• AI For Business
  • Posts
  • Google’s $20m responsible AI fund 💸, OCR for academic documents 📃, RAG at massive scale 🌎

Google’s $20m responsible AI fund 💸, OCR for academic documents 📃, RAG at massive scale 🌎

AI For Business

Headlines & Launches

Google pledges $20 million for responsible AI fund (2 minute read)

Google's philanthropic arm is investing $20 million in the Digital Futures Project to promote responsible AI development. The initiative supports global institutions, addressing AI's societal impacts. Google.org stresses collaboration for AI's potential benefit.

More companies commit to the White House AI safety accord (2 minute read)

Adobe, IBM, Nvidia, and others have committed to the White House’s accord on AI safety and trustworthiness, echoing earlier pledges by Meta and Google. These voluntary agreements stress pre-release testing and risk-sharing. AI regulation remains a step behind innovation under the Biden administration.

OCR for academic documents (7 minute read)

Optical character recognition (OCR) is the process of extracting text from images. It can fail on documents with lots of jargon or special characters like mathematics. This work from Facebook research showcases strong performance across academic domains, enabling the digitization of many old texts.

Research & Innovation

AI gets a sense of smell (29 minute read)

Researchers were able to train a model to accurately recognize smells by training a graph neural network on a custom data set of thousands of hand-labeled molecules.

Is Your Writing Losing Its Uniqueness? The Impact of Using AI on Content Diversity (18 minute read)

This study looked at whether advanced AI writing assistants like InstructGPT were making people's writing sound too similar and potentially making public conversations less diverse.

Turning Real-World Pictures into Stunning Anime Scenes (3 minute read)

Researchers have introduced Scenimefy, a new tool to transform everyday photos into detailed anime-style images.

Engineering & Resources

Retrieval Augmented Generation at Planet Scale (8 minute read)

Arcus scales RAG to planet scale by using a hierarchical retriever. After clustering documents into groups based on their semantic content, you can progressively filter over these groups to narrow the search space. This results in much more relevant context being retrieved, fewer hallucinations, and better reliability over planet-scale corpuses of data.

Whisper Turbo (GitHub Repo)

A drop-in replacement for the OpenAI Whisper API that offers 20x faster transcription written in Rust.

Miscellaneous

Roblox’s new AI chatbot will help you build virtual worlds (3 minute read)

Roblox introduced an AI assistant to help creators design virtual experiences at the 2023 RDC. Set for release late this year or early next, the tool will enable advanced gameplay and 3D model generation.

Why Regulating AI Is So Difficult (17 minute read)

The desire to encourage AI development while limiting potential harm, combined with the breakneck pace of innovation, makes regulating AI an especially tricky situation.

Let’s Get Personal: Why AI Will Unlock a Massive Market for Online Consumer Services (4 minute read)

The idea of personalized services beyond what we have around shopping is really interesting. This post breaks down the opportunities and some of the companies already playing in this space.

New Benchmark Tests Speed Of Running AI Models (1 minute read)

MLCommons has unveiled a new benchmark for how quickly top-of-the-line hardware can run AI models, with Nvidia coming in first and Intel coming in second.

Merlin (Product)

Qualitative research conducted, analyzed, and summarized by AI.

If you have any comments or feedback, just respond to this email!

Thanks for reading,

Brian Hanson.