Deep Models Continue to Set Benchmarks

Estimated Reading Time: 4 minutes

TL;DR

We cover interesting news from around the web, in particular Google’s new Minerva model for mathematical problem solving and DeepMind’s impressive new Stratego game playing engine.

Interesting Links from Around the Web

• https://www.quantamagazine.org/the-scandalous-history-of-the-cubic-formula-20220630/: A wonderful read about mathematical history.

• https://www.cnbc.com/2022/06/29/fcc-commissioner-tells-apple-google-to-remove-tiktok-from-app-stores.html: One of the FCC’s commissioners is calling for Apple and Google to remove TikTok from app stores. TikTok is a national security threat and should be regulated as such.

• https://www.anandtech.com/show/17474/samsung-starts-3nm-production-the-gaafet-era-begins: Samsung has started production of its 3nm node, an industry first. High volume production has not yet started though.

• https://ai.googleblog.com/2022/06/minerva-solving-quantitative-reasoning.html: Google’s new Minerva system achieves astounding performance on the challenging MATH benchmark suite. The reasoning performed is quite sophisticated as the image below shows. Minerva is built on Google’s PaLM 540B parameter model and provides compelling additional evidence for the scale hypothesis. Minerva achieves dramatic improvement over the previous state of the art. Most observers of the field would not have expected dramatic progress on mathematical reasoning tasks in such a short time frame

• https://arxiv.org/abs/2206.15378: DeepMind’s most recent AI system DeepNash has learned to play the strategy game Stratego at the level of a human expert. Stratego requires complex decision making under uncertainty and until recently has been resistant to AI solvers. 

Parting Thoughts

Progress on Minerva and Stratego provides additional compelling evidence that large models bring something qualitatively new to AI research. These models are dramatically harder to train than older AI advances (such as AlexNet or ResNet) though, which may limit their broader impact outside giant industrial labs until hardware catches up.

Feedback and Comments

Please feel free to email me directly (bharath@deepforestsci.com) with your feedback and comments! 

About

Deep Into the Forest is a newsletter by Deep Forest Sciences, Inc. We’re a deep tech R&D company building Chiron, an AI-powered scientific discovery engine. Deep Forest Sciences leads the development of the open source DeepChem ecosystem. Partner with us to apply our foundational AI technologies to hard real-world problems. Get in touch with us at partnerships@deepforestsci.com! 

Credits

Author: Bharath Ramsundar, Ph.D.

Editor: Sandya Subramanian, Ph.D.