When an AI Tries Writing Slashdot Headlines (tumblr.com)
She trained it separately on the first decade of Slashdot headlines -- 1997 through 2007 -- as well as the second decade from 2008 to the present, and then re-ran the entire experiment using the whole collection of every headline from the last 20 years. Among the remarkable machine-generated headlines?
- Microsoft To Develop Programming Law
- More Pong Users for Kernel Project
- New Company Revises Super-Things For Problems
- Steve Jobs To Be Good
But that was just the beginning...
Those five headlines were all derived from the first decade, but it's really nice to see that Steve Jobs made it into both decades. When training on the second set of 82,871 headlines from Slashdot's second decade, the neural network began envisioning the co-founder of Apple tackling even greater challenges.
- Steve Jobs Allowed To Deal With Solar Power
- Steve Jobs Sues Death of the Future
The neural network "did its best to reflect the new topics of the last decade," Janelle writes, adding "Compared to the late 1990s and early 2000s, some companies and topics disappeared, while the coverage of Apple in particular exploded."
But Sun Microsystems also founds its way into several headlines -- especially when Janelle tried to create the "essential" Slashdot headline using the whole 20-year set.
- Sun Sues Open Source Project Content
- Sun Sues New Star Trek To Stop The Math
And as technology continues changing our world, Sun isn't the only company that the neural network saw pushing for new rights in court.
- Sony Sues Apple Server For Seconds Off From SpaceX Project
- Apple Sues Apple To Start The Solar Power Project
Janelle will send you four more pages of machine-generated Slashdot headlines if you subscribe to her blog's announcement list. But after savoring the whole surreal AI-enabled look at the last 20 years, these four headlines were still my favorites:
- Red Hat Releases Linux Games And Moon
- Why Open Source Power Man Sues Java
- Microsoft Releases New Months
- Ask Slashdot: Do We Want To Be the Computers?
If you train your algorithm with whatever raw data, you would get whatever result. Even a model perfectly analysing the given situation becomes useless when not being adequately trained. In this specific case, the problem is clear: that tool was designed to deal with a different type of scenarios. Coming up with names for objects by training the program with many other names of equivalent objects makes perfect sense. Trying to figure out the best title for an article by analysing a big number of past titles about different subjects makes no sense at all.
The only sensible proceeding in this specific case would have been to rely on a tool able to reasonably analyse article contents and accurately determine the associated title; also to analyse a big amount of contents and output a good summary for them. You train that tool with all the articles during the last years, such that it can come up with the best summary and generate a title from that summary. If they did that, the training might have been considered acceptably good and the accuracy of the used model might have been properly assessed. Under the current conditions, these results don't differ much from the generation of random words.
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
Because, well, what they show is what topics really dominate on /., because what does finding the "ultimate" headline really mean? It means that it finds what terms, products, people and so on are found the most in /. headlines. It's pretty much a popularity contest. And what do we get?
Company-wise we get MS, Sun and Apple. Which makes sense. I'm glad to not see SCO anywhere anymore, that used to dominate the headlines a few years back.
People-wise all we get is Jobs. Really? He's the quintessential poster child for our headlines? Not Billy? Not Ballmer? I am not so deluded anymore that it would be Turing or someone important, but couldn't it at least be Stallman? Of all the people that shape the IT world, it really is Jobs? And that guy is dead, unlike the rest of them!
And content-wise? Lawsuits, mostly. And patents. A bit open source, a bit Star Wars, a bit trivialities. Seriously, one could think we're on a board for lawyers and law geeks, not techs.
And this, ladies and gentlemen, sums up what's wrong here.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
The best way to tell is to look at the grammar. If it is unnatural, with weird syntax and and obvious spelling errors, then it was one of the editors.