What could go wrong? Redmond researchers release a blabbering bot trained on Reddit chats

Microsoft researchers have built a chatbot from OpenAI’s text-generating model GPT-2, and trained it on millions of conversations scraped from... oh crap. Reddit. Teaching software to talk using chatter lifted from internet forums, such as Reddit, is risky. What can start out as an innocent discussion can quickly descend into …

[Mike Lewis]

Easier than the Turing test?

Now we have the DIALOGPT test to determine whether or not a computer is capable of thinking like a Redditor.

[404]

Re: Easier than the Turing test?

Typical Redditor or the Reddit Hive Mind? They're the same.

I think the results might be interesting to see if the bot gets buried or will glom on to the hive to be accepted.

[getHandle]

Interesting but...

Mostly harmless pointless.

[bombastic bob]

Re: Interesting but...

ack. Although I admit liking [to laugh at] that AI-written Eurovision-based song that was featured in an El Reg article a while back, I don't think having a conversation with an AI bot [especially in a worthless forum like Reddit] would be even mildly amusing...

[Anonymous Coward]

Ok...

So the robot has been trained to upload memes and shitpost.

Given that it probably won't do either of those well we should dub this thing "NormieTron 3000".

[JetSetJim]

Re: Ok...

I think they should train it on (Monty) Python scripts to come up with new SFW insults on a par with "Your mother was a hamster and your father smelt of elderberries!"

[bombastic bob]

Re: Ok...

I kinda like the sequence "vacuous, toffee-nosed, malodorous pervert!" Sounds better when Graham Chapman said it.

[Anonymous Coward]

Re: Ok...

All I know is, if the robot doesn't invoke Godwin's Law at some stage the experiment is a failure.

[Michael H.F. Wilkinson]

Oh, dear, I think you will find reality has gone on the blink again

I am just waiting for a bot that starts the conversation with: "I think you ought to know I am feeling very depressed"

Doffs hat (grey Tilley once more) to the late, great Douglas Adams

I'd better get me coat

[Chris G]

Re: Oh, dear, I think you will find reality has gone on the blink again

"SSiiiighh! I've got a brain the size of a planet and here I am making inane Reddit comments to myself.

“The first ten million years were the worst," said Marvin, "and the second ten million years, they were the worst too. The third ten million years I didn't enjoy at all. After that I went into a bit of a decline.”

[Totally not a Cylon]

A role for Reg commentators?

Perhaps a chatbot should be trained using what we write?

After all most of us seem to be actual humans......

Having opened that can of worms.....

[Anonymous Coward]

Re: A role for Reg commentators?

I wonder how far AI chatbot technology could be advanced if amanfrommars1 open sourced himself?

[Dr. G. Freeman]

"After all most of us seem to be actual humans...."

*Sob.*.. thank you....

That's the nicest thing anybody's ever said about me.

[Arthur the cat]

"make sure you're alive when you die"

My new goal in life.

[Teiwaz]

Re: "make sure you're alive when you die"

I'm thinking of releasing a new product line

'Schrodingers Coffin'

[Il'Geller]

Re: "make sure you're alive when you die"

I tried to sell "Eternal Life" product, as the creation of Lexical Clones (personal AI databases)... 15 years ago. I was not able to attract any attention then, though.

[IR]

Did they not bother to teach it how punctuation works?

[Il'Geller]

punctuation marks

Not that simple! Grammatical punctuation is responsible for emotional orientations of texts. But these emotions, in turn, are transmitted by the texts' subtexts, i.e. by what is meant implicitly, while only the texts' contexts are explicit.

I found out that the subtexts are 1) dictionary definitions on the texts' words and 2) synonymous clusters from related to the given other texts. That is, only by sifting the subtexts through the texts' contexts, it's possible to find out what the punctuation marks should be.

[IR]

Re: punctuation marks

I was thinking more along the lines of not adding a space before every comma, full stop, and question mark. But what you said is good too.

[Blackjack]

So...

How expensive it would be to train a chat bot using Sesame Street?

[Il'Geller]

Re: So...

Quite expensive because

1) you should annotate everything into Sesame Street with dictionary definitions,

2) find other texts related to Sesame Street texts, extract synonymous clusters from them, build blockchain relationships and annotate Sesame Street.

Only than you can say that Sesame Street is understood and computer becomes AI.

[bombastic bob]

"Gaining enough control to prevent offensive-output issues"

what a bunch of WUSSES if that's their biggest concern.

"Oh NOES, 'teh intarwebs' OFFENDED"

yeah, welcome to 'teh intarwebs'. Grow a thicker skin.

NOT having _that_ is recipe for BOREDOM. But then again IMMATURE (and INSECURE) people _LOVE_ having "disussions" with their fanboi clubs, filled with sycophantic yes-men and echo chambers, merely confirming what they WANT to *FEEL* on any subject.

And if that's all that the AI can do, then it's a COMPLETE waste of time.

[Bitsminer]

Garbage in...

trained it on millions of conversations scraped from... oh crap. Reddit

I note, without malice, they didn't use text from Microsoft support forums.