This content is the introductory meeting of the "Machine Learning Research Group" under "Real Good AI," emphasizing a commitment to rigorous, ethical, and community-driven AI research, distinct from traditional, competitive academic or corporate environments.
Mind Map
Click to expand
Click to explore the full interactive mind map • Zoom, pan, and navigate
Okay. Well, it's officially 2 after, so
let's get started. What a great turnout.
I'm really excited today to have this
meeting. Thank you for taking time out
of your weekend to talk to us today. Um,
before we get started, I want you to
know this is being recorded, not not for
any reason other than for anyone who
can't make it can still join us and see
what we talked about. Um, so, uh, before
we get into the group and introducing
ourselves and having some fun, um, I
would be admiss if I did not advertise.
Um, none of this organization happens
without fundraising, without being uh, a
a bunded organization. Um, so we are
having our first big fundraising push
before in in between this meeting and
whenever we have our next meeting. It's
at the end of October, beginning of
November. Um, if you know any streamers,
you know anyone who might want to
participate in this group, you want
future uh further details, we'd love to
provide that for you. Um, it's called
Full Steam Ahead because Steam, get it?
Okay. Um, but um, I just thought I would
advertise that as we get going. So,
welcome to the machine learning research
group. This is by far the most I would
say like hardcore
your science group that we have by far.
And so congratulations if you're here.
We did not let everyone in this group.
We thought that there was something in
your resume that uh pointed towards you
would be a good fit to help us do this
topic. So I don't want to hear anyone
this time saying, "Oh, I don't belong
here." We made sure you all belong here.
So thank you and we're excited to have
you all here. Um, so, uh, I am leading
this group. I'm Dr. Mandy Mskins. Um, my
husband is Bob who does Distractable
with Mark Plier. Um,
and, uh, yeah. So, the philosophy of the
group, right? So, this group is meant,
it will show less like immediate results
as maybe some of the other groups, but
what we want to work on and the problems
we're concerned with in this group are
the hardcore research problems. they may
or may not have specific applications.
Um, we're happy to help people. These
problems that we talk about and we work
on it as a group. Um, absolutely feel
free to we can publish papers. We can
not we can make blog posts. We can do
whatever we want with that information
and with the research that we get. It
could we could potentially help you do
towards your degree like that could be a
thing that we could work on together.
There are all sorts of opportunities
that could potentially work out if we
make this work. The point of this group
is to do hardcore research. Um, so I do
want to mention again, I see lots of
familiar faces from some of the other
groups we have. I know lots of people
signed up for a lot of groups and maybe
some of them are interesting and some of
them aren't or like it's too much doing
all of them. We let everyone into all
the groups we thought were appropriate
for them. Um, and we will let you self-
select what you want to do. So, if you
need to drop out of a group, no problem.
If you want to uh change your
positionings, that's all fine. Your
involvement is entirely up to you,
entirely what's interesting to you. Uh,
so as with some of the other groups,
there is no specific one goal we are
trying to achieve here. Now, the
research we will point you towards, so
I'm a statistician by trained. Um so uh my
my
uh bias if you will is towards models
that are interpretable and uh have
uncertainty quantification and are well
justified ethical models in that way. We
also have uh they will be here a little
bit later another member of the real
good team who is a computer scientist
who can also tell you more about that
angle and the way that they work ethical
AI in um the computer science realm. But
the the idea is there's almost nothing
that can be off
uh off limits here as long as we're
going about trying to solve these
problems with good faith doing good
research practices and we'll keep all
each other in check. Now, I've been in a
lot of research uh organizations.
Obviously, I spent a lot of time in
academia. I spent some time research
labs. Um the research culture can be
really negative and competitive and um
not very welcoming to anyone who's even
a little bit different. What we're
trying to do here at Real Good AI is
something completely flipped from that.
I would love this to be like a really uh
supportive space. Not to say we don't
question each other. We can question
each other all we want but do it in a
respectful manner and um uh everyone's
opinion can be heard. Uh I don't care if
anyone has a PhD or if you are just
working on your bachelors. Uh your
opinion should be heard regardless and I
would love to hear your opinion. Group
problems work way better when we get
different opinions as opposed to just
one voice. Um and we're all here to sort
of learn from each other. I don't want
anyone to be like protective of
something or keeping people out and um
all that. So, anything we can do to make
the culture not typically researchy and
more community- based, I would love to
do that. Uh and just all of that upfront.
upfront.
Love to answer any questions. But let's
go around the room and introduce
ourselves. Um, how about you say just
like a little bit about your background,
your research interests, and sort of why
you're here, um, what you're interested
in with Real Good AI and the m the
machine learning methods research group.
Um, I'll call people so that it's like
easier to go around. Um,
hey, so Craig, you can go first.
>> I can go first. Simple enough. My name
is Craig Kaufman. I'm the outreach
director for Real Good AI. So, um, I'm
actually the one person not qualified to
be in this, uh, call because I don't
have all of the same credentials, but
I'm the one that does a lot of the
outreach, sets up some of the online
fundraising type of stuff. So, um, if we
if you ever do any in-person
volunteering, you'll probably get an
email from me. Um, and I'm excited to
meet everyone and just hear all the cool
conversations. So,
>> what are your research interests?
all of the interest that you all have.
That is what I'm looking forward to researching.
researching.
>> Thanks, Craig.
>> Hello, I'm Cody Schultz. I'm a uh
undergrad at Ball State University for
their uh computational data analytics
program. Um I joined because I'm I'm
interested in interpretability research.
I haven't done a whole lot of that. Um
but I'm also engaged in a research
project with the US Department of
Agriculture. And I've really liked um
working with these kinds of uh
research problems, you know, like just
getting in to things. And I wanted to
look at more stuff. Look at more stuff
and and do good work for for people who
I do think are trying to do something
good with AI, which I feel like there's
not a lot of uh places out are trying to
do that. Um and so that's I mean that's
why I'm here. That's why I'm interested.
>> Hi, my name is Leslie Matthews. Um I'm a
Penn State graduate. Um my interest in
joining this group was like I always
knew AI was a big u game changer in the
job market and I didn't know it was
going to have such an immediate impact because
because
um the first time I ever was introduced
or understood the concept was in uh 2020
when I attended Google sandbox. Uh that
was when Google was launching or
announcing their first AI and a lot of
companies or at least a lot of people
that were attending were expressing
their interest in AI. How you made AI?
How what does it do? Does it reduce the
workforce and all that? And that brought
up my own questions of asking. It's like
that's like why do you want to reduce
the workforce? Because it's not really
meant to do that. It's supposed to make
the workload easier, more efficient. And
I don't have a lot of experience behind
AI, but I've been working with my mentor
in the last two years for helping her
develop her own app. and she mentioned I
might be very good in the AI research uh
field because I have Asperger syndrome
and um the way my mind works is almost
like a computer which is uh it's it's
own difficulties but I realized if I can
help with this I might as well help an
organization or whoever is most ethical
about it because in that case it can
kind of push back any of the companies
that Let's use the uh capabilities of AI
and it should boost it should boost
humanity as a whole to discover new
possibilities rather than push them back.
back.
That's it.
>> Great. Thank you. Glad you're here. Um
Ian Tidwell.
>> Hey, my name is Ian. Um I am a master
student, potentially PhD, but right now
just a master student at Georgia Tech.
Um, I am doing aerospace engineering,
uh, computational fluids as my, uh, kind
of thing I'm doing right now. Um, once I
get more classes under my belt, I'm
going to be moving to combustion with
fluids a lot. Um, but a lot of my
interest in AI comes from uh when I was
learning how to do uh case setup because
it actually does a decent job of uh the
current models do a decent job of uh
getting a a case setup for you. Uh it
can help you learn, but it's just not
quite there. And I think what uh we're
doing here, which is making it to where
we can uh aside from the ethical uh you
know uh implications of being able to
have your sources here, it points to
sources, it it's also very helpful to be
able to know where it's grabbing
information from uh in general so that
you can uh site back to it and validate
that it's correct and see where it maybe
went wrong. Uh so that's something I'm
really interested in seeing if we can
get to work. Uh like uh Leslie, I don't
have too much experience right now
behind AI yet, but I have um I have a
pretty I' I've been more interested in
it as time has gone on. So um it's
something I may want to incorporate into
a PhD if I end up doing one.
>> Absolutely. There's tons of uh AI going
on right now in aerospace for sure.
>> Oh yeah. Yeah.
>> Thank you. Uh William
Uh hi, my name is Will Starling. I'm
working on getting my bachelor's at the
University of South Alabama in computer
science with a focus on artificial intelligence.
intelligence.
Um through my time there I was given an
opportunity to work as a research
assistant under for like on the topic of
AI and I was able to write a research
paper on using AI in the teleahalth
industry and like I was incredibly
blessed to give it the opportunity to
present a poster for it at the le big
data conference and so that's really
given me an interest in like research
and academia and so like I AI is a
really interesting thing and there's
lots of applications for it but like I
strongly agree with
um, Real Good AI's mission statement
about how it's like it's not being used
properly and it's also like really
unreliable and so a lot of like we're
overrelying on something that's really
unreliable and so I really want to help
us be able to like make something that
is reliable and also like stop all the
unethical practices when it comes to how
it's trained and how it's used because
it really should be something that
benefits humanity as a whole and not
just make the rich richer and yeah so
I'm excited to be here.
Thank you. Uh Ian Blan maybe.
>> Yes, Len. I forgot to change the
spelling of my name again. I'm sorry.
>> You're fine.
>> I'm a I'm a biomedical engineer. I
graduated about two years ago. Uh during
my studies, I had a course on on AI on
CNN's and how they kind of work and the
math behind it. But it was always also
frequently uh described to me by
professors as well as kind of a sort of
a blackbox technology. We know we know
what we we know what we put in, we know
what we get out, but we don't know
what's really happening in between. And
that really frustrated me because
I don't know. I just really don't like
the idea of
us having created something a a tool
that the tool that has the possibility
to be the great one of the greatest
tools that we have since the invention
of the internet. But we have in a sense
a very poor understanding of how it works.
works.
And so one of the things that I want to
do as part of this group is both learn
more about AI, how it works, various
models and how they're constructed, how
they produce the results that they do,
as well as maybe
maybe
put some effort into creating different
models that can actually benefit
people both in the professional industry
and in the regular day-to-day life.
So yeah, the measure is just mainly
echoing what people have said here. It's
too much see too much unethical
practices, too much
lack of good being done with a
technology that seems to have so much
>> Great. Thank you. Uh Justin,
>> hey, I'm Justin Allen. I uh started out
at the Lawrence Livermore National Lab
and worked there for about 6 years. Uh
got to work on a whole bunch of
different projects. The biggest one was
like using large language models and
GNN's to analyze binaries. Um like
training my own on on op codes and
things like that. Uh was really fun. Got
to use the big supercomputers there at
the lab. Um, and just about a year ago,
I uh left and went to Meta where I work
on the friending recommendation team. So
when it pops up with the like people you
may know, uh it some of my algorithms
and stuff going on back there to try and
predict who you might want to be friends with.
with.
>> Oh, so you're who we can blame about that.
that.
>> Keep showing us the people we don't
want. No, just kidding. Go ahead.
>> It's if it's going wrong, then it's uh
not my fault. If it's working great,
then that's all me.
Um yeah, and I I one of my main reasons
for being here is a bit more selfish.
It's just that I've I've been missing
research ever since I left the lab and
uh want to do that again. And I I think
that the uh you know the work I I really
did mostly was with neural networks and
I think that they are uh very they're
being used very inefficiently right now
when we have to throw the entire
internet at an LLM and then after that
have it run a thousand times in a loop
with uh reinforcement learning just to
get you know one or two% better results.
I I'd love to
I I see that and I think I I could I
think I could do better and I I want to
do some research into that to try and
prove it.
>> Absolutely. Oh, love that. And I'm I'm
sorry we missed each other at the lab. I
guess I it's a big place, but um surprisingly
surprisingly
>> I heard your name, you know, like like
Jeff would be like, "Oh, hey, have you
have you met Mandy? She's doing work in
that similar field or something." And
but yeah,
>> so how it goes. Um yeah, so it's really
segmented. Everybody works on their own
individual research projects. So but
>> well glad you're here and excited to do
some research together. >> Um
>> Um >> Braxton.
>> Hey, so uh I'm Braxton. I have a
bachelor's in computer science. Uh I'm
currently working in industry, but not
really with AI or uh data a whole lot.
Um, but my interests are in to make like
DI uh like more transparent. Um, because
like Lon said, it's kind of a black box
as it exists now. Uh, and ideally to be
able to train it um in a more ethical
way so we can have something that's
because I don't think we're able to put
LLMs like back in a box now. And so I
think the only way to um move forward is
to make better models in the ways that
we can uh to make them like less harmful
to the world at large. Uh and that's for me.
me.
>> Hi. Uh my name is Chong Yan. Uh
I graduated with a bachelor's in data
science, concentration astrophysics,
minor in physics and a certification in Japanese.
Japanese.
I previously worked with Petronus
on handling their data stuff like
streamlining the data stuff and before
that I was also a researcher with the
University of Toledo. I was helping with
astro astrophysics research. So what I
did there was essentially uh we made a
machine learning algorithm to sort out
around 500 galaxies from the 500 GBs of
data from the hyper suprime cam. So we
were able to find about 500 polar
galaxies from the typical
typical two galaxies. Yeah. typical two
types but uh my current currently I'm a
online platform executive handling stuff
uh
my current interest on joining this
group is actually to
it's a bit late kind getting getting
stuck here uh but yeah my current
interest to actually join join this
group is to actually try to align myself career-wise.
career-wise.
So, uh and also meet like-minded and
more professional people who kind of
know what they are doing more than me
and just help build my foundation as
well as I grow. So, that's about it from me.
me.
>> Great. Well, we're thankful you're here.
I know it's really late there. So,
thanks for coming. And actually, the the
Hyper Suprime Cam, that's really
interesting that you mentioned that. We
actually our our team did a um when I
worked at Lawrence Livermore did a
project on that where we were using AI
to like select stars and galaxies. So
>> from that same data it's really fun data set.
set. >> Yeah.
>> Yeah.
>> Thanks for coming. Uh Daniel.
>> Right. Hi everybody. My name is Daniel.
Um I'm currently a junior at Penn State.
So what's up Leslie? Um uh my current
research interests are so my biggest one
is assessing bias and fairness in large
language models and also mitigating bias
in large language models. So using
things like fine-tuning um human
feedback, things like that. Um seeing
how effective they are. It's kind of
aligned with the paper that I sent in
the Discord, which you should join if
you haven't. Um yeah. Uh my biggest
reason for joining this group is
probably because I'm so swamped with my
actual labs that I'm in in college, but
I have so many other ideas for things
that I'd love to explore. Um including
like actually, you know, cuz right now
my labs I'm trying to assess the bias,
but I'm interested in trying to, you
know, do something about it. Um so
that's something maybe I'll I'll kind of
try to cultivate here, right? Because
it's a little bit I' I'd hope a little
bit less intense than than the work I'm
doing in my labs. Um, so that's going to
be very fun. I'm looking forward to that.
that.
>> Great. Depends on what you mean by less
intense. More fun, but just as much
work. So,
thanks, Daniel. Kevin.
>> Hi, I'm Kevin. Um, I graduated with a
bachelor's in computer science from the
University of Washington about 10 years
ago. And since then, I've been working
in a variety of big tech companies,
primarily Microsoft. um also interned at
uh Google and Facebook and mainly my
experience is in uh pretty diverse like
backend engineering. So I work on uh
operating systems uh and hypervisors
with Microsoft's HyperV team and also um
uh programming language infrastructure
uh at Google and a whole bunch of other
different experiences um across resume
and yeah my interest for joining uh real
good AI in this team particular is that
um I personally don't have a strong
background in AI ML or data science but
I do in backend engineering obviously
any any sort of software engineering
that doesn't involve like um like HTML
like JavaScript and stuff like um a lot
of different um programming and software
engineering. So I was hoping that I
could join a project that could um use
uh some of that infrastructure
engineering like you need to set up like
web services or do like performance
testing and also learn um learn about uh
data science uh in ML and asked for kind
of like my motivation to join real good
AI. So, I've been working in big tech
for a while and I actually left uh
Microsoft about a year ago uh because I
feel like my um
motivation, enthusiasm for the work has
waned uh over the years because I don't
feel like it's like really aligned with
my values. And so going forward, uh, I'm
interested in either returning to big
tech or switching into like algorithmic
trading and stuff. And then more seeing
the work as kind of like a means to make
money to donate to causes like real good
AI or um, givewell.org or is there's a
lot of different like charity
evaluators, but I'm really interested in
give will because they have like really
good efficacy uh reports and research.
So, so anyways, that's sort of like what
I'm looking at in terms of career and
with real good AI, I want to see if
instead of just donating money, I can
also um put some of my like skills to uh
my engineering skills to use as well.
So, uh, yeah. Yeah, that's, uh, that's
it for me.
>> Absolutely. Great. Thank you so much for
being here. We're excited. Um,
>> thank you, >> Jamie.
Oh, I'm Jamie. Um, I am currently a
master student in computational physics
and I'm here cuz like many of all of
you, I believe in the mission statement
of real good AI. Um, yeah, I think that
that's something we definitely need. And
outside of that, in terms of the
research, I'm really interested in
seeing the crossplay between the
frameworks that we'll end up using here
to construct something and the ones that
I'll use in my uh research and physics
to like model the physical world, to
model materials and various
and things like that. Um, so I'm really
interested to see what I can uh kind of
use and use from one in the other and
vice versa. Right.
Well, this is a great place to be for
like physics-based machine learning type
things because it needs to be
interpretable. It can't be a black box,
right? So, >> 100%.
>> 100%.
>> Hi, everybody. Um, uh, my name is
Sebastian Foss. I'm an undergraduate
senior at in computer science at Seattle
University. Uh I currently work as a
research assistant in computational
astrophysics uh specifically endbody
simulations uh and building high
performance computing clusters. Uh
there's a huge realm of opportunity
within um astrophysics for machine
learning and and in my most immediate
case um there's a huge need for machine
learning to develop um initial
conditions for these simulations. Um and
uh my main interest is going to be high
performance computing. Um I know a huge
part of the AI space right now is the
development of computing infrastructure
optimized for machine learning and
there's definitely a lot of implications
to that I think. Um and then another one
of my interests is um cyber security and
AI. I think that on an international
scale there's there are massive in
existential considerations for that. Um,
and uh, as for my motivation to join the
group, um, I've been watching Markiplier
for about 10 the better part of 10
years. Um, and ever since since I was a
little tiny little bitty kitty itty
bitty kid and um, and his video maybe
like a month ago um, about his like
criticisms of AI and how he was fun how
he was assisting with the startup got me
to to reach out. Um, so I'm really
excited to see what this group can do
and I'm really excited also to see how I
can contribute. So, thank you.
>> Great. Happy to have you and glad Mark
brought you in. I think Mark probably
brought almost everyone in. So, yay
Hello. Uh, I'm Angelina. I have uh I'm a
postgrad from UCSC and where I did my
mers in uh scientific computing and
applied math. During that time I had
done a thesis uh specifically on
filling in mass uh areas in ocean
satellite data.
Um I think that's my reason for being
here is I've of course worked with AI a
lot during my time in research. I've
seen a lot of um different models around
there, you know, CNN's um other types of
classification models, transformers, and
I see that there is a lot of potential
for machine learning to be used for a
lot of good, especially within very
scientifically heavy fields. Um so my
particular interest in being here is of
course when it comes to the more public
facing um side of AI, of course, what we
see more is generative AI. We see a lot
of outcry from you know creators,
artists, writers etc who are essentially
having their information who are having
things that they put on the internet in
good faith essentially being stolen
um for not so great things. And of
course a big part of that problem is how
much data machine learning needs that
leads to inevitable scraping of data. So
my particular interest of course lies
with um finding if there are ways to
reduce that data and to still have like
a machine learning model be effective. Um
Um
and in that way perhaps we can stop
relying less on having to to steal data
from people online. Yeah.
Yeah.
>> Great. That's a very lofty goal, but
let's let's shoot for it.
Uh hi, my name is Alice and I have a
degree in psychology and cognitive
neuroscience and hopefully starting my
masters in biomedical science soon. My
research interest is in autobiographical
memory, but more specifically
understanding it enough to hopefully one
day be able to map out specific events
in our memory similar to how people have
been trying to piece together what
sequence of DNA codons translate to
which proteins or functions. If you then
use that knowledge to engineer devices,
technologies that can tackle memory
related issues. But I do like just
learning about everything and anything
in general. I basically find everything
interesting other than business stuff.
Um and at the same time I just like
incorporating principles and skills of
various other fields like engineering,
pure sciences to even arts and design
into understanding memory rather than
sticking to just pure psychology
theories and stuff. And it's partly why
I wanted to join the team to learn from
everyone's different backgrounds and to
also hopefully contribute something
significant for myself of course. Yeah,
>> absolutely. Thank you for being here. I
I love the different perspective too
because so many of us I mean so much of
machine learning is this like philosophy
almost in how how we're modeling it. So
I think that your alternative
perspective will be really helpful. So
thanks for being here. Um, last but not
least by any means, men.
>> Hi everyone. Can you hear me?
>> Okay, great. I had to install Zoom on
this machine because I hadn't actually
used it. This is a fairly new computer. Um,
Um,
so, uh, sorry I'm late. Uh, my name is
Min. Uh, I, uh, have actually been
working with Amanda for
what, five, six years. Um yeah,
>> at least six almost seven. >> Gosh.
>> Gosh.
>> Yeah, I guess I guess so. Um yeah, six
and a half years. Um
we we used to work together at Lawrence
Livermore National Lab and I I still
work there. Um and uh so my uh
background is in the intersection of
high performance computing uh the theory
of computing and data science. Um so I
have a very mathematical um interest in
like machine learning and AI and as a
result I I find the the current like
mainstream of AI very unsatisfying. Um
Um
so uh you know but we've um you know the
two of us and some of our other uh
collaborators including AMED who is also
a part of real good but isn't here right
now um you know have done a lot of
really interesting work on uh like
scalable statistics applied to various
um uh like various applications
uh and I I joined um in a part-time
capacity in the hopes of continuing
those collaborations, continuing to do
good work and continuing to
uh use um
you know mathematics that you know
centuries of human minds have developed
to do something useful for people rather
than you know further privatizing wealth
in the hands of a handful of very rich people.
people.
Um, so yeah, nice to meet everyone.
>> Well, welcome everyone. So happy you're
here. I realize I didn't properly
introduce myself. You might know some of
this already, but I might as well. I'm
Dr. Mandy Mskins. My PhD is in
statistics from North Carolina State
University. My research interests are in
scaling a specific type of machine
learning model, which you'll be really
tired of hearing me say, um, but maybe
you haven't heard these words yet.
Gaussian process models. Just show our
hands. Who knows about Gaussian process models?
models?
Okay, so only a few. Okay, men. Yeah, we
know. Um, uh, but so I actually have a
jar over here that's like, hold on.
It's my GP idea jar, and I have to put a
dollar in whenever I only suggest GPS as
the solution to everything. But I really
do think that GPS are one of our biggest
saviors that are completely albeit
almost completely untapped in current
machine learning. Um, and I'll tell you
all about them. I could lecture you on
them forever. I don't think anyone wants
that. Uh, but the idea is that they they
only work mathematically for relatively
small problems. and my research in my
PhD tried to get away from it for a
while and then had more ideas and came
back to it again with a completely new
perspective but that's been my entire of
my career pretty much has been scaling
GPS to either large number of
observations and now large dimensions
problems um and so particularly they're
good because they have uncertainty
quantification and you can check the
assumptions and they tell you exactly
where they got the information they came
from so they're sort of the like dream
thing, but instead we use neural
networks because they they work really
well out of the box and they're just
like easier to slap on stuff. So my my
theory is that we can make GPS work in a
lot of these cases. So uh the other
theory that I sort of had that I
feel like is worth mentioning is that
most of the way that I feel like we
process data is a little bit incorrect.
um in that like for example if we're
going to do like a image classification
problem or something right they divide
the data into pixels and the pixels are
what goes into the machine learning
model and there's no extra
pre-processing like you probably use
like a CNN or something that does some
like convolutions or something like that
but the point being that the pixels are
your base data rather than um I think
like when someone would draw something
for example uh they wouldn't go pixel by
pixel coloring ing everything in to
color it. And I think that there's a lot
of comparisons between treating data
like it's what's easy for us rather than
respecting the hierarchy of what
actually makes sense. And that gives us
like the the birth to the fields like
called adversarial AI where you can like
move a few pixels and it completely
messes up your classifier or things like
that even though the picture looks
identically the same. Machine learning
models aren't learning what you think
that they're learning. They're learning
these like weird underlying texture
things. And I I would love them to think
like we think see what we see at least
in some of those ways. And that means
that really we have to think about how
we're putting our variables into the
data, not just throwing data in and
spinning a neural network. Okay, so that
was Sorry about the soap boxing, but uh
we created real good AI. I've been
talking about it with Mark for a while.
I mean, I've been having these
conversations with him. Uh, go ahead, Kevin.
Kevin.
>> Yeah, I had a question about what you're
just saying, like the type of input data
we get for art. So, one of my interests
is the game development in 3D modeling
and for example, the modeling program
Blender, every operation um that you do
in the user interface can also be done
by like the API by the Python API. So
like is there I'm sure there are uh I've
seen like gen generative models for for
blender itself actually like is there
research being done that actually like
collects the it could be like visual
like video capture or API calls that
artists do like you know when they use
like whatever Photoshop or Illustrator
or Blender as trending data. Um, so I'm
not familiar with any. If they're doing
it, it's not that open. Um, and I could
imagine some of these companies are
collecting that data and just not
telling us, but I'm not 100% sure on
that either way. Um, I do want to be
clear though, I use that as like an
example because that's something we can
all visualize. But when I'm talking
about like the applications, I think we
should be interested in. I think the art
community is a little uh, and by a
little, I mean a lot anti-AII right now.
And so for at least the time being, we'd
like to stay away from those
applications until we build enough of a
reputation that they know that the
models we're doing are not harming them
in the same ways that some of these
others or we demonstrate that things can
give credit and some until we show
enough positives. Um I I would love to
stay away from that very specific
application even though I did use it for
illustration. But I think that this
applies to like a lot of different
things about that that we're just
putting in what's easy rather than
what's actually natural to our brains.
>> Yeah. Yeah. I agree. Yeah. So you're
talking more about the the general
structure of the data and part just
happened to be exical. Yeah. Okay, that
makes sense. >> Absolutely.
>> Absolutely.
Um Oh, so I was going to give just a
tiny bit of background on the
organization Real Good AI as a
organization. And so I've been talking
to Mark about this for I don't know a
long time. We used to just have
intermittent conversations about AI and
about you know the science behind it. um
as I finished my PhD etc etc and the
past couple years um as my research was
I I had a re main research project at
Lawrence Livermore National Lab we made
major strides on GPS and no one was
interested to continue it and that was
very disheartening and so Mark took that
up and also I saw a lot of the research
culture being very wrong um not just
there but in academia as well and
thought that real good AI I could exist
as its own thing to do both the fun
aspects of uh research and and outreach
like how we do the streams um and also
continue to do this hardcore research
and work on the problems that we think
are important towards solving this
problem and spend instead of spending
half your life writing proposals because
I know that that is at least for me that
was one of the bane of my existence. you
write so many proposals and they they
don't even respond to them. And now with
the defunding of science, I I don't I
can't even imagine that they're going to
respond to any of them whatsoever.
Right. So, I do want to tell you what we
have the opportunity to do here is very
unique. Now, I appreciate all of your
time. Um we we have the opportunity to
do pretty much anything we can think of,
whether it be a research project,
whether it be I don't know Uh, I can't
even think of all the things it could
be. Whether we want to go to a
convention, whether we want to have a
convention, whether we want to do online
events to talk about research, whether
we want to do some educational things
about the research we're doing. Any
anything and everything is fair game.
And I want people in this group's voices
to be heard, not just mine. I know I'm
talking a lot today. Uh but uh the uh
the the big things that we want to talk
about next time and I just want to
preface this is we want to talk about
like directions and ideas and uh how we
could start working just like
immediately because I'd love to work on
that. And the second thing is how we
want to collaborate together because I
think it's really hard to have this many
people working on maybe one research
project. we might need to like subcategorize
subcategorize
into what things people are interested
in. So I'm thinking next group meeting
will either be if there's interest to
hear about GPS I'm always that's always
on the table or
uh we can we can do some brainstorming
come with ideas about things that we're
uh we'll collaborate on in the future.
So nuts and bolts before before we run
out of time here of my talking. If you
have not joined the discord yet, here is
the uh Discord server. Please join it so
we can talk offline. Um also uh we can
uh we'll probably have all of our
meetings from then on just in Discord
because it's easier. Um and so that will
be our like central volunteer locations.
As of right now, we're just letting uh
staff and volunteers in. It will
eventually be an open server, but for
now um it's special. And uh finally, I
know that this was a time that worked um
and seemed
to work, but here is a survey um to uh
fill out when we should meet next in
November. Um hopefully we can pick a
better time that works for everybody.
Um, I have just like a bunch of times
listed there. Feel free to fill out at
your will and uh we'll I'll post this in
the Discord as well so that people who
missed it can get to it. But um with
that, is there any anything wants to
talk about? We got through all the stuff
I wanted to talk about.
>> Uh I I have a question. Um >> yes.
>> yes.
>> If we want to learn more about GPS,
where should we start?
Good question. Um, let me see if I can
find. Um,
I've definitely done intro to GPS. A lot
of times I give a presentation and I
know it's been recorded several times.
Let me see if I can dig that up and give
here
>> I know there are several hosted
internally at Livermore, but they're not
uh they're not released, so I can't post
them, unfortunately.
Shouldn't there be one externally
Did you record I mean this isn't exactly
that, but did you record your uh any of
your Auburn
>> Yeah,
>> I thought you gave like a general
seminar at Auburn, which is like what
prompted that whole collaboration.
It was on my GPS though. It wasn't on
Intro to GPS themselves. Um,
this is less accessible than what I'm
talking about, but this is my favorite
textbook that is available online, so
you don't have to buy it or anything. Um,
Um,
the electronic print version. Um, uh, it
talks about using GPS as surrogates and
it has lots of code and stuff that goes
with it. Uh,
I think that this is really helpful for
people to start out with. Um, we can
have a meeting where I can talk about
intro to GPS like I'm happy to do that also.
also.
>> I would definitely volunteer for that. >> Okay.
>> Okay.
Yeah, I also be interested in that.
>> Okay. Well, I'll schedule it because I'm
actually doing that next week at one of
the Texas schools. Their SIM chapter
actually asked me to come give a same
talk. So, um I I know it's so silly and
I like to me statistics like GPS are the
unifying theory of much of statistics
because basically it's that you're
defining everything through a normal
distribution with a mean and covariance
and you can write them however you want.
So like all sorts of statistics models
fit into this box like linear models are
obviously in this box. Um uh generalized
linear models all fit in this box. uh
like uh longitudinal data analysis all
that fits in this box like it it can be
anything. It's just how you define your
coariance matrix. So um yeah very useful
in my opinion
and I can send those slides before that
too if people are interested. My slides
don't always have a lot of information but
Okay. Any other questions or burning
thoughts? Yeah.
>> Earlier when you had mentioned um the
data structure or the structure of like
whatever item being lost and how we pick
up the data. This may be far off base or
really on base. I'm not really sure. But
uh would conformal mapping be a good
like pre-processing assessing step since
that kind of bakes in structure preservation
preservation maybe
maybe
>> possibly. Uh is your conformal mapping
through through a neural network? >> Uh
>> Uh
I don't know. >> Okay.
>> Okay.
>> As a whole.
>> Yes. I love that idea because it might
not be possible, right, to collect the
actual information that we're interested
in. It might be some data
prep-processing. I've done all these
things. So like for example, like let's
imagine emnest, right? So that's the
data set where it's handwriting and
you're supposed to identify the zero
through nine um digits, right? Um, it's
always seems so silly to me that we
treat everything as an individual pixel
in that problem when really you would
draw something and there's like a a
hidden dimension over time. And so maybe
the pre-processing step is to embed it
in that higher dimension which is
similar to what you're talking about to
infer that other dimension even if you
don't have it so that you can imagine
what is connected and what actually um
shows that that description. So
absolutely but if if you fit a neural
network at the beginning of something
you kind of lose the uncertainty
quantification is the my only critique
of doing that. So if you fit something
where it's like a direct one to one
>> Yeah, I had a question about um so you
were talking earlier about the like
research culture and the need to like
constantly you know apply do like grant
proposals and stuff. So, so I I was in a
research child for a little bit, but
mainly I've been in industry and one of
the concerns I guess I had about um
organizations like real good AI is uh
funding because I guess like from like
an industry software engineer
perspective like if you want to do any
sort of like impactful work uh
especially with generative AI or even
like infrastructure for gener AI you
need join, you know, the companies that
have like billions and trillions like,
you know, in their war chest. It's just
not like without the without the
capital, your impact will be like really
minimal. So, I was wondering if you had
any thoughts about um like how to, you
know, actually get traction on really
impactful and more like ethical projects
outside of like, you know, like the top
schools or the the most well-endowed
schools and the you know, largest uh
most uh yeah, I guess the largest market
cap companies.
>> Yeah. So I think our approach is super
different than most traditional research
organizations. And so I guess the the
way I think about what we're doing is
sort of like crowdfunded is the way like
we're accountable to the public and
we're trying to get our support and our
financial support and everything buy in
from the public. So we have this
interesting line because we have Mark,
because we have Bob, we have this way to
communicate with like tons of people all
at once. And um that communication and
those communities can be like super
powerful both monetarily in like what
we're trying to do but also in terms of
change and in terms of asking people to
do things and showing people do things
the right way. Now in terms of the
actual research we will do at this
organization it will be small. I mean
there's no way that we can compare to a
hundred billion just given to some other
you know organization right. um the the the
the
relative money is very different. But
what I consider what we're doing is sort
of like proofs of concept because once
something is shown that the research is
there for people to do it one way, I
think that people will be excited by
that and move towards that as it is
communicated to large groups of people.
And so uh I wouldn't be saying that
we're in direct competition with the
tech companies. I would say we're
showing proofs of concept that there are
other pathways to get to the correct way
to actually deal with AI and try and put
pressure on them to do things better.
Like maybe we spend all our time coming
up with one feature that then they
implement in a weekend. Like that's
totally fine. That's a win. That's
something that they wouldn't have done
if we didn't exist, right? So we're
trying to move that needle towards the
ethical side. uh regardless if our
actual like infrastructure and like the
actual models we're running are these
big models probably not in the span of
what we're planning on doing.
>> Yes, Justin.
>> Uh along that that same vein, if we have
stuff that you know is like a little
more computationally intensive than just
our local machine, uh does the lab allow
us to collaborate with them? Because I
still have my collaborator account open
for like HBC stuff. It depends on what
we're doing. So we are on a proposal and
we are collaborating on uh we're there's
an LL project um that's about
uncertainty quantification with circuit
models um that's going on right now. So
if it falls under that and I can
convince the PIs that that makes sense I
think that there's probably something
that we could do there. Otherwise, uh I
think we do have access to like other
things and like if we need to buy server
time, it's not ideal, but if we can we
can either write a proposal or buy time
if we really need big computing to be done.
done.
I don't know if that helps your question.
question.
>> Yeah, that answered it. >> Okay.
>> Okay.
>> Server farm.
>> Yeah. Well, he's already offered that
up. He's like, "It's still doing my
movie right now, but once it's done with
my movie, you guys can use it if you want.
>> I do have a question along the lines of
funding. Um, I'm very much in uh
starting grad school. I'm very much in
my uh trying to find a fellowship mood.
Uh do you know uh companies or I guess
organizations outside of um you know
just like the NSF or the stereotypical
ones that would uh that I could get
fellowships for that would uh allow me
to put more time toward this project specifically?
specifically?
>> That is a really good question. Um, you
know, normally I would say like like the
Department of Energy Labs, but uh I know
that their belt has really been
tightened these days and they may be
difficult to get, but I believe their
fellowship programs still exist. I don't
think that they've ended. Is that
correct, Ben? >> Uh,
>> Uh,
they still exist. Um
Um
I think all bets are off while the
government is shut down and until we
discover how that will resolve >> uh
>> uh
they have definitely at least like the
the fellowships that I
like I'm on the review committee for a
few of these and like those have had far
fewer candidates considered over the
last year or two. Yeah. But that's
that's largely a consequence of some um
um
a very high level a very high level
budget realignment that's been going on
for the last few years because of a
major effort called the exscale
computing project which has now ended.
Um and like the the realignment for the
computing part of the national labs
has been awkward as that has ended
because that was supporting like a huge
chunk of the national lab system for a decade.
decade. >> Um
>> Um
and uh yeah like the these huge
organizations are very slow to pivot. Um so,
so,
uh I don't want to say that there's no
hope. Um but certainly until the like we
certainly until the like the the the the
the shutdown
um is resolved
like there there's there's going to be
no movement on um fellowships any
fellowships involving any kind of
federal money. So that includes things
like NSF and NIH.
>> Yeah. Um,
>> so also I don't want to sound um crazy
here, but um if you have a specific
idea, right? So like let's bring it to
the crowd. Let's try and get it
crowdfunded. Like if there's a
difference between you doing research
and not doing research and the
difference is money, like let's design a
t-shirt that supports that organization
and gets you money. Like I I am totally
serious. This is a completely different
way of thinking about it than we have
ever thought about. I personally have
ever thought about it before, but the
sky is the limit in terms of what we
have here at Real Good. It's kind of a
different backwards thing because we
have the crowd first before we have any
of the rest of this. And so people are
willing to put forward if they know
exactly what they're they're uh
supporting and maybe they you'll go
along the way or maybe we can get like a
business involved like maybe we can get
I don't know
any business to like if you'll give a
talk about what you're doing to them
like talk about the ethical AI maybe
they'll do it like maybe we can work out
a deal like maybe we can split your time
with something else. There's all sorts
of like creative avenues and I really
think this is going to be the way moving
forward not just for us but for research
in general. I don't think it's going to
be so narrowly funded through NSF NIH
anymore. Um and I think we can be one of
the leaders of changing that model. So
>> yeah, we we've had some wild
conversations about like a a streamer
model for academics basically.
>> That's funny. Actually today in a couple
hours we're meeting with our streamer
group. So I'll tell
>> that's awesome.
>> We already have a project for you.
>> That'd be great. I mean that I know
earlier I mentioned I'm masters kind of
in limbo between masters and PhD and
that's that's definitely one of the main
reasons and waiting to see if I and I I
did apply well currently applying to the
NSF fellowship but it's it's like you
said it's not very uh it's not very high
hopes on that. There aren't like TA
ships. Usually teaching is
>> there are but again it's still my even
in Georgia Tech our research group is
having fun having trouble getting
funding. We had a couple of our projects
get u sliced. So
>> yep happening everywhere. >> Yep.
>> Yep. Yeah.
Yeah.
>> Unfortunately a grim moment to be a researcher.
researcher.
>> It's an interesting time to do it.
>> But they will not squaltch us out. We
will continue doing the right things by
the way. So, well, thank you all and
thank you for your questions and your
interest. Please join the Discord. I'll
see you over there and fill out that uh
form and we can talk about when we can
talk again. Um I am moving between now
and then. So, that's why it's a little
bit later in November. So, yay. But, uh
uh thank you all so much and uh look
forward to working with you. Let's talk
brainstorming and I'll schedule the GP
thing. Let's do that also. I'll look at
the times. So,
>> thank you for having us.
>> Thank you everyone very much.
>> Have a good weekend.
>> Have a great one. Yeah. Bye. You too. Bye.
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
