Friday, May 2, 2025

5/2/25: MM presentation on Bringing GPU-Accelerated Computing, Data Science and Deep Learning to the Classroom, etc.

 Artificial Intelligence Study Group

 Welcome! We meet from 4:00-4:45 p.m. Central Time. Anyone can join. Feel free to attend any or all sessions, or ask to be removed from the invite list as we have no wish to send unneeded emails of which we all certainly get too many. 
Contacts: jdberleant@ualr.edu and mgmilanova@ualr.edu


Agenda & Minutes  (161st meeting, May 2, 2025)

Table of Contents
* Agenda and minutes
* Appendix: Transcript (when available)

Agenda and Minutes
  • Announcements, updates, questions, presentations, etc. as time allows
    • MJ writes: Please share this with your Friday afternoon group. Tonight [W 4/30/25] we have an expert on agentic AI, but we are meeting monthly. https://www.intherock.ai/.
    • MJ writes: Monday is the last day for high school and college students to sign up for the Hackathon. Info attached. Please share. Also, if you or others want to help me with the Hackathon, let me know. It's a big lift!  mkjohnson@ualr.edu. 
    • Today is the last day to participate in the AI Campus Climate Survey. The UALR AI Integration team, led by Curriculum and Special Projects Coordinator Nathan Holloway, is seeking information the campus community's experiences, perceptions, and expectations regarding the integration of artificial intelligence (AI). Insights from this survey will inform the development of implementation strategies, programming, and policies to effectively integrate AI across academic, administrative, and student support functions. Survey Link: https://ualr.qualtrics.com/jfe/form/SV_e5up1YI8n1X73Ku. Deadline: Today! Thank you for helping shape the future of UA Little Rock!
      Nathan Holloway
      Curriculum Coordinator & Special Projects Manager
    • 5/9/2025: ET defense (book project). 
      • Book: https://docs.google.com/document/d/1tOhGzVFYnwmkTlb0_brlRSY6Uol6vy3MNFf3Q98ZNDc/edit?tab=t.0
      • What next? Put it online, self-publish, both, neither, ...?
    • Today: MM presentation on BRINGING GPU-ACCELERATED COMPUTING,
      DATA SCIENCE AND DEEP LEARNING TO THE CLASSROOM
    • Soon: EG and DD could do a session on programming AIs (loading AI model into video card, programming, etc.). Could talk about rapids.ai. MJ would like to publicize to some students once we have a definite schedule for this. (We'll do it twice, once just for us and once for external invitees.)
    • Any other updates or announcements? 
  • Masters project on using AI to write a book.
    • A source of information about the problem: https://spectrum.library.concordia.ca/id/eprint/993284/
    • ET: Growing vegetables from seeds. 
      • Committee: DB, RS, MM. 
      • Proposal status. Approved!
      • Report suggestion: The book
      • Defense presentation suggestion: The experience and process, hints for future book authors, observations, etc. Tour of the book
      • Gemini is doing better than others for pictures for the book. They are more realistic compared to the others. 
 We got up to here.
  • We did the Chapter 6 video, https://www.youtube.com/watch?v=eMlx5fFNoYc, up to time 13:08. We can start there next time we do it.
  • Schedule back burner "when possible" items:
    • TE is in the informal campus faculty AI discussion group. SL: "I've been asked to lead the DCSTEM College AI Ad Hoc Committee. ... We’ll discuss AI’s role in our curriculum, how to integrate AI literacy into courses, and strategies for guiding students on responsible AI use."
    • Anyone read an article recently they can tell us about?
    • If anyone else has a project they would like to help supervise, let me know.
    • (2/14/25) An ad hoc group is forming on campus for people to discuss AI and teaching of diverse subjects by ES. It would be interesting to hear from someone in that group at some point to see what people are thinking and doing regarding AIs and their teaching activities.
    • The campus has assigned a group to participate in the AAC&U AI Institute's activity "AI Pedagogy in the Curriculum." IU is on it and may be able to provide updates now and then.
  • Here is the latest on future readings and viewings
    • https://transformer-circuits.pub/2025/attribution-graphs/biology.html#dives-refusals
    • https://venturebeat.com/ai/anthropic-flips-the-script-on-ai-in-education-claude-learning-mode-makes-students-do-the-thinking
    • https://transformer-circuits.pub/2025/attribution-graphs/methods.html
      (Biology of Large Language Models)
    • We can work through chapter 7: https://www.youtube.com/watch?v=9-Jl0dxWQs8
    • https://www.forbes.com/sites/robtoews/2024/12/22/10-ai-predictions-for-2025/
    • Prompt engineering course:
      https://apps.cognitiveclass.ai/learning/course/course-v1:IBMSkillsNetwork+AI0117EN+v1/home
    • https://arxiv.org/pdf/2001.08361

Appendix: Transcript

AI Discussion Group
Friday, May 2, 2025

0:20 - Unidentified Speaker:
Hi, folks.

0:21 - M. M.:
Hello, hello. Good afternoon. Hello, hello.

0:25 - D. B.:
I see we've developed a one here in the upper left-hand corner. I can't seem to get rid of it. I don't know how that happened.

0:42 - M. M.:
M. J. advertised our blog and our group. Yeah, actually, M. is here.

0:51 - D. B.:
Yes.

0:51 - M. M.:
Thank you for coming. Okay, well, I guess we can go ahead and get started. Yeah, so she will see the same presentation that actually I present already.

1:08 - D. B.:
Oh, for the Tech Ted tech.

1:11 - M. M.:
Yeah, yeah, yeah. So many of you actually saw the presentation, but I have one video very interesting to share right now.

1:22 - D. B.:
Okay, well, let's, I'm going to put that, let me just do a few announcements and then we'll get right to it. Okay. So, last Wednesday, there was a monthly meeting of the Little Rock AI discussion group. And this is from M. J. And here's the link, and if you want to join it or whatever, check it out. Just follow the link.

1:51 - M. J.:
Yeah, do you mind if I just say something about it?

1:55 - D. B.:
Sure. So it's a meetup.

1:57 - M. J.:
It's been set up by business people. And E., didn't you go there? Did you go? I thought I saw someone from this group there, but maybe not.

2:07 - M. M. & M. J.:
B. was there for sure. D. Yeah, so anyway, it was stormy weather on Wednesday night.

2:16 - M. J.:
There was a tornado watch. We had over 40 people there. And most of them are in business, but we did have some students and some faculty from, like T. W. was there. He brought his son, who's kind of another budding AI genius. But anyway, it was fantastic. V., were you there? Someone was there. I was not. I got rained on.

2:40 - V. W.:
Yeah, by the tornado warning and so forth.

2:43 - M. J.:
Yeah. I was so surprised at how many people were there despite the weather.

2:48 - M. M.:
Anyway, it was fantastic.

2:49 - M. J.:
Z. M. presented, and he talked about agentic AI. It was really inspiring. And he talked about how he's using AI in his life. And it was just really great. And they had great food and wine and beer. And it was lots of communication with everybody, a really big space where people I got to meet a lot of people. I think that you guys would just really enjoy it. And we really are trying to give a boost to and support industry partners who are also interested in AI here. So I hope you guys join.

3:27 - Unidentified Speaker:
Great.

3:28 - D. B.:
Well, thanks. And also, since you're here, do you want to say a word about this one?

3:35 - M. J.:
Sure. Yeah. The AI Hackathon, this is just an amazing program. I mean, F.'s working on it, S. L. I mean, we've got great courses. Now we have T. W. and N. S., and we have some people helping with the game development and the application development of it. But we also have mental health experts from UAMS, from Arkansas Blue Cross Blue Shield, from the Veterans Administration, and from other organizations We're all participating. We're about to be full. V., we're about to have 42 people attending. That's good.

4:09 - M. M.:
The deadline to sign up is for high school and college students, any high school, any college.

4:15 - M. J.:
It is in person. You do have to be on campus, and we don't have housing. And it's June 9 through the 13th. We're going to have a big pitch competition. We've got a $2,000 donation to give money to the winning pitch, and all of these students are going to it's a hackathon so they'll all be coding around mental health solutions and um if you guys know of somebody we're we're on the wire I mean again deadline is Monday to sign people up but it's going to be an awesome opportunity and um it's probably going to happen again it'll probably happen again next year V. has D. signed up for No, I don't think so.

5:01 - M. M.:
All of these people that are in this group, they are really very good in AI. They can help us a lot. Do you want me to ask him to sign?

5:14 - V. W.:
I think that might be a good thing for him because he's doing medically related AI and this certainly is in that bailiwick and it'd be good connection.

5:27 - M. M.:
that he would develop and so forth.

5:29 - V. W. & M. J.:
And we want to promote that. My email address has an M before the Kay Johnson part. It's either getting scrolled off or it's just a typo. It's a typo.

5:38 - Unidentified Speaker:
Yeah.

5:38 - V. W.:
Yeah. Oh, no, I'm wrong.

5:39 - M. J.:
So you guys, if you want to help me, if you want to help with the hackathon, but you don't want to attend, but you want to help, you know, you want to be a mentor. You want to work with the student groups as they are creating their solutions. Give them any advice, please send me, drop me an email, let me know that. And yeah, you can see if you can, you know, you can help us out. Yeah.

6:04 - V. W.:
These hackathons are really good for the campus. They're good for the energy behind this interest area. And it's something we really want to promote so that it becomes just normal to be at the hackathon on a weekend when you're going through the year. Yeah.

6:18 - M. J.:
And this is a week long and they're going to get some AI coursework from V., if they're advanced or members of her team.

6:26 - M. J. & V. W.:
Or they'll get AI fundamentals training from S. L.

6:30 - M. J.:
Yeah. A lot of progress in AI has been facilitated in the hackathon format over the years.

6:37 - M. M. & V. W.:
The BERT LLM, which was sort of like the preliminary LLM that led to chat GPT.

6:43 - V. W.:
It came up through the ranks of these kinds of contests where people were trying to improve performance. And convolutional neural networks, you know, recognizing dog from cat and all those things also was improved through this process. And you only look once where you analyze a picture and you see all of its elements without having to reevaluate it. Is what it's called. You only look once. Those kind of YOLOs and CNNs and LLMs all trace their improvements at least to the hackathon environment.

7:17 - M. J.:
So, yeah.

7:17 - M. M.:
Thanks, D. We can give them certificates too. And what I want to say, M., that probably what I'm seeing is a lot of interest and probably we will call another hackathon soon, additional to this one. And we're looking right now then for understanding from industry perspective and the participants, particularly health, mental health, what kind of problems they have how actually the AI can help them. So this is- Yeah, I love this.

7:52 - M. J.:
This is near and dear to my heart because I love, I don't love technology for technology's sake as much as I love how it can really make a difference in our community and for people in our lives. And so, yeah, so it's a lot of work, but yeah, we can do it again. We can do another one.

8:14 - V. W.:
yeah maybe different topic maybe different yeah we're really living in a moment where we're going to and you know in a historical moment where we're going to see the impact of llms on counseling on cognitive behavioral therapy on couples counseling as we learned in M.'s uh tech talk and so I'm pretty excited to see and even in uh bioinformatics of pharmacology you know people are given these weird cocktails some of which work some of which don't some of which are titrated over years. And we we don't really have the database together of how to optimize patient care. And it's done still by very much trial and error. And so these hackathons have the opportunity to open up new doors that have more leverage to actually help people, as M. points out.

9:01 - M. J.:
And R., I know you want to say something, but just let me also point out that we have unlocked tremendous demand and interest from people who who are on the spectrum or who are dealing with any number of mental health or behavioral health issues, who are already using AI to try to level the playing field. And even that's what our speaker who's using agentic AI, creating his own agents, Z. M., he's got ADHD, he's got this and that, and he talked about how he was using AI to help him. I am super curious about the last two minutes of this topic.

9:47 - R. R.:
My daughter is actually a cognitive science major and she's been working on a thesis, she's a bachelor's at this point in time, and she's working on a special project for an organization called Simply Neuroscience. Of aligns with some things that you're referring to, but she's not a computer science major, she's not a programmer, she's doing some cc++ right now, she might do Python later, but can she be part of this in any form at all? She's not part of it.

10:24 - M. J. & R. R.:
Please, please, yes, please, yes, you don't have to be a coder. We have purposely, we're making pretty big teams, there will be some coders on each team, but we are working everybody will learn a lot, maybe about application development, or maybe just about identifying audiences and value and, and that kinds of thing. So there's a lot of work to be done outside of the coding.

10:53 - R. R.:
That's a part of the pitch. Yeah.

10:57 - D. B. & M. J.:
So yeah, it would be wonderful. What platform is used for coding?

11:03 - M. M. & M. J.:
We have So the idea to use this n8n.io, the platform that is actually without coding.

11:13 - M. M.:
And additional, we're using Google colab for Python. But yeah. Yeah.

11:20 - M. J.:
And then I'm having a conversation about for game development part, possibly Unity. So anyway, D. We're going to make that decision specifically once we actually have, which will be by next Friday, what the problems are that we're going to solve. And if you guys have ideas for real problems that we could help solve, I would invite those too. So I'm going to put two links in here, or I'll put a link to the form so, R., you and your daughter can sign up. And also, I'll remind you again. Of my email address and you can send me if you have ideas or you want to participate in some. You know for what problems we might solve or you want to be a mentor to the students.

12:11 - R. R.:
You know I would welcome you. Your name came up in one of Dr. S.'s seminar grad seminar a few weeks ago that that Dr. S. was trying to create what is called an AI factory. With all the different projects and I've had a chat with him as well about it. It's fascinating and he did He was asked to talk to you to see how it can go through some legal counsel maybe also. So on that note, so yeah, I did note down your email earlier. Thank you. Yeah, please, definitely.

12:45 - M. J.:
I'd love to hear more about the AI factory. Because we're working on the Build Good AI Institute. I guess you guys know that. I think I told you that. But yeah, so that's what we're scaffolding here. And we're beginning to, and we've just really hit the ground running.

13:08 - D. B.:
I mean, we've already made a lot of progress. OK, another thing is there.

13:15 - E. G.:
I have a quick question. Sure. M., is this all on site or is there any virtual?

13:24 - M. J.:
It is on site. I don't think we're going to do any part of it virtual.

13:31 - E. G.:
I'm sorry. That's fine. The commute made me just a little bit too much for me. Where do you live?

13:40 - M. J.:
I'm just south of Bangor, Maine. That'd be big.

13:43 - M. J. & R. R.:
Sorry. I'm also in the same boat. Yeah, we're in California.

13:48 - M. J.:
Oh, OK. And is your daughter in California?

13:52 - R. R.:
She is, yeah. Oh, boo.

13:56 - V. W.:
Yeah, you know, it'd be really nice if we could open up a pinhole into the firewall of remote participation, because sometimes you can do things remotely more effectively than you can in person, because everyone is already in front of their own screen with their own stuff. So you can actually get a little faster progress than you might think.

14:16 - E. G.:
So maybe maybe we could look on that.

14:19 - V. W.:
If not for this time, maybe for the next or next.

14:22 - M. M.:
This is a great idea, M. We can consider all of my classes are online, actually.

14:29 - D. D.:
Yeah, I think V. just volunteered.

14:32 - M. M.:
Promote is the new here now thing.

14:36 - R. R. & V. W.
Yeah, I would happily help facilitate any kind of online presence because the rate at which you can coordinate multiple people is really almost better than we can in person because in person, number of possible channels grows as the people squared.

14:57 - V. W.:
And with Zoom, that's still true, but we can manage it better.

15:03 - R. R.:
Absolutely. Well, thank you so much.

15:06 - M. J.:
All right.

15:07 - D. B.:
So there's a campus survey. People are being asked to fill out a survey here. Deadline is today. So after this evening or after this meeting, whatever, feel free to go to the minutes and get this link and fill out the survey. Campus Climate Survey.

15:29 - M. M.:
Yeah.

15:30 - M. J.:
D., can you put the survey link in the chat so I can grab it, please?

15:37 - V. W.:
Thanks for adding that link, M. I see it in the chat now.

15:43 - M. M.:
Oh, it's a direct message, I see.

15:46 - Unidentified Speaker:
OK.

15:47 - D. B.:
There's a link in the chat I should put in the minutes.

15:52 - V. W.:
Yeah, that's perfect. That's perfect. Now we're already there because of the instantaneous power.

15:57 - D. B.:
So what is this link, M., that you put in the chat again?

16:03 - M. M.:
Registration link, I think.

16:04 - D. B.:
It's the what? To register for Hackathon.

16:07 - M. M.:
Oh, OK. OK.

16:15 - Unidentified Speaker:
And we appreciate your help, okay?

16:18 - M. M.:
Like I say, all of you are very, very well prepared for helping in AI hackathon.

16:26 - M. J.:
Sorry, I had to step away because my husband showed up to rescue a rosemary plant. Oh, what's the link?

16:36 - D. B.:
I got two links here.

16:39 - R. R.:
One more quick question, you were mentioning platform that non-programmers can use to code. I can send the link too.

16:51 - M. M.:
Oh, that'll be awesome.

16:54 - R. R.:
I'm just sending you an email.

16:58 - M. M.:
Oh, M., can you send the link?

17:02 - M. J.:
Yeah, I've got it. By the way, that speaker about it as well. Yeah.

17:10 - M. M.:
Is it M8IN?

17:12 - M. J.:
M18N? So, Dr.

17:13 - D. D.:
B., when I follow the link, I get up like a readers, a reading list. Which link? Something like this. Is it supposed to be? I thought you, it was some type of a survey. Yeah, yeah, the survey.

17:30 - D. B.:
Yeah, that was, that was sort of The survey is for this, this one. And to register for the hackathon is this one.

17:40 - D. D.:
You put the, uh, you put the survey in chat, right?

17:44 - D. B.:
I guess it did end up in the chat somehow. Something.

17:48 - E. G. & M. J.:
It is in chat, but it goes, it goes to, it goes to a reading.

17:54 - D. D.:
It's not a survey.

17:55 - D. B.:
I need to delete it here. There, it's gone. Now, M.'s link is still there and it won't confuse. Okay. E., any updates on your book project?

18:15 - M. M.:
Hi, I'm almost done with my defense slides. So I'm still working on it right now.

18:27 - E. T.:
Hopefully it'll be done today, and I'll send it to you for your feedback.

18:40 - D. B.:
All right.

18:41 - V. W.:
E.'s defense is next Friday in this meeting.

18:48 - D. B.:
Awesome.

18:49 - E. T.:
This is the front page of her book.

19:01 - M. M.:
Looks very good. From seeds to salt. Yeah, it's beautiful.

19:05 - M. J.:
It's beautiful. And it's much, I guess it's not all loaded, but anyway, we'll hear more about it next Friday.

19:12 - D. B.:
I thought the title was so fantastic. I thought like, my God, it's, I'm too logical. I mean, I've been trained to think too logically to ever think of a title like that. So I asked E. if this was, she came up with it or it was an AI suggestion and she said it was an AI suggestion. So I think the AI, AIs can have more creativity than I have.

19:42 - M. M.:
Yeah, sometimes. Well, yeah, but it's really good at linguistic creativity.

19:48 - D. D.:
And also the picture, E., the picture here is an AI generated picture too, right?

19:57 - D. B.:
Yes, sir.

19:58 - E. T.:
It is also AI generated. And well, actually, ChatGPT is the best one being able to put the text on the picture with the correct spelling.

20:10 - D. B.:
So yes, it is AI generated as well.

20:14 - Unidentified Speaker:
OK, wonderful.

20:15 - D. D.:
That's pretty good. Thank you. Yeah. Great.

20:19 - Unidentified Speaker:
All right.

20:20 - M. M.:
So V., would you like to share your screen and give us Yes, yes, because I will be brief and if you have any discussions after that, go ahead. And what I want is actually to start with one video, but I'm not sure that I can do the video.

20:40 - M. J.:
And D., while she's doing that, do you mind, I mean, is that, was that survey link, I mean, it went to a blog, but do you have a different link that goes to the survey? The link was not right?

20:57 - D. B.:
Yeah, you took it down.

20:59 - M. J.:
It was a reading list to the blog and you took it down, remember? But I'd really like to fill out that survey and see who's put that survey out there. So anyway, just throwing that out there.

21:18 - D. B.:
Okay, so the link I'll put the link back in the chat here.

21:25 - Unidentified Speaker:
Maybe I chatted a bit wrong or something. I mean, I just, it worked.

21:33 - D. D.:
I just tried it and it went to the survey.

21:39 - D. B.:
Now I need to go back to...

21:43 - M. M.:
Hold it right out of your browser, that way. Suggestion there. Okay.

21:53 - M. M.:
So, as you know, NVIDIA is involved in many tasks, but originally they start with computer graphics and graphics card. And this is very good demo of integrated computer graphics with 3D scene development. I don't know if you can hear it or see it. Can you see it at least?

22:21 - D. D.:
Yes, yes, I can tell. Can you hear it? Beautiful concept art of a...

22:27 - M. M.:
Generative AI synthesizes amazing images from simple text prompts, yet image composition can be challenging to control using only words. With NVIDIA NIM microservices, creators can use simple 3D objects to guide AI image generation. Let's see how a concept artist can use this technology to develop the look of a scene. They start by laying out 3D assets created by hand or generated with AI. Then use an image generation NIM, such as Flux, to create a visual that adheres to the 3D scene. Add or move objects to refine the composition. Change camera angles to frame the perfect child, or reimagine the whole scene with a new prompt. Assisted by generative AI and NVIDIA NIMH, an artist can quickly realize their vision. So are you able to see the video? Yes. Yes, and we can hear it.

23:35 - M. J.:
Can hear it or not?

23:39 - D. D. & E. G.:
Yeah, we heard it. Yeah, we saw the video, yes.

23:44 - M. M.:
Yeah, so because this is the generative AI actually are doing extremely good job with... This is a really important area moving AI tools into the third dimension where spatial relationships are being explicitly represented and not embedded in the language.

24:05 - M. M. & V. W.:
Because geometric and And spatial relationships are embedded in a different way than we typically use in language.

24:12 - V. W.:
If you're trying to describe the structure of a building, you have to use geometric measurements and relationships in such a way that they can be easily apprehended by the machine. And LLMs are not doing the job. And programs like mid-journey 2D image synthesizers are not doing the job when it comes down to rigid specifications of engineering objects that we need to actually make stuff that works. So yeah, this is a very important area.

24:40 - M. M.:
Exactly. This is why I'm showing you this demo. Because like V. mentioned, we have diffusion model before, diffusion model from text to image or to text to video, text to 3D object. But this is more. You can use the 3D graphics from Blender.

25:04 - D. D. & M. M.:
or scratch or whatever you decided or open GL so this is how I know yeah yeah it's integrated with actually do you have more information on that oh yeah I have more information on email me some of that okay yeah I have the link for this yes and it's everything related why I'm showing you you is because it's related with my microservices.

25:33 - M. M.:
These microservices, they are actually like a hug and face models that you can use. OK, so I will skip this.

25:46 - Unidentified Speaker:
The job opportunities that I told before, like,

25:51 - M. M.:
170 billion new jobs will be created, so this is not true that the job will Many of them will, you know, change the way that they are working, but new jobs will be created. So I will send you the link, yes, for microservices and particularly for this one. But I know that most of you already are familiar with Deep Learning Institute, but they offer a variety of resources So not only workshops that are with instructor, but also self-placed courses, teaching kids, and also they are offering a lot of solutions for enterprises, for the companies. So right now we're preparing work, professional certificates for large language models and professional certificates for agentic education. Specialists. So we're preparing these certificates right now. So they're particularly for industry. So the certificates are in different areas, in data science, in conversational AI, in edge computing, you know, the Jetson, and not only you can use for any edge device, embedded devices, and so on. In robotics, they have applications right now. So why I'm doing this talk, because I was thinking in the past that only faculty can become certified instructors, but I understand now that actually graduate students also can become certified instructors, okay? So the process is become developer, and all of you are actually developers, because you already get one certificate or participate in one of our workshops before. So you are developer. So the application process is to apply, and I have the links. All of the links are actually in the slides, and I shared the slides with D. So you have the slides, or if you need the link, I will share the link again. I think some course, one course or two. After that, there are some exams, but they are not so difficult exams to emulate your teaching performance and how you are using the Zoom or their platform or whatever. And in the end, there is one face-to-face or online exam with person in NVIDIA. So when you get this certified certification and become instructor, actually all of these advantages are available. They give you resources, they can give you the teaching kits, they can give you the free self-learning courses, and even they pay some money if you participate in, how many they want, three, I think, three per year, and they pay you some money. So it's not about the money, it's about actually how you, I mean, your people benefit in our society. And I really want our university to have more students certified and more faculties certified. So the areas are multiple areas. So AI foundations, deep learning, accelerating computing. This is very helpful. My student, R., my PhD student, become one of the leader in high performance computing in Fayetteville right now. Many of my PhD students actually graduate with, I mean, they are working with RAPIDS. It's a data science. I don't know. It's not here, but we have a data science. Multimodal courses.

30:20 - V. W.:
These courses are really well designed from both the, you using it as an interactive platform and then taking the little tests they give you for progress measurement. And you can just move this right into your Jupyter Colab notebooks. If you want to work on Google, So, whether you're working on the NVIDIA platform for learning or transporting this knowledge to Google Colab, it works well because the NVIDIA libraries are also available in Google Colab, so you can just run the code there as well.

30:53 - E. G.:
If you get a chance, I think, sorry to interrupt, but D. and I, D. D., we were playing around. I've got experience with rapids.ai. Yeah, it's it's one of the It's a difference between throwing a bullet and shooting a bullet in trying to do machine learning.

31:17 - M. M.:
Yes, yes, E. Yes, RAPIDS is included in many companies right now. I included in. What what? Yeah, the the workforce development additional certificate for data science. So RAPIDS is included. I don't know why it's not included here, but it's in data science. We have the data science. And you have this self-learning course, self-place. But I want to emphasize again, this is the microservices. These are actually the models that we're promoting and integrating right now. So PDF to podcast, another is text to image, several models and the link is here too. And right now, this is from the talk, probably everybody listened this talk about the time that we're in a Gentic AI and we try and hack right now to implement actually the multi-agent approach. So the difference, some people ask me, you can become certified instructor without to be university ambassador. So for all of you, you can do certified instructor and work with industry, academia, public workshops. So all of these are the guys and applications And this is the courses that I offer if students are interested. And that's it. And I have a very good video generated from text, but I already show you the video, I think. Another one for Prometheus. Okay, so R. has a question.

33:17 - R. R.:
No, I just, thank you so much for sharing. Yeah.

33:22 - M. M.:
and I have to find this link to send you. It's actually the presentation that I give recently, and most of you are already familiar with this, but the new is, I think, this very impressive demo. So I don't know if I go Is the link there? Generative AI synthesizes amazing images from simple text prompts, yet image composition can be challenging to control using only words. With NVIDIA NIM microservices, creators can use simple 3D objects to guide AI image generation. Let's see how... This is just the video, but I have to send you the original. Yeah, I have to send you the... But these are all of the NIMS agents. Why are you not investing in tax links? No, seriously. Why are you not investing in tax links? That's... Can we... Can everyone...

34:44 - R. R.:
get a copy of the deck that you shared?

34:48 - M. M.:
I think that, yeah, I sent to D. I can send again. OK.

34:54 - R. R.:
I'd love to get my hands on the deck.

34:58 - M. M.:
Yeah, because they are very good links. And the link for this 3D generation, I will share right now in the chat.

35:08 - R. R.:
I was on a meeting with NVIDIA two days ago at Seattle.

35:15 - M. M.:
OK, so great.

35:17 - R. R.:
And NVIDIA, I went to the GPU factory maybe about three months ago at NVIDIA.

35:30 - M. M.:
Yeah, I'm teaching for NVIDIA from 2018, OK? So every year I'm at least to us. Yeah.

35:43 - R. R.:
Just FYI, I'm sure everybody has heard about the project Stargate from two months ago. So I'm actually working on that. About 200 engineers are working from OCI on that.

36:00 - M. M.:
Wow. Yeah. Excellent. They will create the companies here, new companies. In the States.

36:08 - Unidentified Speaker:
Yeah.

36:09 - R. R.:
Phoenix and Houston. Yeah. This one is in a place called Abilene near Texas. Yeah. It's primarily a data center for OCI, primarily to, I'm sorry, for OAI, OpenAI, to manage large AI workloads. It's not close to Houston?

36:32 - M. M.:
I don't know.

36:34 - R. R.:
it might be, yeah, I only saw it in the map, where is Abilene.

36:46 - Unidentified Speaker:
Okay.

36:46 - D. B.:
Okay, so, D., if you need me, I will send the slides again. And as usual, I'm getting I'm a little bit confused. Is it, what was the, yeah, can you chat the link again? It's not the link. It actually has a deck.

37:13 - M. M.:
It's the presentation.

37:15 - D. B. & R. R.:
Oh, well, it's hard to upload that stuff.

37:19 - D. B.:
You can hit a reply on the blog, because I have to send, bring up the PowerPoint, save it as individual, images, and then upload all the images.

37:32 - R. R.:
Is it possible to hit a reply all to the Zoom invite? Or not the Zoom invite, to the invite that you created?

37:41 - D. B.:
Pardon? Is it possible?

37:42 - M. M.:
Or the email. I can probably reply to the email.

37:46 - R. R.:
Yeah, yeah, to the invite. Yeah, that would be awesome.

37:50 - M. M.:
To the invite email.

37:52 - D. B. & M. M.:
Oh, exactly. Sure, yeah. Yeah, I can.

37:54 - G. S.:
Yeah, Dr. M., this is G., I mean, sorry, I joined like 15 minutes before, so I probably would have missed a lot of what you guys discussed. Is it possible to, for someone based in, I mean, I'm based in Australia, so can I also get certified, or is it only for people who are based in the US?

38:17 - M. M.:
No, no, no, this is everywhere. We have people from everywhere, from India, from UK, from, there is a list of Certified and ambassadors for each country. Okay, you can you can get it. Yes, I will share the The links right now Which link this is for a certification or for the video? Yes, you just apply they will be happy. Oh You should What they have a web page and they list it but I also do when I am going to new country I'm going to ChatGPT and ask who is the certified NVIDIA instructor in this country or in this area or in this university and this is how I met people, you know. You can do it from LinkedIn or NVIDIA resources but you can do it in ChatGPT. Probably you have already several people from Australia that are already certified instructors or university ambassadors, one of these. Got it. Okay.

39:26 - G. S.:
And I work for Informatica. So Informatica is an ETL and a product organization focused on data and AI, right? So I would be very keen to understand how we can collaborate together from the Informatica perspective as well. So glad to think on those lines as well.

39:48 - Unidentified Speaker:
Perfect.

39:49 - M. M.:
Perfect. Yeah. Yeah. If you send me email or I'll try on the invite to share these links right now. Yeah. OK. All right. Thanks. If you are included in the email or just send me the email. Yeah. My name is there in the list. So I don't attend every week, but sometimes when I get a chance, to join in.

40:18 - V. W.:
I'm going to have to bug out. I have a graduate student who is having a committee presentation today, and I have to get to it. But it's really been wonderful, as always, to be in this meeting, and I appreciate it.

40:35 - M. M.:
So, yeah, thank you. Thank you, V. And I also have to leave early. So I'm trying to send you the links.

40:44 - D. B. & R. R.:
And thank you. Thank you, Dr.

40:47 - R. R.:
M. Thanks, everybody. You're welcome.

40:51 - D. D.:
Thank you so much.

40:54 - Unidentified Speaker:
Bye. Bye.

40:55 - D. D.:
I wonder if he was talking about that Stargate, the infrastructure project for the AI. I wanted to ask him.

41:08 - Unidentified Speaker:
I guess he's gone now.

41:12 - M. M.:
Yeah. And I cannot find right now, but I will send all of these links later, probably.

41:26 - Unidentified Speaker:
I cannot find it now. At least one is here.

41:34 - E. G.:
The link that you sent earlier, Yeah, I was checking that out.

41:46 - M. M.:
It looks pretty cool.

41:51 - D. B.:
All right.

41:53 - M. M.:
This link is here. Okay, link for the video. Not only video, explaining AI generative images, but using 3D guidance. So 3D models, how to integrate them with generating image or video. Dr.

42:26 - E. G.:
M., I think this is the first time I've seen an amalgamation of both a visual, where you create some artwork, and verbiage, where you describe what you want, and it would amalgamate the picture, or generate the picture. But you've got two mediums that you're deriving from, which... It's powerful.

42:55 - M. M.:
It's powerful. But NVIDIA, like I say, they start with computer graphics and games. So they are very good in graphics and they are very good in rendering images. And now it's even generating from text new images. So it's a very good study. Did you get the link, E.?

43:22 - E. G.:
I'm pulling it up now. In the chat. Yeah, I've got the 4090 card. I just ordered the 5090. I guess you've got new stock out. Must be nice. Thanks.

43:38 - D. D.:
Thanks for rubbing it in. It's not rubbing it in.

43:44 - E. G.:
It's survival.

43:46 - D. D.:
It's survival because I actually run a lot about the 5070 TI with 16 gigs, it's an upgrade from my regular standard. 12 gigs. I got two 12 gigs on the server that I use. I've got a 16 I can do larger batch sizes. Take care. Bye. Bye for now and I will send you the... There's some news from Google I guess. I put a link in there. I've been reading that today where you can just came out today they anyway I put I shared a link if anybody's interested I'll read but actually Dr.

44:37 - E. G.:
brilliant on our Wednesday meeting you're able to take multiple Nvidia cards and group them together so you could have a model span memory of two cards and we're SLI but only use the memory of but the processing of both. This uses the memory of both, which is a huge leap.

45:02 - D. D.:
So what it said, say that again.

45:06 - E. G.:
Previously on the NVIDIA cards, you had to drop a special cable which would allow you to link cards together. Like Crossfire?

45:18 - D. D. & E. G.:
Like Crossfire, but NVIDIA. had their own called SLI.

45:23 - D. D.:
Yes.

45:23 - E. G.:
But in our Wednesday meeting, I forget his name, I'd have to look it up, but they're able to link cards together and actually use the memory of multiple cards so you're able to run much larger models. Like a 2080?

45:44 - D. D.:
There's no SLI on the 50 or the Are they even the 30 series?

45:51 - E. G.:
No, no, no. This is an SLI. From my understanding, it's actually linking the cards. As soon as I find out, because I'm going to beg my wife for a second 50-90 so I can go up to the 60 gigabyte model. You want to.

46:13 - D. D.:
Everybody needs to.

46:14 - D. B.:
I don't think you mean the way you don't mean the Wednesday meeting that I know of is not related.

46:25 - E. G.:
No, no, it's it's the multiple myeloma. He said that he was able to link multiple cards together. Who said that? The primary person who's getting me data access. Oh, at the medical school?

46:44 - D. B.:
Yes. M. B.? Okay. I think it's M. B.

46:49 - E. G.:
Yeah, M. B. sounds right. So, yeah. So, at that point, you can... I really want to know.

46:58 - D. D. & E. G.:
I do too. It's just, you got to play with it.

47:04 - D. D.:
You get to play with it Let's take him out to eat something That that would be there's ways you can make them run parallel But if you could extend the memory from the GPUs and so you if you had to 12 gig you could go up to 24 you did That that is significant What's the 90 what's the 32. 32 gigs? Yeah. Ouch.

47:34 - E. G.:
If he had two of them, that'd be.

47:37 - D. D.:
64, because most of the models that I'm dealing with now are 60 gig.

47:43 - E. G.:
Right now, since I have 256 gig, if you're able to, if we're digressing here, Dr. B., let us know, because.

47:51 - D. B.:
Well, I mean, essentially, we're not doing anything else on the agenda, so just, you can stick around and talk if you want.

48:00 - E. G.:
So with these large LLMs, if you want to go to larger LLMs that exceed the memory of your video card, it uses system memory. And that system memory, you can actually see how much slower it is. I had done a side-by-side where I ran a model all on my GPU, my 4090, which is 24 gig, and I 4090 on my Threadripper, which is 64 processors on my computer with 256 gig of memory. I ran that, and you could see the words coming in, dropping several seconds between each word. In the GPU, it was, you couldn't read that fast.

48:57 - D. D.:
Yeah, you were absolutely right when we did those experiments that running through the GPU is a lot faster.

49:09 - Unidentified Speaker:
I mean, it's measurable. It's been a while since I looked at the measurements.

49:18 - D. D.:
I was inferring some of the smaller models that you download the whole model and host it. I was hosting a wizard coder on my server and sending back, getting it to help me code. I was surprised at how well even the smaller ones could do. But I mean, you could get to the point where you can infer what's the 4.0 knockoff model that we talked about. I forgot its name, but you can download that whole model.

49:57 - E. G.:
We'll connect and I'll show you, but I actually run Olama now because for me to get into those other ones to run the NVIDIA, you have to go into WSF. I configure your Windows for WSF.

50:14 - D. D.:
It's not open source, is it? I thought Olama was specific for certain use.

50:20 - E. G.:
No, it's open. It's completely, it's open? Yeah, I've been running it.

50:25 - D. D.:
I looked at it, I was looking at it and it looked like it was like, you can use this for research purposes and, you know, that's it, or.

50:36 - E. G.:
Oh, that's all I'm using it for, so.

50:39 - D. D.:
Yeah. Yeah, it's, if that's all you're using it for, then. But I mean, it's not that hard.

50:47 - E. G.:
and make anything proprietary, I wouldn't think. If anybody would like to learn how to actually get a model and write Python to load this up, D. D. and I can host a quick programming session.

51:04 - D. B.:
Oh, you mean in this meeting? In this meeting?

51:08 - Unidentified Speaker:
Yeah.

51:09 - E. G.:
Outside this meeting? Good. Because we're talking about AI. Might as well know how to actually go in and code something to do it, how to load up a model, how to execute it, how to pass the parameters to it.

51:28 - D. B.:
OK, I'll put it on the agenda.

51:31 - E. G.:
So, Dan, let's get D. D. Let's connect and do that because I can show how to do it using and PyCharm, that's usually what I use for my IDE. And I think you use Jupyter, don't you?

51:50 - D. D.:
Yeah, I use Jupyter. Let's connect.

51:53 - D. B.:
To a session on programming AIs, would that be a good topic name?

52:01 - E. G.:
Programming AIs, loading up and loading up a model into a video. Card, and how to program. And it may please, Dr. M., we give a short presentation on how to do RAPIDS, RAPIDS.ai, and some of the functions on it.

52:30 - D. D.:
But yeah, we'll get out of just talking the theory of it and actually do a... Programming is somewhere like inferring them. Are you talking about training one?

52:47 - E. G.:
No, I'm actually talking about loading up one.

52:52 - Unidentified Speaker:
Right.

52:52 - D. D.:
And then where you can... Right. Then building out the predicate.

53:01 - D. D.:
prompt how to get the result from the prompt automatically. It's inferences.

53:10 - E. G.:
Inferences.

53:11 - D. B.:
OK, well, why don't we adjourn? Next week will be E.'s defense. And we'll go from there. So thanks, everyone. We'll see you next week.

53:29 - M. J.:
And D., I can think of some other people who would want to join us, probably for that course or whatever that little workshop that E. is talking about putting together.

53:42 - D. B.:
I'll make a mention.

53:43 - M. J.:
Yeah, so if I know when it's scheduled, I think I could bring some, there's some students and others who I think would want to join.

53:54 - Unidentified Speaker:
Okay.

53:55 - D. B.:
Mr.

53:56 - E. G.:
D., you still have my phone number, don't you?

53:59 - Unidentified Speaker:
Maybe.

54:00 - D. D.:
I don't know what I've got. Let's connect. All right.

54:04 - E. G.:
I will contact you across a medium soon. We could do it.

54:09 - D. D.:
We could do it twice if you have, depending on how much material you have, we could do it twice.

54:17 - D. B.:
And maybe the first time we could kind of test run it here, and the second time M. could invite a bunch of students.

54:27 - E. G.:
I think that's a better better approach because let's do that whatever is gonna go wrong it's gonna happen when you try to present it.

54:38 - D. D.:
So I'm not showing your phone number I'm showing I've got your email.

54:43 - E. G.:
Then hit me up on email let's connect and we just talk about putting together your cell phone and email. Actually, I don't mind.

54:58 - M. J.:
It is being recorded. Oh, I can stop recording.




 


at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)