Matterport Cuts AI Data Prep by 80%, 32 Experts Beat a LLama (v3), Meta Advances VLMs

Before we start, share this week's news with a friend or a colleague:

Key Takeaways

• Matterport slashes data prep times by 80% and enables Multimodal AI with Activeloop.

• Activeloop named a Cool Vendor in the 2024 Gartner® Cool Vendors in Data Management: GenAI Disrupts Traditional Technologies Report.¹

• Intel, Meta, Google, and Microsoft formed a group called Ultra Accelerator Link to develop alternative AI chips to challenge Nvidia’s market dominance.

• Sam Altman secured a partnership with Apple to integrate advanced AI functionalities into Apple products, while Opera partnered with Google to integrate Gemini into its Aria AI assistant.

• Google AI Overview produces bizarre suggestions like eating rocks for nutritional benefits, leading to a lot of public scrutiny on social media.

• Yuan-2.0-M32 is a MoE architecture with a new network router that outperformed models like Llama-3-70B on the ARC-C benchmark.

• EPFL researchers proposed a constant learning rate with cooldowns and stochastic weight averaging to improve training efficiency and model performance, based on Chinchilla's scaling laws.

Got forwarded this newsletter? Subscribe below👇

The Talk of the Day: Activeloop is named a Cool Vendor in the 2024 Gartner Cool Vendors in Data Management: GenAI Disrupts Traditional Technologies,

and How Matterport Delivers Multi-Modal AI with 80% Less Data Prep

First off, we've been recognized as a Cool Vendor in Data Management by Gartner.

We believe that this is a significant milestone for us, as few companies are acknowledged as Cool Vendors (specifically for Data Management, there's only 5 in the current batch).

We think that this is a strong step forward in our mission to help enterprises organize their unstructured, complex data while ensuring accurate knowledge retrieval.

If you have a Gartner subscription, you can read the full report here.

Gartner, Cool Vendors in Data Management: GenAI Disrupts Traditional Technologies. Nina Showell, Ehtisham Zaidi, Aaron Rosenbaum, 30 May 2024

How Matterport Decreased Data Prep Times by 80% and Enabled Multimodal AI

Matterport, a leader in 3D digital twins has digitized more than 35 billion square feet, making them one of the largest players in the domain. Before adopting Activeloop, Matterport faced challenges in managing their colossal datasets (comprising opt-in data). With over 7 million scanned spaces, the sheer size of the data posed significant logistical issue.

The company's Vision & Learning team partnered with Activeloop to streamline how they store complex data and retrieve it to train machine learning models at scale.

Results?

- 80% less time spent on data prep.

- Time to train on a new dataset reduced from hours to seconds.

Here are the challenges Matterport tackled with Activeloop:

• Rapidly evolving vast data

• Lack of standardization

• Multimodal support

Learn how by reading the full case study below.

The Latest AI News

Last week, we saw some pretty notable moves from all the major tech players. OpenAI has a partnership that allows it to integrate its AI technology into Apple products. Intel, Meta, Google, and Microsoft also didn’t show any signs of slowing down, as they formed a new group to develop AI chips.

Meanwhile, Google was ridiculed all over social media since Google AI Overview produced some…interesting suggestions like using non-toxic glue on pizza to keep the cheese in place (where was this when I was planning my stand-up routine?).