Xinyu Zeng, a PhD researcher, introduces F3, a new file format aimed at overcoming the limitations of existing formats like Parquet and ORC. He highlights several issues with these widely used formats, such as CPU-bound decoding, excessive metadata overhead for wide-table projections, and inadequate random-access capabilities for machine learning tasks. F3 tackles these problems with a reimagined approach to data layout and encoding, focusing on efficiency, interoperability, and extensibility. Two core features define F3: its decoupled layout that separates IO units, dictionary scope, and encoding choices, and the self-decoding capability that incorporates WebAssembly (WASM) kernels directly into files. This allows for easier updates and adoption of new encoding methods without requiring all processing engines to be upgraded simultaneously. Zeng also discusses the need to decouple table formats from file formats, suggesting that this separation can enhance performance and adaptability. He envisions future developments for F3, including expanding WASM applications beyond encodings to functions like indexing and filtering, which could further enhance the format's utility in data lakes.