Keith McCready

Even then if we go from 90% to 95% correct in 26 years, (half the error) - how long from 95% to 99% (5x less error!)

But all the models (older Hidden Markov Model and new transformer LLM) have the same “Zipfian” problem - huge “tail” of low-frequency tokens & contexts

I think more likely is differing standards of measurement - more specific: measuring over a limited domain of vocabulary vs. a larger domain.

I think we all too easily forget the fundamental mathematics. 90% correct vs 99% correct means a 10x reduction in error.
Kurzweil claimed 97% accuracy in the limited domain of ASR in 1998. What is the best Word Error Rate (WER) in ASR today? 🤔

Reminds me of the late 1990s

Ray Kurzweil said this Feb 1998:

“Computers have gradually moved towards human ways of communicating. I expect we will see over the next 12 to 24 months this technology will become relatively ubiquitous.”

web.archive.org/web/19990225...

Sharing this link because of your other “status” postings - and also so we can keep our eye on everything flowing Westward under the Mackinac Bridge into Chicago 🙂 www.youtube.com/live/qBH-cer...

No one told you that Chicagoans love their Malort in much the same way as New Yorkers love their Pizza Rats?

I *presume* the pun is good

ORD -> SFO

Screenshot: they put Christmas sweater on the t-rex at the national history museum in London and it's ridiculous and adorable

Aww. *waves tiny arms*

earthquake.usgs.gov/earthquakes/...

Got this…but I am in Chicago currently

earthquake.usgs.gov/earthquakes/...

earthquake.usgs.gov/earthquakes/...

There appears to be one metric fuckton of bots on here saying Jimmy Carter pardoned his brother Billy and George HW Bush pardoned his son Neil. Those didn’t happen.

Biden should have pardoned Hunter and I’m glad he did, but there’s no need to lie about nonexistent pardons to justify it.

Slate

Sandhill cranes divert for air traffic

Left rainy SFO for snowy ORD