Erica Lee, Women of AI: Artificial intelligence development, large language models, and Women of AI

welcome to another AI wisdom pod with Erica Luna Lee with us today Erica is a lot of things uh had an interesting career uh but she's the nonprofit founder at wom of AI So today we're obviously gonna talk about AI Erica worked at a bunch of cool companies like upward ex Zillow Apple etc etc etc really accomplished very ambitious uh interesting background and honored to have her

today Erica can you introduce yourself tell us about what you're up to nowadays uh thanks for having me Charles um my na middle name I Go by Luna now but a little bit about me I've been doing AI most of my career I started initially in research at Harvard and Hopkins but I moved into the private sector both um kind of starting to start with my own and then I got the chance to work at Apple um and the AI group mostly on Siri language models and photo recommendations later with Zillow mostly on forecasting algorithms like the zestate and uh eventually at upwork which is known as the uh platform for hiring remote contractors uh leading kind of all Ai and data strategy so it's been great and I feel really blessed um right now

I'm working um as an adviser to a couple of startups in large language models and um doing some more Consulting work but it's been great to get more Hands-On and nerd out so I feel very blessed why so ambitious um I think you know ambition for me it's not something I do it's kind of a natural calling um you know when I was younger I was determined you know to get straight A's and be the president of all these clubs and you know I I really pursued a lot of art U my parents enrolled me in piano M um violin and a lot of instruments um outside of kind of academics um and then I also did taond do and track and all these kind of things um I think for me I have a

lot of Creative Energy so I like to be able to just focus it somewhere otherwise it'll eat me as I say why AI um I actually you know most of my career I got into as an accident kind of like how all discovery sometimes happen um I you know in in college a lot of people recommended me to follow engineering but more like mechanical electrical or chemical

more traditional ones and I saw software engineering as a growing industry ever since the doccom bubble so I actually moved to um Silicon Valley a few years after school and then I ended up living in one of those hacker houses where you pay $500 to like live in a bed with like little like kind of these curtains to split up and everybody's working on a startup or

uh you know they're an engineer at Google and um I ended up getting into AI because everybody I talked to and lived with was working in it there were tons of marketing and billboards and Tech talks and I remember Google had released their self-driving car was that small two-seater one that kind of looked like a cute little pod and I was just fascinated with the industry and

when I was about 13 um I went to Barnes & Noble and they were promoting a book called The singularities Near by Ray cwell uh so I sat down there and over several weeks I would go visit the Barnes & Noble and read it and I really fell in love with one of the six EPO or Technologies he mentioned which is AI so um I got really lucky to end up in the field and um I started my nonprofit really to bring other women um and people of different backgrounds into it is the singular near um I think it's near and probably closer than people are thinking um you know when I was growing up I remember having one of those flip phones and you would text by pushing the numbers sever veral times to

get to the third letter and now we have you know AIP pin is something released by Humane which is kind of like a smartphone that attaches to you and projects a light display you know so in less than 20 years we're seeing such a huge change so um I think we're a lot closer um so really preparing for that and uh you know seeing the next phase of evolution is that what I'm exit about Humane at first I was a Humane hater but then there was the video that they released two weeks ago and as a contrarian most people hated on that video by the way I was like this is one of the best video presentation I've ever seen amazing amazing amazing the so natural yeah so and I mean the production company did

or the media agency that they hired did an amazing job and to me one it was an amazing presentation two I got really bullish on the founders and tree even if the product um has a bunch of features I'm quite bullish on them and apart from the fact that alman's an investor right the M majority investor what do you think them um well I didn't hear about

them much earlier till recently but when I saw their product video I knew right away they were two ex Apple Founders I said the way the product looks the intuitiveness the Simplicity and the ecosystem and you have your own cell line and it's very much separate not a supplement to existing products um I could tell that it was very special so really excited for it to come

out I think they're shipping in January I really wish them the best I I really think they're going to change how we do personal Computing again as we moved from kind of phones with buttons to phones with screens now it kind of reminds me of the movie her if you've seen it where um he pins the phone onto a shirt it's like a magnet and it kind of goes from handheld to uh shirt wearing I just feel we're steps back in where we should or could potentially be especially Apple since you mentioned that company this year as I was traveling back from a Spartan Race Ultra of mine in Mexico I was in taxi and just wa watching you know the pesage by the window and I was like why can't I have smart glasses

tell me how much that company makes on a monthly basis you know I see all these PL yeah or this human what's her story you know how can I connect with that stranger with with the glasses and it seems that human is advancing Us in that direction why do you think they chose the the VIS the the audio and not the visual um yeah you know they have it's mostly audio like you talk to it and then occasionally you can see like the light display to change and do a little bit of navigation on a menu um my guess is you know the VR and the AR headset wave which came in the past five to sevenish years um a lot of people adopted that for more immersive technology so gaming and you know Microsoft

is working to do a productivity Suite with the the the new I think is the quest 3 or I forget the name of the new headset was Facebook but um I I think they're just moving it where it's a bit more natural um a lot of people I know use a voice assistant on their phone to set alarms or send text messages and on carplay you know I have't read messages and I just talk um so talking maybe is their approach to a more simple interface than reading a bunch of light visuals that might have interactions with like too much dust or there's too much background light and you can't see it or gets blurry um because usually when things are blurry or hard to read it causes nausea or confusion so language models specifically

these large language models know as llms uh they're getting much more better and faster than image and kind of computer vision technology so it could be just because it's more accurate and much more faster is that like online reality on uh well the virtual reality on Twitter LinkedIn is so much more interesting the the physical reality it seemen I mean I'm a creative person so sometimes I'm

in my head and I think of scenarios and and deeper stuff than what is projected in front of me but it seems that bolt world would need to to sync up and there could be an interesting opportunity for that I think Microsoft uh the their device called the Hol lands um I'm not too bullish on Microsoft but talking about Microsoft um what happened at open Ai and

what what do you have to say on uh governments governance and and board that at these uh Humanity changing companies yeah um you're referring to the whole Sam Alman and the Greg okay so first of all hearing that in the news um it was it's probably the most drama AI drama just Tech drama we've had in many many years um my recap in terms of what I heard was that you know Sam was called and basically kicked out of the board and then Greg his co-founder um uh basically resigned and then uh it was really the board who voted them out and some of the rumors are because Sam had warned them of AGI and that open AI was getting very close to creating kind of this independent more conscious uh creature as calls it

in the video not necessarily like a system or a device um and that you know he was kind of warning like let's take a level of responsibility with this and uh then eventually I think Sai announced he was going to hire Greg and Sam um because uh you know they were closely as Microsoft's a huge investor they put 10 plus billion dollars in open Ai and then

um it was announced that EMT from twitch the co-founder twitch was the interim CEO for a day and he kind of said he was going to do an investigation into why the happened and then Sam came back I think a lot of the employees petitioned to resign uh because you know he is the People's Choice leader and they wanted him back um so you know there's democracy

yeah yeah weird and anyway you developed a bunch of gamechanging uh ml related alos I would call these like Z Zillow estimates Apple Siri uh 2.0 upor search and fraud engines um what's next for you like what what is your next big uh contribution to humanity um I would probably say I'm I'm doing more work and research um not in the large language models but a lot

of the infrastructure that supports it so you know hugging face is one of the largest AI startups um you know hundreds of millions in funding and they have open sourced a lot of ml models and so has Facebook with llama and llama 2 and there's mistol so I think models themselves um are becoming commoditized with the ability to fine-tune them and then also do all the data

pipelining data annotation and just kind of data augmentation I really focus mostly on the fine-tuning and the data pipeline since I think those what really differentiate a company's ability to maximize their data there's bias in AI like I ask mid journey to come up with an image of an Olympian surrounded by its coaches and it the Olympian is a male surrounded by 10 males as coaches um

I'm not the best biased person that is you know like I mean I'm opening my mind to it and learning on a daily basis but what's the problem with that where could that potentially lead us and what would be potential solutions to that problem yeah um one of the problems with machine learning is it's it learns from the data provides so um you know years ago there

was kind of an exposure that you know black people on photos were getting detected as gorillas and they weren't being seen because their skin skin tone was much darker and the problem when they did some research into the Google ethics team was they didn't have a large enough data set of many people's variations of skin color whether it was brown or black or different types of caucasians

so um I think the most responsible thing is to increase the data set quantity and quality um so it's more inclusive and representative of the world that we live in and also you know to really encourage um better testing so when the engineering research teams develop these algorithms uh to actually kind of stress test them hopefully in like smaller beta groups and try to use kind of

every Edge case and see what are kind of the hidden corner is that when we release this publicly what we might see or not see and and I think if people are just a little bit more not cautious but they've been more more rigorous with their testing and rigorous with their data collection and data governance um we'll hopefully be able to see more inclusive and accurate results

but can't AI learn like where humanity is going and the the progress and you know like the yeah the the growth Trend in humanity and the fact that we're trying to be more inclusive we're trying to give people more freedom can't the AI detect that yeah um I think it's starting to detect that so with open AI you know when they release chat GPT if you ask

like um you know what are the most common populations or um e ethn ethnicities in the US believe or not Latinos are you know 40 plus perc and they're really growing in size and you know if you asked like show me a photo of like US citizens if it was all white people then it wouldn't actually be very representative so I think the more algorithms have access

to public data or data that is you know more representative to today so something published in the past year versus 10 years ago it gives you a bigger clue I also think you know there's a lot of social media in kind of user generated content and if you can look at all the images of what people are posting or articles or blogs that they're writing even though

they don't get a lot of widespread reach maybe they have 200 followers or 200 viewers a month um I think those are still critical pieces of content so um I do I do see some features where like when you get a response on chat gbd you can thumbs it up or thumbs it down um I think that feedback also improves the algorithm and um with Elon musk's new he rebranded Twitter into X um there's actually this like reply feature on people's comments and if somebody makes a statement and people disagree you can actually disagree and like add a reference check and like correct it and then you can up vote and down vote if you think that reference check is right so there's a little bit more user generated feedback on um verification of quality

and I hopefully think it'll prevent more misinformation which is the term used for fake or inaccurate information that's being widespread but do you want to accomplish with wom of AI um yeah woman of AI .org uh I always clud the.org because we're a nonprofit um is um you know we're right now rebranding and um relaunching kind of the content site which is the newsletter um I think

I want to change it from more of like a traditional Meetup and workshop Group which there's a lot of those educational groups and turn it more into a mentorship and like a media channel where we post content on women in their careers uh kind of helpful tips or or or strategies they've learned and do more kind of like memor style content and I think you know the

more women are out there and the world sees that they'll hopefully start thinking it's more normal and younger girls of Tomorrow will be able to see that and hopefully be inspired for alternative career so ideally I always joke my dream to you know the mission to be so successful we actually can close the nonprofit where having a woman in AI is not a minority thing or a

weird thing that we don't have to have to like advocate word it just becomes you know integrated and it's a balanced economy so ideally we can retire it one day and then I don't have to do anything what are your top goals for next year um for 2024 um you know my goals always start with um I kind of like M hierarchy period physical um my health so I'm at a point where been able to get very very healthy and kind of get rid of any inflammation or chronic illnesses so I really want to actually build more muscle so I'm doing a lot more strength training um and then um in terms of economic kind of work um seeing myself getting more involved in kind of the fine-tuning and the data pipelining work specifically now

that I've done more applied ml algorithms and ml infrastructure and then in terms of my community um probably host a big um fundraising or annual conference for women of AI to reconnect the community um and then for um probably family and friends to spend more you know kind of community dinners and time Outdoors now that I live in LA more access to the beach uh live music

and I get to know people more people in the Fine Arts I really like fashion um but I get to know the people in this community better very cool where can people find out more about you Erica um you can find me on LinkedIn under uh Erica Luna Lee I think my handle is Erica Luna Lee I did change it um and yeah you can shoot me a message or you can kind of put a comment that you heard me from the podcast and I'll be able to know