• Starter AI
  • Posts
  • From Zero to LLMs, Salesforce’s Einstein 1 Studio, Meta’s giant AI model

From Zero to LLMs, Salesforce’s Einstein 1 Studio, Meta’s giant AI model

What is Meta up to?

Author

Sthefania
March 07, 2024

Hello, Starters!

Thursday has arrived after a few days of ongoing AI news. We'll be back on Monday with more insights, as we're sure there will always be another surprise development to cover.

Here’s what you’ll find today:

  • Training LLMs from ground zero as a startup

  • Salesforce unveils Einstein 1 Studio

  • Meta is building an AI model for its “entire video ecosystem”

  • Teachers embrace ChatGPT

  • Gemini lets you fine-tune answers

  • And more.

🤖 Training LLMs from ground zero as a startup (9 min)

Yi Tay, Co-founder of Reka and former research scientist at Google Brain, recently shared an insightful blog post about the experience of building an LLM from scratch, or as he puts it, "from ground zero in the wilderness."

After leaving Google and starting to work at Reka, where they've successfully built strong multimodal LLMs, Tay states that the challenges often revolve around compute scarcity and unreliable compute providers, making the process feel like a hardship. 

However, as a conclusion from his experience, he believes that there's a lot to learn and embrace about building and training LLMs from a completely clean slate.

Salesforce has introduced Einstein 1 Studio at its TrailblazerDX developer conference. This suite of tools serves as a boost for their recent Einstein Copilot assistant, aiming to provide developers with complete access to customization.

Einstein 1 Studio features three main tools: Copilot Builder, Prompt Builder, and Model Builder. This aligns with Salesforce's vision of an AI-first future, where developers are equipped with the right tools to build applications for specific needs.

Meta's continuous push for AI is undeniable as the company consistently invests in the technology. This time, Tom Alison, Head of Facebook, has given a glimpse into their roadmap until 2026, which includes creating a single model to power Meta's entire video ecosystem.

This model aims to enhance user interaction by providing personalised recommendations of their "Reels" video service, as well as long traditional videos. The goal is to improve engagement and add more relevant content to people's feeds.

👍Google is giving more control to Gemini users on the web with a feature that allows them to modify a selected part of the chatbot's response to fine-tune the content and match it with exactly what they're looking for.

🧑‍🏫AI is still finding its way into the classroom, with models such as ChatGPT being at the centre of discussions as to whether it is a good ally for education or not. Nonetheless, according to reports, teachers are turning to chatbots to streamline the grading process.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!