[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Fwd: AI BoF



Hi,

I am not a DD or DM, and I'm not familiar with the challenges that for instance people like M. Zhou face in getting software packaged officially for Debian. Also, I won't be attending in person, only online if that's an option.

However, I have been campaigning as part of my FRDCSA project since 2002 to get many of the tens of thousands of open source AI tools in the wild packaged for Debian, whether that be (possibly roughshod and unofficial) auto-packaged packages, or manually packaged. I have been dabbling with ML/DNN/DL/LLM/etc lately to try to package these tools and/or use these tools to auto-package.

https://frdcsa.org

For instance, I wrote the mailing list at the beginning of 2008 promoting this idea.

https://frdcsa.org/~andrewdo/archive/debian-ai-thread/

Recently, I started a team of four people for a three day hackathon, some of which who have more experience than I do with ML/DNN/DL/LLM/etc, to create an LLM-autonomous agent (such as SuperAGI/BabyAGI/AutoGPT/MetaGPT/BeeBot/etc) capable of packaging software for Debian. Of the around 400 teams that entered, we placed in the top 25, however, we neglected to include a required business plan.

https://lists.debian.org/debian-ai/2023/08/msg00021.html

https://lablab.ai/event/autonomous-agents-hackathon/frdcsa/agent-speak-toolkitbuilder-and-autopacker

At the end of our presentation video you can see a demonstration that two of our team members got working of SuperAGI rolling a Debian package automatically.

Our team is now debating doing a second much long (three week) hackathon to develop these technologies further.

Debian would be the perfect universal operating system to package all this software for, if there was the interest.

If needed, I can speak about some of the technical challenges that we ourselves have faced. I could possibly work on a short to medium length presentation on this topic, if anyone was really interested (otherwise, my time would be better served by continuing to work on FRDCSA), by adapting some of my previous attempts at presenting at a DebConf on the topic.

For instance here is a somewhat cringey partially finished (but never shown) video for DebConf 2020. I don't remember what's in it, and I was definitely going to finish and revamp it, such as with using Text To Speech instead of my own voice. It's embarassing to share this:

https://frdcsa.org/~andrewdo/output.webm

However, I haven't rolled a .deb in years, and have not been brought up to speed on the latest best practices.

I apologize if I have erred in sending this to the mailing list. I am just offering some help if people should want it, on the topic of packaging-with-AI and packaging-for-AI.

Cheers,
Andrew Dougherty







On Fri, Sep 1, 2023 at 5:48 AM Joost van Baal-Ilić <joostvb-debian@mdcc.cx> wrote:
Hi Aigars e.a.,

On Fri, Sep 01, 2023 at 11:04:38AM +0200, Aigars Mahinovs wrote:
> IMHO we are missing one big topic in the current DebConf schedule - the
> explosion of large language models and other generative AI models. This is
> raising questions of what is or could be Debian role in these developments?
>
> So the question from my side is - is anyone who will be there at DebConf
> and is feeling competent enough to chat about this topic, maybe also give a
> quick summary of what the landscape of this functionality and its
> challenges is currently, to start the discussion?
>
> I expect that tooling for development of such AI tools is in Debian or is
> easy to get into Debian and that training of models is too cost-intensive
> for Debian to do (currently) but what about distribution of pre-trained
> models - what are technical and legal challenges there? What about hosting
> execution engines of pre-trained models - is that within Debian compute
> capability? What about distribution of tools with AI integration - local,
> Debina-hosted or externally hosted? What are technical and legal challenges
> there?
>
> If we don't have a better candidate, I can volunteer to host/moderate the
> BoF, but we for sure need some more competent people who can actually talk
> there. Any ideas?

Such a BoF could be very useful indeed, thanks for bringing this up.

People could read https://salsa.debian.org/deeplearning-team/ml-policy to
get some feel of some of the current challenges.

My € 0,02,

Bye,

Joost

PS: No, I do not volunteer to give an introduction.



--
Andrew Dougherty
Independent A.I. Researcher
https://frdcsa.org

Reply to: