Sujet : Re: If you're a fucking moron
De : bowman (at) *nospam* montana.com (rbowman)
Groupes : comp.os.linux.advocacyDate : 10. Oct 2024, 22:38:47
Autres entêtes
Message-ID : <lmqvr7FiegjU1@mid.individual.net>
References : 1 2 3 4 5
User-Agent : Pan/0.149 (Bellevue; 4c157ba)
On Thu, 10 Oct 2024 18:00:07 -0000 (UTC), candycanearter07 wrote:
AI, or more properly, pattern matching, can be quite valuable in image
sharpening, background removal, and other image processing tasks.
Agreed, and /some/ tedious tasks, like captioning.
Unfortunately image recognition starts with the tedious task of
captioning. The 'hello world' of image recognition is a huge dataset of
dogs and cats, each labeled 'dog' or 'cat'. Guess who does the labeling.
Voice recognition is as bad if not worse. There are a variety of accents
to deal with as well as languages. A real problem is the very skewed
nature of the available data. English has the most data while Tagalog data
is sparse/
If you have some spare time:
https://commonvoice.mozilla.org/enThe 'listen' side is revealing. You're shown the words and listens to the
speaker's pronunciation and score it. Some people try really hard but what
they produce is unintelligible. The 'speak' side can be humbling. I
wouldn't make a good politician since it's apparent I can't read from a
teleprompter :) At least you get a redo.
There are unsupervised learning techniques that do not require massive
amounts of stat but they seldom are in the spotlight. An example would be
a system receiving sound, temperature, vibration, and other data from a
car engine. In time it learns the baseline parameters of a healthy engine
and can raise a red flag when you're getting ready to drop a rod.