#311: Sequencing The First Complete Human Genome Heralds A New Era For Human Genetics, & More

In ARK’s Big Ideas 2022 report, we highlighted how long-read sequencing (LRS) helped generate the first complete human genome. This week, Science magazine released a peer-reviewed special issue with several articles not only about the first complete human genome but also about epigenetic changes hidden in the previously “dark genome” and open-source tools to turbocharge scientific discovery. As we delve into this new body of work, ARK plans to publish a more comprehensive analysis of how these discoveries could alter the course of human genetics.

Last week, we learned that the Ronin Network, an Ethereum-linked sidechain launched to support the popular play-and-earn game Axie Infinity, was exploited for 173,000 ETH and $25.5 million USDC. At $620 million USD, this hack is the largest in DeFi history.

Unlike the Wormhole hack which exploited a bug in Solana’s code in February, hackers took advantage of Ronin’s security validation and absconded with “private keys in order to forge fake withdrawals”. The Ronin network relies on a 5-of-9 multi-signature scheme to validate ether deposits into and withdrawals from the network. During the hack, the perpetrators gained access to the private keys of five validators by finding “a backdoor through [their] gas-free RPC node,” and gained control of all the funds locked in the Ronin network.

Nearly a week passed before developers learned of the hack when a user’s attempt to withdraw 5,000 ether from the treasury failed. As the developer team works with law enforcement, deposits into and withdrawals from Ronin Bridge will not be possible.

As ARK has highlighted, cross-chain attacks should spur interest in alternative scaling solutions like layer 2 rollups and blockchain networks designed for interoperability like Polkadot and Cosmos. They also highlight the resilience of blockchains with more narrow use cases, like Bitcoin.

At its bi-annual GTC conference, Nvidia unveiled not only next generation datacenter hardware that will accelerate artificial-intelligence (AI) computing but also updates to its Enterprise AI and Omniverse software suites. The star of the show was the “Hopper” architecture underlying H100, Nvidia’s new datacenter AI accelerator chip that will boost performance 9x when training large AI models. H100’s performance gains stem from several technological advances, including:

• a new Transformer Engine tuned for faster calculations when training AI models
• faster GPU-to-GPU and node-to-node communication, enabled by updated NVLink and NVSwitch interconnect systems, to scale-out compute cluster for training the largest models

In what Nvidia claims could be the largest performance leap ever, the H100 could accelerate the decline in AI training hardware costs from, according to ARK’s research, 43% per year to the mid- to high-40% range.

Nvidia also is pushing its business increasingly into the AI software stack with Enterprise AI, giving customers access to machine learning frameworks and operational tools to deploy AI models optimized for its hardware. Its Omniverse platform for building virtual worlds collaboratively also is gaining traction with 105,000 active users and 82 connectors to third party tools, the latter up from 8 at its launch last year. Early adopters of Omniverse include BMW, Amazon Robotics, and Siemens.

During an investor event after the GTC keynote, management highlighted that Nvidia is well-positioned for an AI hardware and software opportunity that could scale to a $1 trillion addressable market.