and our entire team stays dedicated to caring for more than 3.4 million californians and counting. nationally recognized expertise. a whole team on your team. sutter health it's not just shag carpet, it's shag carpet and more. m is for their magnificent selection of flooring and carpeting. o is for overall price, value and affordability and outstanding warranties. r is for smg's reputation for a no surprise guarantee. they've been a respected leader for over 70 years. and e is for exceptional quality and expertise and their experience. that's s and g carpet and more. s and g more than carpet where quality is affordable. raj mathai moving the bay area forward. this week a fascinating interview with space scientist annie lennox, who made a discovery on mercury

and then discovered something even more striking right here on earth. plus, his ai running out of things to read. and what do we do if that happens? expert pablo villalobos. plus, a chatbot success story and an uncertain future for video game programmers. that's this week on press here. good morning everyone. i'm scott mcgrew. 2% of the craters on the moon and 2% roughly of the craters on mars are named after women. the rest, of course, named after men. and there's a fairly simple reason for that. the group that sets the rules, the international astronomical union, says things in space have to be named after generally speaking, scientists or artists, and they have to be dead, and their work has to be more than 50 years old. well, you see the problem here? a scientist or an artist who was working before 1974 is probably a man. he's

probably a man of european descent. annie lennox is a third year phd student at open university in the uk, and she joins us this morning. annie, your opportunity to name a crater set off quite a quest for you. you are trying to show the iau that these naming conventions are kind of unfair, right? absolutely, yes. this whole thing sprung about from me doing my phd project, which involves geologically mapping mercury's south pole. mercury is split into a variety of different quadrangles, and i'm mapping the one down there in the south. and i found this one crater that had really interesting ejecta that being the material that is excavated when an impact hits the surface of a planet. and so it had this scientific interest to it. and when you make new discoveries that have scientific interest, these need naming. and that's so you can talk about them. so you

can describe them better. so you better communicate them. and for this one it's also because it can be a target for future missions. the mapping projects in general are happening because bepicolombo the next mission to mercury, arrives at 2025. so part of our job is to map the area. but also look for targets for this bepicolombo mission. so i found this interesting crater and very excitingly got the opportunity to name it. and i felt like it was it was a bit challenging for me to name this discovery after a woman. i think that perhaps the rules that convene what we are able to name these things have some integrated implicit bias that make them unknowingly easier to name things after men. and as a result, we have a massive overrepresentation of male names across the system, the solar system. that's not unique to mercury. it's everywhere that we name things. and you can prove

that because one of the things as a scientist is you didn't just tell the iau, hey, it's unfair. you wanted to go with a list that said, look at all these names. they're all named after men, but that means you have to go through what thousands of valleys and rifts and craters and phenomena in planets and moons all over the solar system. absolutely. there's more than a thousand just on our moon. so when you look at all the planetary bodies, it's really is a massive undertaking. and that's why at the start, when i first started looking into this, i took a kind of gendered focus because even just trying to work out the gender of every individual that's named, that's as an individual, quite a lot of work. so it's prompted me to run these data entry hackathons. i've i'm organizing the third in this series, which is due to happen at the end of this month, coming up very soon. but the whole aim is to bring people from from across the world really together to look at the eponyms of these

named features and look into that individual and look for all sorts of diversity. while i've had a focus so far on gender, you can be rest assured when women are missing from the narrative. other forms of diversity and marginalized groups are as well. and so we're considering all forms of diversity with these hackathons. yeah, there there is not much diversity. i found a list. this is just craters on mercury. and this is just selecting from the b's bach, beethoven, beckett, botticelli, barry as in chuck berry. by the way, and brahms. cesar cervantes, cervantes, chekhov, chopin and copland. i even i know those are all white men. yes, absolutely. yeah. and it's it is the problem with the conventions is that in order for a real person's name to be used, that person has to have had 50 years of fame. they must be deceased for at least three years. so we're looking at a historical pool, but also they have to have demonstrable fame

or recognition. and when we look at the type of people who were allowed to achieve fame and recognition and status, that's not a diverse pool. and that's because these metrics are very much set by and set to benefit the patriarchy. and so while our rules are consistent with that structural imbalance, we're not going to have good representation. so what is it you're asking the iau to do? because you did send them an open letter? i did, yes, the open letter really wanted to point out that we do have this imbalance. like i said at the time, i focused on a gender imbalance, but i did allude to there being other sorts of diversity that are underrepresented and so i wanted to make that and make them aware of that fact and down the line. but i would love is if we can change the nature of the these conventions to try and take away that need for recognizing

historical celebrity, celebrity status or recognition or whatever it is, and try and make the rules as fair as as possible. so it's as easy to name a feature after any type of person than it is a cis white man. now, fortunately, looking ahead, we'll have plenty of women scientists to choose from. now that we're in 2024, there has to be a jennifer doudna crater somewhere in the world. are there still things to discover or did men get all the names? oh, totally. i mean, because men kind of came there as a first come, first serves, it does mean that we have this de facto male occupation of space. but space science is evolving at such an incredible rate that new discoveries are constantly being made for me on mercury, we've got the bepicolombo mission coming up, and i cannot wait for that better resolution data to see the surface as clear as we can.

and undoubtedly with that better data, we'll see more interesting things and more things that deserve being named. and of course, i need to ask, did you ever get your crater named? i have had some craters named, yes. so that first one that prompted this whole project, i ended up calling the nairn crater. it's after a scottish poet and singer called lady carolina nairn, who was a contemporary, actually, of robert burns. and she's a really interesting woman in her lifetime. it wasn't considered proper for a lady to be a poet, so many of her poems were wrongly attributed to robert burns, and it wasn't really until after her death that she got proper recognition. recognition. so i think there's this lovely story here of her not getting recognition in her own lifetime, and now getting recognition on another planet. i think that's really cool. and i've had a couple more since then. i've had i've named one after audre lorde in incredible poet as well, and also bessie

macnicol, one of the original glasgow girls. i'm scottish myself, so it's lovely to have a little bit of a little bit of myself up there. well, it's just it's a problem. i don't think a lot of people were aware of, and it's just an amazing an effort on your part and everyone else's to get that problem solved. and, and congratulations on everything that you've discovered. annie lennox is, of course, an astronomer. i do want to mention thank you, annie, for being with us. i do want to mention that i first learned of annie's research through the fabulous podcast the illusionist by helen zaltzman. the episode is called craters and i highly recommend really anything that helen zaltzman does. it's again the illusionist. we'll be back in just a moment. hi, i'm renovation expert scott mcgilvray. i know bath remodeling and i've seen all the gimmicks by now. half off free

installation. this is why i work with bath fitter. with 40 years of experience and 2 million installs. i trust their integrity. bath fitter gives you a fair price and sticks to it with a lifetime warranty. you don't need gimmicks, just a partner you can trust. bath fitter it just fits. go to bath fitter .com to book your free consultation. if you're an asbestos victim, pay close attention. billions of dollars are being paid out to people like you. we should know we're whites in luxembourg. the law firm that already won over $9 billion for men and women just like you. that's right. over $9 billion for mesothelioma sufferers and their families. we know you're short of money, and we know you're short of time. so we fight hard. call whites in luxembourg at 917 lawyers. that's 917 lawyers. whites in luxembourg winning big for the little guy. reignite your passion with custom anniversary bands from the jewelry exchange. two carat fancy lab bands, 1491 carat classic natural bands 990

half carat natural 499 unlimited choices guaranteed to appraise for double the jewelry exchange. redwood city. we keep seeing every night. the big ticker going up of all the money you save, whether it's work, family or finances. kris has your back. that's when i said this is not resolved. i'm going to call kris. you responded so quickly. nbc bay area responds with chris chmura. good morning everyone. i'm scott mcgrew. i suspect you know how large language models ai is trained. it's basically fed a huge amount of language books and magazines, newspapers, really whatever. they can get their hands on. the new york times discovered that openai was using their newspaper articles to train its chatgpt. google's gemini scrapes the web. elon musk's grok reads every tweet on the former twitter podcasts, youtube videos, wikipedia, you name it. you would think there's a nearly infinite supply of human writing out there, but in

fact, ai is probably already seen it. the new york times report meta was so desperate for language data it thought about buying simon and schuster simply so the ai could read all of its books. pablo villalobos and his colleagues at epic predict that ai will have scanned every piece of data. worthwhile information on the internet in the next 2 to 4 years. pablo, let me start with that. what happens if the language models run out of things to read? well, for a while we could repeat the same data we have a bit of time and that gives you some extra a runway to keep improving them. but at some point you add with the current paradigm, it would stop being useful at some point, and then they would stop improving. i think we've all joked about getting to the end of the internet, right? like you click on a webpage and you're like, yep, that's it. that's how you've seen all the internet. that hardly seems possible, but

i guess it shows how much i can absorb in one sitting. yeah, well, it's hard to imagine the a huge amount of data that is fed into these systems, but they are ever hungry for more and, you know, companies want to scale them as much as they can possibly afford to. so yeah, you've written you've written that maybe only 10% of what's on the internet is actually worthwhile. is that i mean, we all know there's some pretty junky stuff on the internet. yeah, that's right. most of the web content that you can download is very redundant, full of fragments of sentences or things that don't make really much sense. but the best 10% of that is roughly comparable to things like articles or papers. sure, facebook has access to instagram, and of course, facebook posts, but those aren't really all that instructive for ai. like, hey, look here are pictures of the salad i had for

lunch. or, you know, hey, we just had a baby. doesn't get an llm very far. well, it's hard to tell because usually models are until now have not been trained on this. a lot of social media posts and that sort of thing. and the newer models, we don't really know what they were trained on because the companies don't report that anymore. so yeah, intuitively, it seems that this would be less useful, but we can't know for sure. and there are things that i mean, companies really should not have their hands on. i mean, google would have a wealth of information if it could read all of my business email because that's, you know, intelligent stuff that an llm could learn from. but google has no business reading. my business email. yeah, that's also a huge issue, right? whether training on this data is infringing on your privacy. and it's something that's really being discussed now. so gpt three and then gpt

four, the jump from one to the other, the amount of data that had to go, you know, that was fed three. and then the amount of data fed four. and of course five is next. that's got to be just a staggering amount of data to make five better than four. yeah. usually it's a jump of around ten times more data. so it's basically impossible for humans to keep up with that rate of consumption. so tell me about synthetic data. i mean that's a possible solution, right? yeah. yeah, that's totally right. and it's probably the most promising approach. and what companies are surely trying right now. yeah. the idea is using the same models or previous generations of models to produce the data that will be used to train the next generation. so, you know what? honestly, it sounds gross, pablo, but you know what? that reminds me of is mad cow disease. when you feed a cow a

cow, you end up with bovine encephalitis. this is a an ai producing data that other ai would use as its data. just sounds like the beginning of a scary movie. yeah, and you know, it can go really, really poorly. like the ai feeding on itself. it can get lost somewhere and lose lose track of reality and start spouting garbage. so you need to control that process very carefully to make sure that it is actually improving the model. and not just leading to nowhere. so, i mean, i guess i guess bottom line, this for me, i mean, is i stuck can i go no further because it's, it's read all the data there is to read. i wouldn't say so. you know, there's billions of dollars at stakes companies are trying and researchers are trying lots of different approaches. and even if none of that works, we still

have steady improvements over time in efficiency. right. so you wouldn't see ai progress coming to a complete stop. well, pablo villalobos, thank you for joining us. i know you've got a lot of hard work being an ai researcher over at epic. pablo villalobos, our guest this morning. press here will be back in just a minute. you. there you go again. looking so handsome. you have your way. i'm doug hopkins with brothers by homes.com. and i want to buy your house. i'll make you an as is cash offer on your house within 24 hours. whether it's a

total fixer upper or in perfect condition. brothers by homes.com is the easiest way to sell your house. when you sell your house to brothers by homes.com. there's no fees, no commissions, no banks, and no repairs. just go to brothers by homes.com for your free, no obligation cash offer 24 hours a day. that's brothers by homes.com. hey claire-bear, i've got a surprise for you. yeah, babe, you know how i love our resilience collection, right? well, behold. football, football, football, football. hahahahaha! hahahahaha! love the room password needs a little work though. introducing the kittles collection exclusively from zenni. welcome back to press here if you're contacting customer service, it's because you need help with something you have not been able to figure out on your own, an answer you have

not found on the company's website, or it's something unique and complicated and an automated system. you know where you press one for new customer, two for existing. that's not going to cut it. my next guest runs an insurance company that's been experimenting with artificial intelligence to run its chat, help dustin, star of brightcove, says it works about half the time. dustin good morning. you know, we don't normally put people on television for something that works half the time. i think in in your case, i believe it's 58% of the time. but that's good progress for ai, right? absolutely. and you know, when it doesn't work, the other 42% of the time we have a customer service rep there to take over and help the customer. so we're making sure that the resolution rate is 100% at the end of the day. now, i can handle, you know, the easy stuff. how do i download my policy or something? it's the special circumstances where i really has to shine. and there's an example that i've seen of your customer who had

asked, i'm moving in a month. what address should i use for my quote? and then the machine understands it, which is impressive because there's a lot of, you know, subtlety. there's a lot of complexity to that question. yeah, absolutely. so we're training the generative ai. you know, on our back end systems that have all the information about how a policy gets formed. so in that situation, it was great that the bot was able to answer that question. and what's great is the bot can be there to help even the simpler ones, even if it's 2 a.m, right. and a customer service rep isn't available. well, and i'm sure the customer service reps love this. you know, when the bot can handle the simpler questions. another complex one i saw was you got a question for somebody who secretly gotten engaged, and you ensure jewelry as a business. and she was worried that she would get mail from the jewelry insurance company and give away your secret. you know, in that case, i mean, here's an ai that's that's really understanding, an unspoken part

of the of the conversation. you know, i secretly got engaged and the bot understood what its role would needed to be in that. yeah, absolutely. so, you know, as a jewelry insurance company, one of the main things that we do insure is engagement rings. and we never want to ruin the surprise. so one of the things that we can leverage the ai there is, you know, if the ai got it wrong the first time, right, a customer service rep can step in, but also we can train it to make sure that in those situations, the customer always gets the answer that they need. right. well, that's what i was going to ask you. what have you learned over time? you know, the learning is up and to the right because it's gotten better and better. but but what have you and the ai learned over, over these last few months? yeah. the important thing is to start it with lots of guardrails and sort of on the simpler queries. and then as it goes, we've learned that you can sort of open it up, allow it to really enable and provide that customer support and hours that normally, you know, a customer service rep wouldn't be working. and then we know that there's

always times that they will need to speak to somebody that you sort of need that white glove handholding touch. and that's well and that's you know to transfer it. yeah. and that's the, the definition right of customer service is, you know, as i said in the intro, i don't call customer service to find out what you know. what hours are you open. right. it's going to be one of those when the person at american airlines or whatever it happens to be gets on the line, i say, okay, i wonder if you've ever heard this problem before, but here's the issue. like customer service is all about the unbelievably complex, unique situations. even if you have seen them before, they're unbelievably complex and unique to me. the customer. yeah, absolutely. and so, you know, even though when sometimes an experience like the shipping and knowing whether your fiance could be notified about it feels so unique to you. there's many times that it's been, you know, asked and answered before. so just by providing that training data to the ai bot, a lot of times you can answer the question that feels so unique in

the moment. so is this replaced employees or contractors? do you anticipating replacing employees or contractors? the way we see it is this is just simply a way to enhance the current employees, you know, give them more power and allow them when they need to spend extra time with someone, not feel rushed to get off the phone or answer chats because, you know, they got emma or a bot in the background helping free up their time. fair enough. well, dustin starr, i appreciate you being with us this morning. dustin starr is the head of growth and marketing for brightcove. and as you heard him say, there ai is called emma. we'll be back in just a moment. i know a few things about a rivalry, but this one hits different. cowboys offense. very exciting. i give them that. but the steelers defense is so powerful. you want to run home crying. sunday night football on nbc and peacock. sunday night is always niners night on nbc bay area. after you watch sunday night football,

stick around for a deep dive into this week's niners highlights on sports sunday, presented by xfinity right after sunday night football. only on nbc bay area at sutter. caring never stops. our over 220 locations are close by when you need them most. our doctors are here to listen and to answer your questions. and our entire team stays dedicated to caring for more than 3.4 million californians and counting. nationally recognized expertise. a whole team on your team. sutter health it's not just shag carpet, it's shag carpet and more. m is for their magnificent selection of flooring and carpeting. o is for overall price, value and affordability and outstanding warranties. r is for smg's reputation for a no surprises guarantee. they've been a respected leader for over 70 years, and e is for exceptional quality and expertise and their experience. that's s and g carpet and more.

s and g more than carpet. we're quality is affordable. welcome back to press. here you are probably too busy when you're playing a video game to notice that the hard work that artists put into it, the gleam of a football helmet or the puff of dust from a near miss of a bullet, all of these effects were created by artists, human artists who recreate the real world on the screen. so if you wanted a character in your game that was, say, a knight made of gold armor, that is something you would need as a programmer to ask an artist to do until now. now i can just do it myself. create a medieval knight with gold armor and a few seconds later i get this with the helpful message let me know if there's anything else you'd like me to adjust. kaveh vahdat says the future of video game programing is that his company, rai angle, is in the early

stages of what they call text to gaming, creating a game simply by asking a computer to make it. early examples include these games for mobile, where the art the dialog are created, or assisted by ai kathy that could. that's amazing because that could open up game development to just about anyone, right? yes it should. and that's the goal. we basically have less than 20 million game creators in the world right now, and maybe like a few million only just professional game developers. now we have more than 3 billion gamers and most of them, they don't even find themselves thinking about like possibility of creating a game because it's like so complex. our goal is to enable everyone to create games, basically. and you know, when we're talking about very big games and, you know, maybe the technology isn't there yet, but red dead redemption two cost something like a half $1 billion

and took eight years because it painstakingly has to be put together by hand. but now we've got ai that would make an enormous difference. yeah, exactly. and then like, basically like games could take from a few months for a simple mobile game. two years, as you mentioned. and like very large teams, is a super multidisciplinary actually development, like artists, developers, testers, sound musi, story writers, rules of the game, game mechanics, etc. and then it's going to be become much, much simpler to do that. it has been happening though, like for many years with game engines and improvements of game engines. it's just we think that generative ai is going to be like a much faster pace in that direction. and you mentioned story writers. the technology you have out now to make, you know, the smaller mobile games, those the ai will create the dialog, it'll you give the

character some ideas to who it is, and then the character will take it from there. yes, absolutely. and then one of the concepts that is going to be very exciting for gamers is going to be this adaptive gaming, which can be adaptive stories, adaptive textures, adaptive soundtracks, or sound effects. in general. and these adaptive stories meaning that you are playing this game and then ai creates the dialogs on the fly based on your previous behavior. so every time you play the game, you see different characters, different stories, different events, and your game is different from anybody else's game. actually. kathy vara of rai angle, thank you so much for joining us this sunday morning. that's our show for this week. my thanks to my guests, and thank you for making us a part of your sunday. excellence in action is presented by stanford medicine children's health. thankfully, cures for childhood cancer have improved over the past decade, and at stanford

medicine children's health, we fight alongside children with cancer and their families. until every childhood cancer has a cure. at stanford children's, we use the latest research to help our patients. we offer more than 100 clinical trials, including some not found elsewhere. we lead collaborative efforts with hospitals across the nation to pioneer new treatments for our patients. our team includes doctors who have spent decades treating childhood cancer scientists who are developing new ways to fight it, and compassionate nurses, social workers and therapists who help support children during treatment. we provide every child with exceptional care, backed by groundbreaking research and hope for the future. learn more at chase center dot stanford children's dot org. hi, i'm scott mcgilvray. i know a bathroom models should cost because i've done hundreds of them. i use bath fitter in my investment properties because i know it will be worth every penny and

then some. after 40 years and over 2 million baths, they're accountable for everything from design to manufacturing to installation, all backed by a lifetime warranty. if you want a bath remodel you can trust with your money. it just fits. bath fitter. go to bath fitter. com to book your free consultation. need to sell your home fast. worried about the hassle? fix ups, listings, showings and surprise fees? they'll take too long and cost stress but you don't have to deal with any of that. let us help you sell your home quickly and easily. john buys bay area houses, provides an all cash offer for your home in minutes. we buy as is, no repairs, no fees, and you close on your timeline. just call 510462 6000 or find us online. bay area houses, sell your . narrator: when you see this symbol, you know you're watching television that is educational and informational. the more you know on nbc.