What I Read This Week…

BYD produced its first solid-state battery cells on a pilot production line, Bybit lost $1.5 billion in Ethereum to North Korean hackers, and researchers created the largest AI model for biology

Watch All-In E216

Read our Deep Dive: The Current State of AI and Follow-Up Conversation with Sina Sojoodi

Caught My Eye…

• Chinese automaker BYD has produced its first solid-state battery cells on a pilot production line. Unlike conventional lithium-ion batteries that use liquid electrolytes, solid-state batteries use solid materials to transport ions between electrodes, eliminating flammable components. This improves safety while enabling tighter component packaging, which increases energy density for longer vehicle range. The solid electrolytes also resist degradation during repeated charging cycles and withstand higher voltage differentials, both of which enable faster charging capabilities. BYD's achievement places them among frontrunners like Toyota, QuantumScape, and CATL in developing this technology. However, all manufacturers face challenges in scaling production. The industry struggles with manufacturing complexity, material costs, and ensuring consistent performance across large-scale production – obstacles that have repeatedly delayed commercial releases despite promising announcements. BYD plans a "mass demonstration" of the technology by 2027, with full commercial-scale production likely following after 2030. The company's strategy involves initially deploying these advanced batteries in high-end vehicles where premium pricing can offset higher production costs. Meanwhile, BYD will continue using their established lithium iron phosphate batteries for mainstream markets for 15 to 20 years as solid-state technology becomes more affordable.

• Singapore-based Bybit, one of crypto's largest exchanges, lost $1.5 billion in Ethereum to North Korean hackers in an attack targeting their cold storage wallet. What are "hot" and "cold" wallets? Crypto exchanges protect assets through two forms of storage: "hot wallets" that handle daily trading operations and user withdrawals, and "cold wallets" that secure larger reserves and require multiple authorized personnel to independently verify and sign any withdrawal. In Bybit's case, the attackers compromised the computers of Bybit's cold wallet signers via phishing emails and then manipulated what the signers saw on their screens during a routine transfer and redirected the funds to themselves. In other words, the attack succeeded not by breaking cryptographic protocols, but by exploiting the humans operating the protocols. Onchain investigators have traced the attack to the Lazarus Group, a North Korean hacking collective that has stolen over $3 billion in crypto since 2021. Of the stolen funds, approximately $550 million has allegedly been laundered through various techniques that obscure transaction trails. The remaining $950 million remains traceable in wallets linked to the attack, though these funds are likely to be gradually moved through other laundering channels. This breach exposes how even the most secure cryptographic systems can fail when attackers focus on deceiving the humans in charge rather than attacking the technology itself. While traditional banking fraud can be reversed, crypto's irreversible transactions means there's no way to claw back these funds.

• Researchers at Arc Institute, Stanford University, and NVIDIA have released Evo 2, the largest AI model created for biology to date. Evo 2 was trained on 128,000 genomes spanning the entire tree of life, from humans to single-celled bacteria and archaea, encompassing 9.3 trillion DNA letters. The model can write entire chromosomes and small genomes from scratch while also interpreting existing DNA. Previous AI models could only process and analyze short stretches of DNA at once, meaning they couldn't identify correlations between distant parts of the genome. Evo 2 was designed to overcome this limitation, as it can detect and model relationships between DNA segments separated by up to one million base pairs – a critical capability for understanding how distant parts of the genome work together to control gene function. To test the model's capabilities, researchers used it to predict the effects of mutations in the breast cancer-associated BRCA1 gene, achieving near state-of-the-art accuracy. The model's code is publicly accessible on Arc Institute's GitHub, making it the largest-scale, fully open-source AI model in biology to date.
  
  

Other Reading…

• Speaker Mike Johnson to Deliver His One-Bill Budget Plan (Politico)

• AI Algorithms Create More Efficient Wireless Chip Designs (Live Science)

• OpenAI's ChatGPT Surpasses 400M Weekly Users (VentureBeat)

• Alibaba Plans to Spend $53 Billion on AI Over Next Three Years (Bloomberg)

• Apple Will Add 20,000 US Jobs and Produce AI Servers in the U.S. (Bloomberg)

• Starbucks Laying Off 1,100 Jobs and Trim Menu by 30% (Axios)

• MSNBC In Process of Programming Overhaul (NYTimes)

• SEC drops investigation into NFT Marketplace OpenSea (Digital Watch Observatory)

• Studying Materials in Action (Chemistry World)

On X…