#350: Crypto Lender Genesis Global Files For Bankruptcy, & More

Thursday night, Genesis Global—a subsidiary of Digital Currency Group (DCG)—filed for bankruptcy, joining a list including BlockFi, FTX, Celsius, and Voyager. The filing comes after Genesis laid off 30% of its staff and yet failed to secure funding. According to the filing, the bankruptcy impacts Genesis’s lending business, not its trading and custody businesses.

Genesis owes its top 50 creditors more than $3.6 billion, including $766 million to Gemini Trust, its largest creditor. Other creditors include Miranda Corp, Babel Finance, VanEck, Cumberland DRW, and Abra.

According to interim CEO Derar Islim, Genesis hopes to exit bankruptcy as quickly and efficiently as possible. Under a Chapter 11 bankruptcy, an “automatic stay” prevents Genesis from satisfying any creditor claims until the court determines the claims process.

Gemini CEO Cameron Winkelvoss has raised several questions that the bankruptcy process should answer: Did Genesis misrepresent the 10-year promissory note from DCG as a “Current Asset”—a category which applies only to cash, cash equivalents, and other assets exchangeable into cash within one year? Did DCG commingle assets or offer sweetheart deals when borrowing ~$500 million from Genesis to purchase GBTC shares?

Based on the 5% jump in bitcoin’s price after the announcement, the market already seems to have discounted Genesis’s bankruptcy. Although the fallout could impact Gemini and others, the crypto-related deleveraging that started after the Terra/Luna collapse more than six months ago does seem to have run its course.

Last week, Nature published a paper in which scientists pushed back on the practice of listing ChatGPT as an author on academic publications, and journal editors offered guidelines for proper citation. In ARK’s view, while the use of ChatGPT in text generation, classification, and translation is likely to revolutionize projects and tasks, from grant writing to scientific communication, accurate and proper sourcing will be crucial to credibility.

Last week, after a year of testing, Microsoft rolled out its Azure OpenAI service. Offering a variety of AI models, including GPT-3.5, Codex, and DALL-E, Azure OpenAI will enable developers to integrate them into their own products.

After Microsoft invested $1 billion in 2019, the partnership evolved and strengthened, ultimately incorporating OpenAI’s models into its own products. Azure OpenAI, for example, powers GitHub Copilot. Now, rumors of another $10 billion investment suggest that Microsoft is gearing up aggressively for the AI race.

In our view, while foundation models like those trained by OpenAI will serve a wide range of use cases, smaller models trained on proprietary data have advantages and will enjoy significant adoption as well. More computationally efficient, and therefore faster to train and deploy, smaller models can enable real-time applications on devices with limited resources. Moreover, self-trained models do not need to share data with model providers, an important consideration for competitive data like patent records. Those advantages do come at a performance cost relative to larger foundation models in certain use cases, like general text summaries. In other words, the tradeoff between model size and proprietary data is likely to depend importantly on performance demands and the competitive sensitivity of data.