Books are fair use
Source: The Verge on Anthropic ruling
The federal court ruled Anthropic's use of books for LLM training is fair use. The judge called the process “sufficiently transformative”, and added that the Copyright Act “seeks to advance original works of authorship, not to protect authors against competition.” However, he also noted that Anthropic’s choice to store the books in a central repository is not fair use, and that the court will hold a separate trial to assess liability for pirated content.
Despite storage liability concerns, this ruling is a significant win for all AI-related companies. The downstream effects of a negative result could have been staggering, and the court's decision represents an important attempt to understand the technology at hand instead of implementing blanket bans. The ruling essentially creates a blueprint: training gets protection, but data acquisition remains legally exposed.
I wonder how this will affect compliance going forward. Will AI labs require lawyers overseeing data pipelines? How will companies setup guardrails to protect against violations? Will new forms of compliance architecture emerge? Will companies handle compliance in-house or outsource?
Tesla proves scale beats perfection (maybe not)
Source: CNBC + The Guardian Tesla robotaxi coverage
Tesla finally debuted robotaxis in Austin, Texas last week. Initial press was all positive - the stock jumped 10% the following day. However, as the launch progressed, the outlook darkened with videos revealing traffic violations and operational issues.
Tesla’s approach is ambitious - a self-driving car that can be deployed anywhere, by anyone, with no specialized equipment or extensive infrastructure. Waymo, on the other hand, spent 15+ years mapping every intersection, building intricate safety systems, and creating purpose-built hardware. I frequently take Waymos in SF, and the experience is nothing short of fantastic.
Despite Tesla’s continued issues, the implications of an operational robotaxi are significant. If Tesla succeeds, transportation will change forever. Overnight, Tesla could deploy thousands of vehicles, create an entirely new market, enable everyday owners to pay off their own cars, and completely disrupt the car industry.
Republic's private market arbitrage
Source: Republic token launch
Republic is offering "SpaceX exposure" without SpaceX approval through tokenized claims on future performance. They're bypassing accredited investor requirements, minimum investments, and company consent entirely.
The structure is interesting. You're essentially buying options on potential performance, not actual equity. I imagine the premiums will be massive - retail investors will pay 2-3x institutional rates for inferior downside protection.
I spent almost 5 months going deep on RWA (real-world asset) tokenization. It’s a fascinating space with a lot of opportunity, but it’s extremely regulation heavy. How did Republic structure this to avoid securities regulations? Is this crypto? Derivatives? Something entirely new? Private market access is important, and tokenization may solve the liquidity issue, but Republic’s execution screams ‘let’s concentrate the risk on the least sophisticated investors.’ We've seen this movie before. I’ll be following. And judging.
Other commentary
Quantity > quality for general purpose LLM training (Source: Clement Delangue, CEO - HuggingFace)
Jane Street funded a coup in South Sudan. lol.
Obviously Harvard never have the juice, but what an embarrassing L for Jane Street that they couldn’t even successfully orchestrate a coup in a midsize African country with a deeply unstable government
#ThingsYouNeverWantToHaveToSay
