Meta and Amazon Generative AI Moves
Sundays, The Sequence Scope brings a summary of the most important research papers, technology releases and VC funding deals in the artificial intelligence space.
Next Week in The Sequence
Edge #269: We start a new series about federated learning. Can’t miss that one! We discuss the basic concepts of federated learning as well as Google’s original paper in the space and the TensorFlow Federated framework.
Edge #270: We deep dive into Claude, the LLM built by Anthropic which has already raised over $1 billion to compete with ChatGPT.
📝 Editorial
During a recent conversation with a few VCs, I was asked whether I thought generative AI could be the trend that propels Microsoft and Google ahead of other tech incumbents. Google and Microsoft have quickly integrated large language models (LLMs) into their search engines and business app suites. Given that progress in generative AI is anything but linear, it's fair to assume that first movers could have a unique advantage.
When asked the question, I mentioned that I believed Meta AI had the research and distribution backbone to be highly competitive, especially with their active research in areas such as text-to-image and text-to-video. Amazon also has significant experience in this space with Alexa.
This week, both Meta and Amazon demonstrated early signs of their strategies in the generative AI space. Meta AI unveiled the research and early access to LLaMA, a 65 billion parameter LLM that works efficiently across 20 languages. While LLaMA is significantly smaller than GPT-3 (175 billion parameters), Meta AI claims it outperforms the OpenAI alternative on most benchmarks. Meta AI also enabled early access to LLaMA, and we plan to discuss it in depth in a future edition of this newsletter.
Amazon also made moves in the generative AI space, announcing a strategic alliance with Hugging Face to enable access to its foundation model suite in the SageMaker platform. As part of the collaboration, Hugging Face will also gain access to 1,000 GPUs available for a supercomputer cluster running SageMaker. This integration will make Hugging Face a first-class service in AWS, just like you can currently access the OpenAI API via Azure.
Both Meta and Amazon's announcements demonstrate their intent to be competitive in the generative AI space. However, I couldn't answer the question of "What about Apple?"
🔎 ML Research
LLaMA
Meta AI published a paper introducing the Large Language Model Meta AI( LLaMA), a 65 billion parameters language model able to master tasks across 20 different languages —> Read more.
Foundational UI Understanding
Google Research published a paper detailing a computer vision model able to master UI tasks from raw pixels —> Read more.
Few-Shot Region Aware Machine Translation
Google Research published a paper introducing a dataset to evaluate machine translation systems in their ability to support regional linguistic varieties —> Read more.
Text-To-Image for Robot Training
Researchers from Meta AI and the University of Washington published a paper discussing GenAug, a text-to-image method for imitation learning in robots —> Read more.
🤖 Cool AI Tech Releases
AI-Bing on Edge Mobile App and Skype
Microsoft previews its AI-powered Bing experience for Edge mobile apps and Skype —> Read more.
OpenAI Foundry
OpenAI unveiled details about Foundry, a program that allow developers to run instances of GPT-3.5 on dedicated infrastructure —> Read more.
LandingLens
Andrew Ng’s led Landing AI unveiled LandingLens, a new cloud platform for computer vision —> Read more.
🛠 Real World ML
Data Drifts at Uber
Uber discuses the architecture used to monitor data drift in their ML pipelines —> Read more.
Responsible AI at LinkedIn
The LinkedIn engineering team discusses some of the responsible AI principles that they try to enforce in their applications —> Read more.
📡AI Radar
AWS and Hugging Face announced a strategic partnership to enable next generation ML models in the AWS cloud.
Metomic announced a $20 million Series A to use AI to protect data leakages in SaaS applications.
Generative sound startup VoiceMod raised $14.5 million to continue expanding its text-to-music platform.
South Africa’s startup Envisionit Deep AI raised $1.65 million for its AI medical imaging platform.
Roblox is planning to leverage generative AI to enable the creation of virtual worlds.
AI sales startup Autobound raised $4 million to expand its generative AI for sales communications platform.
HireLogic raised $6 million Series A to bring conversational analytics to hiring processes.
Curated raised $2.4 million for its AI ecommerce platform.