Do you believe there is a limit to what models we can train with layers of differentiable units?

Do you have info materials about OpenAI I could use in AI events?

Will you work with 3rd parties to rise awareness and support, recruit talent and external collaborators (e.g. research groups)?

May I suggest Kaggle?

The AllenAI Institute is currently running a pretty interesting contest there at the moment, with pretty direct links to OpenAI's mission.

Of course Kaggle often degenerates into a "who can drive xgboost[1] the best", but some of the feature engineering is quite novel.

A decent Winograd schema challenge would be pretty interesting.

[1] https://github.com/dmlc/xgboost#whats-new

No particular materials right now besides the website.

We expect to collaborate with many third parties over time. We're just getting started, so not sure on the specifics yet!

Okay. May I write you an email to establish the connection? I intended to contact OpenAI later this month, but now that we're talking...

Feel free to get in touch: gdb@openai.com!

I'll send you a mail tomorrow. Thanks.