a detection accuracy of 98.2%, bringing hope to early detection and advancements in diagnosis and treatment.

📚Continuing with its goal of making AI learning more accessible, Andrej Karpathy recently shared a new course destined for those who want to explore their creative side with the help of AI. Available on GitHub, users can learn how to build a Storyteller AI LLM to create, refine, and illustrate stories.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!