What file/directory should we state solutions in?

We have already worked on dqn+RAM (http://arxiv.org/abs/1605.01335).

I'm trying to download the dataset for Description2Code and the link provided only has one of the 9000 examples promised.

Hey there,

No content seems to be rendering on the right-hand side of the page at https://openai.com/requests-for-research/ in either Chrome or Safari on OS X.

I cloned things down locally, ran make, and content seemed to render, so I'm imagining this is something on the server operations side.

Hi!

I recently tried Train a language model on a jokes corpus research and for that purpose I've crawled few jokes sites and trained char-rnn language model on that. However, the output wasn't really funny.

Here's the dataset that I've used. However, please note that some of these jokes are really creepy. I tried to filter some of them out, but... well, it's just crawled from jokes site, can't do anything about it.

Hello @gdb and @joschu,

I wonder whether you have considered adding project on intrinsic motivation. Here are a few papers that have stood out to me:

1. Causal Entropic Forces
2. Empowerment introduction
3. Variational Empowerment
4. Unsupervised Real-time Control through Variational Empowerment
5. Unsupervised learning of Goal Spaces

I'm currently working in this area and have worked on practical variations of the Causal Entropic Forces paper in the past. Personally, I think that this field has a lot of potential both in terms of practical and theoretical contributions.

For concrete projects, I have several experiments and extensions/generalisations of the above papers in mind, and I'm working on such an extension, but before making this issue longer it would be useful to know whether this research area interests OpenAI.

We should add an "author" section to each problem.

OpenAI is “OPEN” but gpt-4 is not even available for free users, even for only
ONE TIME PER DAY

A bit of background info about me, I co-founded a startup in the security industry, my main focus is to collect security related data (spams, attacks, ...) to identify infected or in other way compromised hosts on the internet. Therefor we built a sensor network to collect ~500 million security events per day, ~70% of it Spam alone. I did a lot of research based on the data to find out how Spams look like, where they originate from or what mechanisms where used to send them. I'd like to share my findings about Email Spam with you because they might be helpful for your "Spam the spammers" idea.

First I'd like to stress that Spammers send Spam, not because they want to annoy you, but because they want to make money. The business model behind each Spam can be very different. Spams are used to advertise real e-commerce platforms or services, phish for account information, distribute malware or trick you into behavior to for example share information you shouldn't share. So the purpose of Spam sometimes would be to directly drive sales of products where others would be to expand resources. Only a fraction of the different business models of Spam require a reply from the recipient of the Spam and most Spams are designed to provide that reply or feedback through other communication channels than Email. This is because of the very nature of how Spam is distributed. Sending the amount of Spam we currently face require mechanisms very different to traditional Email infrastructures. Spammers don't use they're own infrastructure to send Spam, they use other people resources. They use malconfigured Email servers, compromised accounts, vulnerable websites, infected computers, compromised servers, they get very creative I saw printers and TVs send Spam. Using other people resources often mean you need to impersonate a valid sender to send out your Spam. Impersonating valid senders or using other people services for Spam often mean you can't reach out to the original sender of the Spam anymore. Replies on Spams would end up on real people Email inboxes or would be denied by real Email infrastructure. The protocol used for Spam is purely for getting your content out into the world, not actually to use it in the intended way.

So, as much as I like the idea of Spamming the Spammers I think replying on Spams in a large scale fashion would hurt the wrong people, who are not even aware that they send Spams.

What I noticed in Spam content was that different campaigns often use similar or even the same template. Templates would often use Spintax combined with static blocks of text. Of course there are more sophisticated techniques to generate the Spam corpus, but this applied almost only to highly targeted content. The high volume Spammers don't care about quality they care about volume and throughput. So to kill 90% of the Spam out there template based detection mechanisms should be sufficient. The techniques currently used by the industry to detect Spam on the content level are very primitive I would say, that's why Spam is still an issue. What I would like to see is an approach to detect Spam not on a content level, but on a communication level. I noticed Spammers often use the same software underneath to send out Spam. This software leaves traces in how a Spammer communicate and behave on a protocol level. There are traces in all layers starting from IP, TCP up to the application layer. I would like to see learning mechanisms, which learn how Spammers communicate and kill them before their content hit your inbox.

I don't know where I should put this. It's not a PR because I haven't build anything yet. I want to clarify some points first and lay out my thoughts and procedures before I decide whether to actually build this thing. It is possible that I am completely missing the point of this challenge.

• Send Automatic Emails - As soon as you receive an email, wait a few minutes, then send a response...
• That Are Replies Of Previous Emails - Spammers have solved this AI-Hard problem by using "generic comments" when responding to blog posts. You can see an example of generic comments in this gist, written in "Spintax". These comments attempt to portray the existence of a reader who have consumed the content thoughtfully, and wants to raise some points about the content. (A new corpus must be designed though, possibly based on experiences in 419eater.com.)
• Waste Time - Lenny is a loop of pre-recorded responses that plays whenever a telemarketer calls a phone number. The responses appear very generic, though they present the illusion of talking to an elderly person that is interested in the telemarketer but is constantly forgetful. Usually, the telemarketer will converse for a certain period of time (can vary) before the telemarketer either hangs up or realizes they are being fooled.
• Not Easy To Detect - This requirement is vague and unclear. Obviously, the deception should not be impossible to detect...if you give the "defender" unlimited time, and unlimited interactions, the facade should be easy to detect. Furthermore, if you are dealing with the same program multiple times, detection should obviously come easier to the "defender" (simply because the "defender" have now saw the "patterns" that the AI uses to generate the email). At the same time, the "defender" starts off at a severe disadvantage simply due to the novelty of receiving an email. Usually, most emails are believed to be sent by a human, so a "defender" would assume that an email is good, and so long as the context is human-like, everything should be okay. Thus, most decent programs whatsoever will "not be easy to detect". This inherent disadvantage should go away after multiple interactions though.

So my questions are:

1. Why should machine learning be seen as appropriate venue for this problem? After all, spammers are able to write up a corpus of comments properly that are effective at fooling a few bloggers. Why not just write up a corpus of prepared emails ("that's a good email, but I have one minor question...") that can be sent instead? Relying on a machine learning algorithm means you have to teach the machine algo how to write, which seems like a rather time-consuming process (and possibly error-prone). The "Request For Research" links to the paper Exploring the Limits of Language Modeling, which had generated pretty good sentences (so long as you exclude the "short and politically incorrect" sentences)...but there doesn't seem to be any sort of larger-scale organization involved in sentence generation. This is...problematic. Without any larger-scale organization, it's just a bunch of random sentences that seem rather useless. I think this issue could probably be alleviated by writing a pre-written corpus of email replies and then training the ML algorithm on that corpus, but I'm afraid that the results may not be fairly pretty...and since you lack large-scale structure, I feel that the resulting emails would be very incoherent babble. If you could convince the ML algorithm to generate random sentences that appear to fit a certain "theme" though ("I am interested in your topic, tell me more!")...then it might work, but this would still be difficult. TL;DR: Using ML algorithms on this problem seems to be equivalent to using a toothpick to dig a hole in the ground, especially when people have already discovered shovels.
2. Would a complete reliance on apophenia count as a "new idea" that is used to solve this problem? Apophenia is the tendency of humans to see patterns where none exist. The goal would be to produce an email reply that appears to respond to the spammer's email, and so the spammer would assume that the email is indeed legitimate...even though there is no real reason to assume the email is thus legitimate. Taking advantage of apophenia is a useful way for helping machine-generated text go farther, but it doesn't really help this organization succeed in producing 'smart' machines (after all, the machines doesn't know what it's writing, much less what it's responding to.)

Don't get me wrong. I am somewhat interested in this problem, and I think it's a solvable one, but I think the solutions I have in mind may not be the solutions you would prefer to see.