Distillation attacks have likely become prevalent, and @OpenAI needs to defend against it.
This method was actually rejected from NeurIPS 2014. A great reminder to just do good work.
The distillation paper was actually rejected from #NeurIPS2014 : ) As I'm sure many people are working towards #NeurIPS2019, optimize for long term impact, not probability of acceptance.
Folks that use @gurobi out there in industry - do you care about duality gaps? Or is a time limit generally used?
Iβm trying to understand how real-world problems are solved.
π€£ π Apology accepted (so this was @_sholtodouglas then? π ), but I think you missed checking the 'Underestimated the cracked Google DeepMind team' box. Amazing work from the Gemini & GDM teams, always great to see folks enjoy our cooking!π¨βπ³π§ͺ
ML conferences have people that are great at theory and math.
We know that a publication there has been a lottery at some point - why did they think that a qualified reviewer is one who has a first author publication there?
Isnβt this just a poor design of the review system? Someone should write a paper that can optimally solve this.
Whatβs even more infuriating perhaps, is that some of the chefs present food that they bought from a grocery store, microwaved it, plated it nicely, and served it.
I could go onβ¦
MasterChef ICLR Edition.
Suppose 100 chefs compete, but Gordon Ramsay and Michelin-starred chefs are not the ones judging.
Your fellow chefs judge you. But only 20 can win. 80 can lose. Isnβt this a strange contest?
ICLR rebuttals were painful. It seems like the whole idea of doing science on paper has broken down.
On the contrary, if you deal with something physical - it is hard to deny it does not work.
Iβm thinking of building a startup next. Should I:
1) Chase the customers? A paying customer says more about your tech than 3 reviewers.
2) Build better optimization software than Gurobi and aim for an acquisition?
3) Something else?
Comment if you have ideas. Iβm in Boston and would love to get coffee in person.