Sean (Jan 23 2024 at 18:03): Sean (Jan 23 2024 at 18:03):FYI, Google has a thrust this year to incorporate and consider AI organizations into GSoC.
At the kick-off meeting for admins, it was indicated that if our org has AI projects, we should mark ourselves as an AI organization as that will likely help with allocations. BRL-CAD just completed a pretty big one just last semester with Texas A&M University doing neural rendering, and I'm thinking to propose its continuation as a GSoC project. As such, I do intend to mark our org as having AI project(s).
Towards that end, it would likely be beneficial to have more than one in our project list. If you do have an AI topic you'd be willing to mentor, please mark it in the title as "AI" very clearly and adding an AI tag.
To be clear, you don't have to have AI projects, but having one in your list certainly won't hurt...and might help.
Torsten Paul (Jan 23 2024 at 18:26):My first reaction to that (Google policy) is: I'm out then. May need to have some sleep before actually deciding anything though.
Sean (Jan 23 2024 at 18:28):@Torsten Paul it's not a requirement to have them. It's if you do have an AI project, to make it clear which are AI with a tag/label is all.
Sean (Jan 23 2024 at 18:29):My thought is that we'd end up with 1-3 AI topics in the full list is all. It was just a suggestion.
Sean (Jan 23 2024 at 18:37):There isn't a policy change as far as i know. Google management merely asked OSPO to accept more AI orgs, undoubtedly because that's an industry hot topic. The org application will simply have a checkbox or statement whether we have any AI work, which lets Google include our org in their aggregate metrics.
Sean (Jan 23 2024 at 18:40):I did specifically ask if we're definitely not an "AI org", but we have AI projects, whether we should indicate in our application and they said we should.
Torsten Paul (Jan 23 2024 at 19:12):Pushing industry hot topics backed by closed source and indicated by corporate management instead of supporting open source projects to bring in people is a huge policy shift in my book. This also contradicts the main claim on the OSPO website.
Well, I guess we'll see how things will play out.
To be clear, I'm not opposing adding AI project ideas or anything. If there are projects that can benefit from AI, that's great (if it's fully open source). I'm just concerned the original GSoC idea is getting lost.
Sean (Jan 23 2024 at 20:15):I'm not sure I fully understand, but there absolutely have been changes and shifts to OSPO and Google's internal support for their advocacy of open source (some good, some not so good imho). Mass-firing last year notwithstanding...
As for the AI orgs, they do still have to be open source communities as before -- it sounded more like they want to make sure Google is supporting/inviting the various AI communities to participate like Pytorch, Fast.ai, Keras, Theanos, scikit-learn, OpenNN, etc. Tensorflow and OpenCV (and maybe Pytorch) have been involved in GSoC for many many years already, but many others not so much.
Also, don't know if the AI push was from internal-to-OSPO management or at some level higher -- that wasn't stated. What I do know is that GSoC has now been under threat of cancellation for several years now and microchanges have helped sustain the program. Expanding to allow game projects was one of the first changes I helped push. Expanding to non-notable research projects.
Expanding to more than students. Expanding to different sized projects. Ensuring there's an equitable inclusion of AI projects doesn't seem terribly dissimilar to me.
GSoC's original goal was "flip bits, not burgers" -- a summer job for college kids to get involved in select open source communities.
starseeker (Jan 23 2024 at 20:22):@Torsten Paul I could be wrong, but I think AI is getting pushed across the board right now, not simply by closed source interests - the recent practical results that can be understood and appreciated by the general public have pretty much got everybody paying close attention. Indeed, it could be argued that by encouraging open source focus on AI they're trying to help open source projects and software avoid losing ground to closed source systems.
starseeker (Jan 23 2024 at 20:29):I'm old school myself - I just want a modern Qt interface and robust old school NURBS and triangle mesh geometry operations.
Sean (Jan 23 2024 at 21:20):I bet we could train an autoencoder to solve the np-complete aspects of geometry conversion.
pca006132 (Jan 25 2024 at 05:43):I feel that AI project is not very aligned with GSoC... It is hard for beginners to make a model and train something novel, e.g. geometry conversion. What beginners hope for is to let closed-source tools like ChatGPT/Copilot to do something for you. At best what we can hope for is probably some prompt engineering stuff, new model is just too hard for undergraduate to do in a few months.
pca006132 (Jan 25 2024 at 05:47):probably more use for arts, not sure if AI is suitable for CAD...
Last updated: Nov 15 2024 at 00:49 UTC