I was curious why Amazon can't develop anything that holds a candle to ChatGPT despite 5B+ year over year losses on Alexa and 10k engineers. I used the phonetool to check one of the NLP teams and I google searched the names to see what papers our scientists have... and I a peasant SWE had a first author paper at a better conference during my MS. Does Amazon not care about academic credentials? You can hire 10k engineers but what does it matter if the models are crap? Now look up OpenAI scientists on LinkedIn and they have presented several papers at NIPS, IMCL, ICLR, etc
Bias for action, which is otherwise a beautiful principle for business outcomes is a nemesis of research. You can't attract and retain research level talent by putting WBR/MBR targets.
It’s probably the shorter term focus of public companies that prevents longer term research.
I don’t think “bad models” is the reason alexa loses money. Hardware biz was subsidized heavily hoping that customers will buy more with voice which is a ridiculous idea. I do not trust a voice assistant to do anything except turning on lights (which is also fails 1 in 10 times). Your product managers are to blame for the greedy monetization. A good PM would have never approved “by the way”. I ditched my echo devices the day they started talking back to me unprompted and switched to nest
But this is shortsighted thinking. Some day voice assistants will become good enough for you to trust them, and that day isn’t far. At that point amazon will have a buying kiosk in almost every other house front and center The queries might be for items that do not need much search or discovery but it indeed will give amazon sales a lift. It’s a great long term asset
There's a meme of a young girl talking to her siri like a real person and holding a whole conversation. I feel like if Alexa brought out a tutoring feature for homework - it would increase utilitarian uses inside the household.
If you keep people incessantly busy they lose creativity, it’s that simple.
It’s a sampling bias. However , I do agree we have enough talent to be able to build something like ChatGPT. In my opinion it’s messed up organization structure and focus on fade projects that create the buzz but fail to deliver.
No one wants to work there. It’s tempting sometimes because of FOMO about cash but that’s about it. People who have talent work at either Google or FAIR afaik
Not as familiar as OpenAI but you may be right.