If data (words, images, music, stats, etc) is the new code (in the sense that it is required for creating computer programs which achieve a desired output), shouldn't it be similarly valued by our society and economy?
Shouldn't the creatives whose works enabled the existence of technology that would otherwise be impossible with hand-coded algorithms be, at the very least, credited and compensated for their contribution?
"GEM: Now here's the deal. That's fun. But it's very much like pen testing. You cannot pen test your way to security. We all know this. We've known this for years and years and years. And now we're saying we can pen test or we can prompt engineer our way to security with an LLM."
#job#jobalert#Postdoc researcher in Machine Learning, working to develop a method for identifying endangered heritage sites in forested mountains of South America as part of the Mapping the Archaeological Pre-Columbian Heritage of South America (MAPHSA) project.
Data oceans required to build generative models are not getting any bigger, nor are they getting less polluted over time. Less focus on computers (GPU cycles) and more focus on recursive pollution would be nice at this point @cademetz
Last week I attended the 6th Perspectives on Scientific Error Conference at @TUEindhoven
I learned so much! About #metascience#preregistration#replicability#qrp questionable research practices, methods to detect data fabrication, #peerreview, #poweranalysis artefacts in #ML machine learning...
I'm impressed by the commitment of participants to improve science through error detection & prevention. Thanks to the organizers Noah van Dongen, @lakens@annescheel Felipe Romero and @annaveer
What I'm hearing is that in a month and half they got 2/11th of the way to spelling "Shakespeare" (S----s-----)
So I think you really need to scale up either the number of monkeys and computers or the duration of the experiment before you can say anything with certainty! 😝
@SwearyMonkey there is simply not enough data. so far ZERO black box LLM foundation models are cleaning up the data ocean before they suck it down. Hence it is incredibly valuable if you believe that the current crop of "bigger is better" LLMs.
[Image is supposed to be robot hands pulling a plant from the ground but may still be a creepy ass array of stepford women with conspicuous complexion variation]