30 comments

[ 3.0 ms ] story [ 70.5 ms ] thread
Hi, To test it out just pick a random article from CNN, BBC etc and select the main sentence. This is an actual useful headline generator so it needs something substantial and real-ish else it doesnt generate anything.
(comment deleted)
Interesting idea. Is there a description of the tech behind it somewhere? Also, seemed to take a while to generate, not sure if it's getting stressed with visitors.
Yup. Brutally stressed. 32 people on it now. Running on a single k80. sorry about the delay guys!
Which framework/architecture did you use?
Suggest you spool up several more GPU servers and throw a load balancer in front.
Methinks it is not that easy for them :)
It still seems pretty stressed. I only wanted to see some examples; I did not want to enter fresh sentences myself. If you could add a section to your website to show previous queries and results across all simulations, that might be good enough for most visitors.
Sure. It's supposed to take 20 seconds to generate all the examples even normally. Anyways here's an example:

Sentence: The University of Notre Dame in Indiana is donating $100,000 toward the renovation of the landmark Notre Dame Cathedral in Paris, according to a statement post on the school’s website today.

Headlines: 1) Notre Dame Catholic University to renovate iconic Notre Dame Cathedral to $100,000

2) $100k donation for renovation of Notre Dame Cathedral in Paris

3) Notre Dame University to donate $100K toward renovation to new Notre Dame Cathedral in Paris

4) Notre Dame University funds ’restoration of historical historic building’

5) Notre Dame pledges $100,000 to refurbish Notre Dame Cathedral as part of renovation

6) Indiana University to contribute $100,000 to renovation of historic Notre Dame Cathedral

7) Indiana University pledges to support Notre Dame Cathedral restoration

8) University of Notre Dame renovates Notre Dame Cathedral in Paris

9) Notre Dame Catholic school funds renovation of Notre Dame Cathedral in the Parisian city

10) Notre Dame Cathedral in Paris donated $100,000 towards renovation

Please offer your honest feedback! Looking for applications!
if it can compare headlines and tell if an article is a click bait.
Have you considered satire as a possible use case?

Could be fun!

I'm not kidding. This created a fake headline "Trump on Obama 'He is an asshole'".

Had to google to be sure this hadn't happenned.

Some of us do not like that comment. I, on the other hand, think it's hilarious!

Keep having fun.

I picked the sentence "The S&P 500 and Nasdaq Composite had their highest closings on record Tuesday, marking a resurgence for the stock market after a rout last year hampered its climb." from the first article on WSJ and got some weird results

1) Wall Street Returns After a Year of Slowest

2) Wall Street Resilient on 2016 High Level of Return

3) New York Stock Market Haunts a Return

4) The S&P 500 Is Back, Again, on a Recovery in the Stock Market

5) Wall Street Revamps Wall Street

6) Federal and Securities Markets Leaders Clinch Over Wall Street’s Stale Trends

7) Financial Times Crashes Close of Shares, S&P 500 and Nasdaq as Largest to Close

8) Wall Street’s Longest-Hired Wall Street Sellers Set Back Year

- - -

On a separate note, I find the phrase "I made an AI for [x]" incredibly off-putting, but maybe that's just me...

Usually only 20%-30% of the results are useful at all. I would suggest using this for ideas only.
>On a separate note, I find the phrase "I made an AI for [x]" incredibly off-putting, but maybe that's just me...

Well, “I made a fancy statistical model for generating text” doesn’t quite have the same ring to it does it?

If only you had some AI that could generate a catchy headline
I like to think the latter would actually be better received on HN
Doesn't seem to work... i tried sentence: President Donald Trump met with Twitter CEO Jack Dorsey at the White House Tuesday to discuss the social media giant, which he has often accused of political bias..

And the CNN and Fox headlines were identical.

What did it return? This is just a demonstration. Use for quantity not quality
This can become a great source of satire I’m sure of it.
Absolutely. It's cute people are taking it very seriously.
Getting a 504 timeout
The page seems to load for me. Unfortunately didn't plan the launch too well, so some capacity issues were left unaddressed.
I can’t get access to the website. It’s under too much load. Could you describe the model a bit?
Given "Lyra McKee revealed plans to propose to her partner Sara just hours before she was murdered, mourners heard at her funeral", I get

Breitbart: 1) Former Breitbart News executive revealed plans to “marry her dead”

NYT: 1) People Talk of Murder in Twin Peaks

2) Survivors, Conducents

3) Reuniting Family in New NYC in Love Pitch

4) ‘Mother of Murder’, Murdered In New Jersey, Is Reported as Dead

Guardian: 1) Widow and Mother in Police Killing of New York Times Co. Reporter Tells of Past and Present

2) Dead Teen Photographer’s Love Affair Is Cited in Murder Investigation

I mean, I don't think it's working right...