> But at the same time, it feels shortsighted, doesn’t it? If we can build software that can speak to other software more efficiently, shouldn’t we use that? Couldn’t there be some benefit?
Then why are we putzing about with strings in the first place? Let them use binary formats if that's the goal.
Somehow this doesn't seem like a desirable property in a language. One of the purposes of (human? inter-sentient?) languages is the reliable transmission of information. Verbosity and repetition in language have a purpose similar to that of redundancy in engineering: they ensure that if someone misses out a part of the message, the whole message isn't ruined.
If your message being get across is predicated on successfully transmitting an arbitrarily large sequence of symbols where every one of them might be crucial (supposing that word frequency is inversely proportional to meaningfulness, which is rather reasonable), this "talking" thing would get hard pretty fast.
> “Getting the data into a format that makes sense for machine learning is a huge undertaking right now and is more art than science. English is a very convoluted and complicated language and not at all amicable for machine learning.”
Is there an advantage to using less irregular human languages - say German - in machine learning?
German, where noun gender doesn’t make any sense? and things like dative case have to be specifically marked? German and English have the same lineage anyhow, both have irregular verbs (in many cases English irregular verbs are also irregular in German).
Analyzing any real language (not Esperanto) is going to be extremely difficult. I would think English is marginally easier than say Korean where everything is based around context, including dropping the subject itself and particles. Otoh, some aspects of Korean are probably easier than English. And all 3 have irregularities despite Sejong the Great purposefully creating Korean Hangul.
Facebook is no where near the level of what Google has accomplished and you're comparing them with OpenAI, a non-profit. A multi-billion dollar company competing on the same level as a non-profit. Facebook own AI research site lists 4 projects, not including a broken chat bot. How could I have possibly arrived at that conclusion?
What a junk article. Link bait. "Humans can't understand" is ambiguous, and the article wants you to believe this means "above human comprehension," when really it just means "makes up some new language that we could absolutely reverse engineer."
I submitted, so it's my fault. I ignored the title and thought the content itself was interesting. That AI appeared to be communicating with a method it figured out itself.
That's what interested me enough to want to share it here.
I'll use more discretion next time i submit something.
Here's what you need to do before publishing an article on AI:
1. Hire an expert that actually knows this stuff
2. Have him read the article
3. Throw the article out when he tells you to throw it out
4. Feel shame
Stop writing these imbecilic, clickbaity shit articles about AI. This leaves people that don't know anything about AI or neural networks thinking that we're somehow developing skynet, instead of just writing a computer program which essentially implements a mathematical model that continually adjusts itself based on some metric.
15 comments
[ 3.2 ms ] story [ 47.7 ms ] threadThen why are we putzing about with strings in the first place? Let them use binary formats if that's the goal.
If your message being get across is predicated on successfully transmitting an arbitrarily large sequence of symbols where every one of them might be crucial (supposing that word frequency is inversely proportional to meaningfulness, which is rather reasonable), this "talking" thing would get hard pretty fast.
Is there an advantage to using less irregular human languages - say German - in machine learning?
Analyzing any real language (not Esperanto) is going to be extremely difficult. I would think English is marginally easier than say Korean where everything is based around context, including dropping the subject itself and particles. Otoh, some aspects of Korean are probably easier than English. And all 3 have irregularities despite Sejong the Great purposefully creating Korean Hangul.
https://research.fb.com/projects/
I still assert that a relative black box will lie at the core of any "strong" AI.
That's what interested me enough to want to share it here.
I'll use more discretion next time i submit something.
Here's what you need to do before publishing an article on AI:
1. Hire an expert that actually knows this stuff 2. Have him read the article 3. Throw the article out when he tells you to throw it out 4. Feel shame
Stop writing these imbecilic, clickbaity shit articles about AI. This leaves people that don't know anything about AI or neural networks thinking that we're somehow developing skynet, instead of just writing a computer program which essentially implements a mathematical model that continually adjusts itself based on some metric.