Don_Patrick

I don't imagine that minor code optimisations would make as much difference as shortening the time the user spends on their device.

The goal would then be to make websites as concise as possible, toss out all ads, comment sections, social media tie-ins, videos, needlessly long-winded articles, anything that slows the visit down, and kick the user out as soon as they have what they were searching for.

But capitalism doesn't approve of that.

XD ...I had not thought of it that way yet. It's because the bot itself is naive.

I expect so. As a next-word prediction algorithm, it knows the probability of "red big" occurring in that order is very low.

This is a very interesting test. It seems to me that, aside from the prompt bias, it may be detecting "yo be real" category questions by the low statistical probability of the words occurring in sequence. Those would be especially low for gibberish words that had an occurrence of zero in its data, which is why it's so good at spotting those.

One could also try and detect nonsense with quad-grams using the same principle, but in any case this metric would tend to mark unique questions and foreign names or brandnames as nonsense.

So far most of this pollution is effectively tagged through accompanying mentions of "GPT", but filtering that from future training data would mean GPT can never learn about itself.

I know people who talk like this, and the user history checks out. Either way, the comment is misinforming: Those quotes are from Descartes, not Turing, whose views as to whether machines could think were opposite. https://plato.stanford.edu/entries/turing-test

I don't think it's just coincidence. A lot of "A or B" questions tend to have B as the correct answer with the first option just being there to confuse the reader. I find it likely that a statistical algorithm would pick up on that.

Good observation. It is possible that GPT picked up on this statistical tendency. I participated in seven Turing tests, and most of the "X or Y" questions had the latter as the correct answer, so regularly that I set my program to pick it as default when it didn't know. As GPT picks up on statistical word orders, I find this likely to be the case here as well.

Turing only proposed his game as a hypothetical argument, so he barely specified any rules on how to actually perform such a test in practice. He referred to the judging party either as "the man" or "the average interrogator", he never specified a number. Official Turing test events have had anywhere between four to hundreds of people judging.

https://artistdetective.wordpress.com

I write about Artificial Intelligence, Turing tests, language processing, and a bit of robotics, for a layperson audience.

You should read it if:

- you enjoy critical commentary on popular AI news and practices,

- want to know AI myths from reality, or

- want to see a participant's view on Turing tests.

If this is what your self-driving car asks itself when approaching a ZEBRA CROSSING, you've got bigger issues than just ethics.

Neural networks as usual. What I make of it is that the programs noticed that some words had more effect than others, and just started spamming those for maximum value. Source: https://code.fb.com/ml-applications/deal-or-no-deal-training...

The command is "Alexa, laugh". Audio-wise, a mere cough could be interpreted as "laugh", that's why Amazon's solution is to change the command into something more pronounced.

Other unprompted activations have been reported in abundance when Amazon installed chatbot functionality after the Alexa Prize: Alexa would start having conversations with herself in the middle of the night. One can expect bugs for newly added features, and of course the creepy ones are more likely to make the news than any random glitch.

Thanks for posting (author here). In retrospect I should have written "A.I. news" in the title, I hadn't expected to see it posted on boards with more general topics.

I participated in the WSC with GOFAI, though I am not representative of the state of the art. I would recommend GOFAI logic combined with a knowledge database assembled through machine learning (my database was virtually empty). However I found the main problem in this challenge to be that one had to solve all aspects of language processing before one could begin to solve the pronouns. My home-made parser just wasn't up for processing relative clauses of relative subclauses. https://artistdetective.wordpress.com/2016/07/31/winograd-sc...

The highest scoring entry used a deep neural network: https://www.cc.gatech.edu/~alanwags/DLAI2016/(Liu+)%20IJCAI-... I'd say it's not doing great because deep learning here merely observes the statistical co-occurrence of words, and hence are only correct insofar as the sentences are common. I'm sure with more data they could get a good 75% of it right. More of a problem is that pure textual comparison is oblivious to the underlying logic. Sometimes the verbs may correlate as a common sequence of events, but a preposition like "in" denotes a logical physical constraint that should overrule the mere correlation of verbs.