Pinned
LFG! That's some nice and intuitive DevEx in @isographlabs
Robert Balicki (👀 @IsographLabs)
@StatisticsFTW
@isographlabs framework author. Ex-React Data Team @Facebook. Co-organizer of #RustNYC. I like Rust, Relay, stats, GraphQL, React, JS
- Several watermark avoidance schemes that come to mind: - ask the LLM to generate text with a specific pattern e.g. interspersing N random words between every real word (hello dog jentacular swanky world), or another such pattern. Then, reverse that pattern. Don't disclose thatseangoedecke.com/text-ai-waterm… watermarking seems hopeless
- with good data structures and guarantees enforced by the type system, I can spend a lot less effort reviewing the implementation. so cool
- Random speculation. Claude seems to accurately reproduce files on disk. Others have noticed that Claude doesn't know that it rewrites output to apply a watermark. So, perhaps a simple workaround to watermarking is to ask the agent to write messages to disk then to print those 🤷♂️
