John L. Gustafson's claims for a more convenient variable-precision numeric format than floating point are compelling. I might have to write an implementation and see how it goes in practice.
Also, as a scientist, interesting to see how the formats and arguments have evolved in the last 5 years. This guy must have had some interesting post-conference bar talks with hardware engineers.
@baseballpunt They seeeeem to promise numerical operations that are a bit easier to reason about and more stable/reproducable, over a wider range of values.
And you don't have to worry about NaN, which is always nice. Though whether it's *desirable* is debatable.
@icefox What makes them so interesting? I've only heard of replacement for floats recently