The zdnet article is critical, but I think we should pay attention to what China is doing in AI.
Power optimization is a big problem. Neuromorphic computing essentially reduces a synapse operation from 32 bits to 1 bit, which reduces power by 1000 (assuming O(n^2) multiplication hardware) but you need more operations to make up for lost resolution. It's not appropriate for all applications. In ZPAQ I use 20 bit weights (+-8 with 16 bits after the decimal point) and 12 bit activation levels (ln p/(1/p) in +-8 with 8 bit fractional part) to produce 32 bit products. Then I discard the low 8 bits before summing, and discard the low 16 bits of the 32 bit sum. Any lower precision results in measurably less accuracy, although I could probably use probabilistic updates. On Sun, Aug 4, 2019, 6:45 PM Alan Grimes via AGI <agi@agi.topicbox.com> wrote: > Secretary of Trades wrote: > > When it's not even able to run a robot simulator, but gives ears to the > > wall and eyes to the bike, it's called an Artificially Junk Idiocy > > > > > https://www.zdnet.com/article/chinas-bizarre-chip-pumps-up-the-hype-in-a-i/ > > ugh... > > We need a governing committee and a standing army of mercinaries to go > after ppl who missuse AGI... > > I was perfectly happy to surrender AI but we need to stand our ground > somewhere. > > -- > Clowns feed off of funny money; > Funny money comes from the FED > so NO FED -> NO CLOWNS!!! > > Powers are not rights. > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T56143ffbc19522e8-M3aeecdab2328db5e32d54c7f Delivery options: https://agi.topicbox.com/groups/agi/subscription
