yhjc2692

What would be a better name?

This is something that can be addressed by file managers - showing a metadata entry on whether it's lossy or lossless. File extension (along with the bifurcated mainstream adoption of JPEG and PNG) has been a great proxy to do this, but like others noted, can be deceptive, which is arguably worse.

Also curious. Especially since handwritten (or worse, LLM-written) assembly is somewhat on the opposite direction of provably memory-safe

quite the delightful maintenance too

Very much agreed

Extremely cool stuff. I’m excited for neural approaches to further improve the bitrate frontier of “visually lossless” compression

Matrix multiply-accumulate operations, possibly at reduced precision (8- or 16-bit). The most common name for this is NPU but the description applies to other things too (eg NVIDIA’s tensor cores, Google’s TPUs, Apple’s ANE to a lesser extent, etc)

Evidence?

Great food for thought. Thank you.

Afaik mobile electronics’ power circuitry is designed so that power goes straight from charging port to the processor, without constantly charging/discharging the battery, once charging is complete. Unless the system suddenly draws a lot of power (more than the charger is supplying) - which is unlikely in a smartphone.

Interestingly it didn't seem to happen with cars

On latency: that is true now but likely false in the near future. There is so much potential left over for optimizing inference further.

On PID: the field has been stuck trying to do analytical/optimization-based control for decades, and end-to-end control has shown incredible performance (e.g., SoTA cost of transport in legged locomotion) and robustness (e.g., not falling over when stepping on a pile of leaves) - while being far scalable (in terms of how fast it is to get a new robot up and running). Which is not to say it's perfect but it seems like it's a step in the right direction.

Better yet, companies like Physical Intelligence are doing good with hierarchical ("fast-slow") architecture to address both the intelligence and latency fronts.

falling down and breaking things?

I do think robotics would come up with more safety mechanisms (provably safe motion planning etc) just because the risk is a lot more serious than LLMs spitting half-truths

must?

This is an extremely good advice - thank you

This is super interesting... thanks for sharing!

Why is that?

Thank you for sharing this awesome project!!!

Woah this is so cool - awesome job!!