Skip to content

Training Datasets

Feedly is a good product – a worthy replacement for the (still-missed) Google RSS reader. It has a ton of capacities for free, and does not badger you into upgrades. As such I’m very inclined to want to give back by helping out with feedback.

In an article by Fred Wilson about the need for corporate boards to have executive sessions I was presented with this as a post-script:

Feedly calls this capacity “Leo.” It’s strange that the most elementary disambiguation is still not done straight out of the box, or even with more than a month of use. The word “Jeff” is not anywhere, nor is “Attorney General” or “Trump” or any other common phrase disproportionately associated with “Sessions.”

Feedly help shows a bunch of other users from a month ago ticked off about this coda to their RSS, and they’ve enabled an option to turn it off. The help further says they’ve trained a dataset already. I doubt it.

(The simple click to say this was “Other” doesn’t do anything further.) I’d like to help, but freemium products need to do a little more heavy lifting first, before I’d like to help put them over the top.

That well-managed startups with good product management (they turned around a fix quickly to enable turning this off for the aggrieved, likely still free users) have such a steep NLP hill to climb underscores what an enormous advantage Google, Amazon, and Apple are building with their exhaustively large trained datasets. Feedly confesses even they’re really just working on English right now. It’s almost unimaginable a startup could catch up in a field like voice recognition.

Published inBusiness

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.