100% automated ML doing this well seems far off to me.

Google's ML for rich snippets thinks a quarter is worth 50 cents. http://imgur.com/1nNreR2

Really parsing language for meaning is hard. Sentiment analysis, which seems simpler than this problem, still has relatively poor accuracy.

I could see it perhaps thinning out how much has to be done manually, but I don't see it removing that need entirely.

That's amazing. Nearly every fact presented there is completely wrong.

Plus, the picture is a standing liberty quarter that is nearly 90 years old and pretty much completely out of general circulation.

Could you specifically address his concerns for those not in-the-know about ML?

Sorry, my comment was ambiguous -- I was referring to the Google result. Thanks for pointing it out. The commenter that I was responding to made a great point that I agree with and was highlighting as a particularly good example of ML gone awry in the wild.

Other than the subjective statement that quarters are "useful", every fact in the google response is wrong.

Quarters are:

- Worth $0.25 (not $0.50)

- Not the largest. (Dollar coins and half-dollars are larger)

- Made of a blend of copper and nickel. (No silver content for circulating coins)

The picture depicts a standing liberty quarter that is no longer a circulating coin.

It's amazing that Google would put that in production.

I believe "Nearly every fact presented there is completely wrong" was referring to Google's snippet in the screenshot, not my comment.

I thought this screenshot may have been from an early and outdated rollout of the Rich Snippets, but no, I just searched and Google still says a quarter is worth 50 cents.

Easily explained as an artifact of just how accurate Google's ML systems really are.

What you're seeing is a hyper-accurate snapshot of the exchange rate between "a shave and a hair cut" and "two bits"[1]. Typically you hear about the value over a much longer period of time and the average value of "a U.S. Quarter Dollar" == $0.25.

/snark ;)

[1] "Two bits" -- Noun. Slang term used predominantly by Neo-Victorian communities living in California's Bay Area megalopolis.

Edit: Fixin' of the grammar

That information is so bad for such an easy question, I can only wonder if google is more interested in using their userbase as a training ground than actually providing a useful service.

It's not an outlier either. I can find about 1 obviously bad answer a minute just randomly typing fact type questions that pop into my head.

Last time facebook tried that they hired a bunch of lefties (collage kids are cheap labour) and they got pilloried for partisanship.

Are they kids that make artwork from newspaper and magazine clippings?

Do you have a source for this?

This report explicitly blacklisting and injecting news people weren't interested in. No mention of college students, so that is probably some bullshit I picked up somewhere, I wasn't following the story too closely.

https://www.gizmodo.com.au/2016/05/former-facebook-workers-w...

> These reviewers will also help us get better at removing things we don't allow on Facebook like hate speech and child exploitation. And we'll keep working with local community groups and law enforcement who are in the best position to help someone if they need it -- either because they're about to harm themselves, or because they're in danger from someone else.

The workers aren't there to be a basis for machine learning (though that is a nice benefit). They are there to communicate with local community groups / law enforcement on an ongoing basis.

Imagine the horror that awaits the 3000.

Do things that don't scale... Until you can scale. I think you are absolutely correct on this point.

This way you will have even less way to find a solution when you are hurt because of a false positive.

I mean, the GAFAs are already not known for their stellar customer service, but if we get rid of all humans in the process, it's going to be Brazil on steroid pretty soon.

You are describing the past. That work is already done. All of these "community moderation" jobs at big companies are just checking the AI's work. If you have a billion posts an hour you're not hand triaging that with a team of 4500 people.