💾 Google’s new AI chip

🧵 In today's edition:

• 💾 Google’s new AI chip

• 🦙 Llama 3 is coming soon

• 📰 AI exam graders

• 🛠️ AI Tools to Check Out

• 🤑 AI Fundraising News

• 🐦 Tweet Post of the Day

💾 Google’s new AI chip

As the competition for AI chips intensifies, Google has entered the race by introducing its proprietary Axion chip.

This chip is designed to manage the vast data needs of AI applications, likely as part of a strategy to lessen its dependence on leading chipmakers such as Nvidia and Intel.

Here’s the TL;DR:

• According to Google, the Axion chips are 30% more potent than similar products from Arm.

• These chips will be grouped in the thousands to enhance performance for AI tasks.

• While Google has developed chips for other business sectors before, Axion is its first chip designed explicitly for AI in data centers.

• Instead of selling the chips directly to customers, Google intends to make Axion accessible via its cloud services later this year. This strategy avoids immediate competition with long-standing partners Intel and Nvidia.

The impact of this announcement on the stock prices of some major chip manufacturers is also notable:

• Nvidia's stock declined by 2% following Google's announcement, despite its plans to launch its next-generation H100 chip soon due to high demand.

• Alphabet, Google's parent company, saw its shares initially surge by 2.4% before reducing gains.

As AI capabilities progress, the competition is escalating for hardware optimized for AI tasks' computational intensity. Google's Axion chip is a clear sign of its ambition to be a strong competitor.

Read more: Fortune

🦙 Llama 3 is coming soon

Meta has confirmed plans to release its next-generation language model, Llama 3, as open-source within the next month.

This is interesting for a few reasons:

• Llama 3 is Meta's latest attempt to compete with OpenAI and others in GenAI.

• Prior versions of Meta's open-source Llama models received criticism for being too limited compared to models like ChatGPT.

• This new model is expected to have significantly enhanced capabilities, allowing it to handle a broader array of queries and topics more accurately.

• Meta's Chief Product Officer, Chris Cox, stated that their goal is for an AI powered by Llama to become the world's most helpful assistant.

However, Meta has yet to provide detailed technical specifications for Llama 3, including the model's parameter count. It is anticipated to be around double the size of Llama 2, with approximately 140 billion parameters.

Releasing Llama 3 as open-source is part of Meta's strategy to build goodwill among developers and adopt a different approach than the closed models of companies like OpenAI and Google.

The company is planning a careful rollout, postponing releases like its image generation tool, Emu, due to safety concerns.

Read more: TechCrunch

📰 AI exam graders

Students taking Texas' STAAR exams this week are unknowingly participating in an interesting experiment—their open-ended responses are being graded by a new "automated scoring engine" instead of human graders.

The Texas Education Agency (TEA) is implementing this AI-powered natural language processing system throughout the state. Though firmly denying it's real AI, the engine is set to replace thousands of temporary human scorers, potentially saving TEA an estimated $15-20 million annually.

Here’s the TL;DR:

• This change follows the redesign of the STAAR exams in 2023, which added many more open-ended questions requiring human grading, a process TEA describes as overly time-consuming and expensive.

• The automated system was trained using 3,000 previously human-graded responses.

• Safety measures, like human re-scoring of 25% of the AI grades and responses that confuse the system, are in place.

Despite this, some districts have reported problems. One superintendent noted a "drastic increase" in zero scores from the automated grading during December 2023 trials.

While AI scoring engines for essays have been used for years, their accuracy is still debated. TEA seems aware of this perception issue, even claiming in small print that its "closed" automated system is "inherently different" from true AI, which can learn and adapt independently.

Read more: The Verge

🛠️ AI Tools to Check Out

• Gotalk: With Gotalk AI Voice Studio, you can generate lifelike AI voices for YouTube, Podcasts, and Phone System Greetings. Check it out!

• Mindpal: Create AI Agents that help boost your productivity at work. Check it out!

• Reclaim: The smartest calendar app that optimizes everyone’s time for better productivity, collaboration, and work-life balance. Check it out!

_{Have anything you’d like to share with over 40k AI enthusiasts? Sponsor Not A Bot}

🤑 AI Fundraising News

• PVML raises $8M in Seed funding to provide a data access platform that gives trustworthy AI-powered interfaces to enterprise data, secured by Differential Privacy.

• Sortium raises $4M+ funding to offer AI tools that automate content creation tasks in game production. These tools speed up production times, reduce costs, and allow studios to focus on innovation and design.

🐦 Tweet Post of the Day

Pretty cool - check it out!