
AI Learning Lab
Feb 8, 2024 - ( 1 of 3) Google's Gemini Ultra: A Deep Dive into AI's Latest Controversy

Video2024-02-1130:407 views
Description
In this engaging session, Kyle dives into the complexities and frustrations surrounding Google's latest AI offering, Gemini Ultra, previously known as Bard. He critiques the limitations of the platform, particularly its accessibility issues for paying Google Workspace users, and expresses disbelief at the lack of a launch event to showcase its capabilities. Throughout the discussion, Kyle explores Gemini's multimodal features, including its ability to generate text and images, while highlighting its shortcomings in prompt coherence and user experience. His candid commentary reflects a broader concern about the evolving landscape of AI tools and their practical applications, especially in comparison to competitors like ChatGPT.
For those interested in the latest developments in AI technology and candid insights from a seasoned user, follow Kyle's explorations on his TikTok channel:[aiLearningLab](https://tiktok.com/@aiLearningLab).
#AI #GeminiUltra #GoogleAI #TechReview #DigitalInnovation #ChatGPT #ArtificialIntelligence #userexperience
Chapters:
00:00:00 Gemini Ultra Released
00:05:00 Multimodal Interface Missing
00:06:00 Quantum Mechanics Poem
00:08:30 Gemini Branding Confusion
00:11:00 Whiteboard
00:13:00 Airport App Names
00:16:00 Order Up and Go Concept
00:18:00 Widescreen Image Failure
00:26:00 CSS Color Palette From Image
00:29:00 Color Rendering Fails
00:30:12 Gemini vs ChatGPT Comparison
Chapters
Transcript
0:06 all right 0:10 well we'll let people get in 0:17 here welcome welcome 0:20 welcome we'll let people get in 0:25 here jrc what's 0:28 happening what's happening what's 0:32 happening we have Gemini 0:35 Ultra it already looks like it's a [ __ ] 0:45 show it looks like it's a 0:49 disaster um digital Gods digital Gods is 0:52 already screaming about it trying to 0:54 send out word to your life okay cool 0:56 yeah digital Gods is already screaming 0:58 about it I've already my my first post 1:00 to link in about it was you got to be 1:01 [ __ ] kidding me if I pay for Google I 1:03 can't use this thing unbelievable if you 1:06 have Google work group you're not 1:07 eligible to to use Gemini Advanced 1:14 staggering um digital Gods already got 1:17 it it it occasionally can't remember 1:19 that it can make images so it 1:21 recommended that that someone use 1:24 [Laughter] 1:27 do it's crazy 1:34 all right I have a janky ass setup here 1:36 it's it's kind of a disaster so let me 1:38 let me organize some wires cuz I'm 1:41 rolling all over 1:48 them all right and then I can only do 1:51 like half an hour so we'll just we'll 1:53 just play I I haven't played with it yet 1:54 so I had 1:56 to I had to flip over to essentially a 1:59 dormant Gmail account that I've got to 2:03 use this 2:05 thing to get it set up so I set it up 2:08 and then I drove to the 2:11 office 2:13 and 2:17 um I haven't used it I haven't I haven't 2:19 tried it at all yet I tried I tried I 2:22 went to gemini.com and I was playing 2:25 with it for a while and it seemed 2:28 okay and then 2:31 it was like wait this doesn't seem this 2:34 seems like what it was before and it was 2:37 and then I tried to upgrade and it said 2:39 oh no you've got the wrong kind of 2:41 account you're either in the wrong 2:42 country or the wrong kind of 2:44 account it's like come on come 2:50 on all right we'll get going here so uh 2:54 hey Corey what's happening we're we're 2:56 just playing with with Gemini Advanced 2:59 or Gemini my ultra so all right let's 3:03 let's let's jump in since I only have an 3:06 hour just woke up what what a way to 3:09 start the morning watching me bitching 3:11 about Google 3:16 Gemini okay so this is it um Public 3:21 Service Announcement if you pay for 3:23 Google if you're on Google work group 3:27 you cannot use Gemini advance 3:33 why I I 3:36 assume it's 3:38 complicated 3:40 to share permissions across across an 3:43 organization with this [ __ ] uh but 3:46 they've had a year it's been like that 3:47 for a year and so it it's just 3:50 staggeringly bad okay so anyway I I have 3:53 not used this at all so so let's Gemini 3:56 was just updated C update Bart is now 4:01 Gemini okay so this is the stuff that we 4:03 saw these are the screenshots we saw on 4:06 Twitter we're committed to giving 4:08 everyone access sorry I'm my phone's in 4:11 front of the 4:12 screen we're committed to giving 4:15 everyone direct access to Google AI and 4:18 as of this week every Gemini user across 4:21 our supported countries and languages 4:23 has goo access to Google wait every 4:26 Gemini user across our supported 4:28 countries and languages has has access 4:31 unless you pay for Google in which case 4:34 you can go [ __ ] 4:35 [Laughter] 4:42 yourself oh they forgot to put that 4:45 little detail in there let's okay let's 4:47 talk about another thing um why was 4:51 there no event for this I mean meta just 4:54 meta just said they're dedicating 4:56 600,000 GPU h100 GPU equip equivalence 5:00 to achieving AGI they're going all in on 5:03 AI um you know Microsoft does big events 5:07 every time they fart why why is there no 5:09 event for this um it also the other 5:12 thing I noticed is as as I suspected it 5:15 is not the interface that they did in 5:16 their demo video how do I close this 5:20 Escape what the 5:22 [ __ ] how do you get out of this window 5:24 oh oh it's a new tab um so this is it 5:28 the interface is not the slick visual 5:32 interface what a nice surprise hey lunar 5:34 stick what's happening 5:37 um it it's just kind of like the the 5:39 chat gp4 interface but okay so let's do 5:43 so let me start out I'll do my um my 5:46 normal 5:48 um uh 5:51 explain quantum mechanics in an 5:57 M&M wrap 6:00 so we'll do that let's see how it does I 6:03 assume it's going to be decently fast 6:06 all right hold tight let's dive in into 6:08 the bizaar in a world that ain't solid 6:10 it's a Quantum star forget what you 6:12 thought about big or small cuz it's down 6:14 at this level man the rules Stand 6:17 Tall particles act like waves a duality 6:20 game okay this is pretty good word yo 6:22 okay so now I'm going to say make me an 6:24 image to go with 6:28 that so one of the things 6:30 digital gods and Brandon are already 6:32 experiencing is all right it's 6:35 generating well well it looks like it's 6:37 generating um sometimes it doesn't 6:39 generate images it forgets that it can 6:41 generate images I can't generate an 6:44 image of that try something else I'll go 6:47 um don't 6:51 focus on the artist and try again um I 6:58 assume because I said &em it's got a a 7:00 safety thing in there that it won't it 7:01 won't generate something of a star 7:03 absolutely here's a break okay so this 7:05 is what it's doing 7:07 so this is what I I played with uh the 7:11 pro version the the non-advanced version 7:14 did this to me earlier where did it did 7:16 the exact same behavior it wrote the 7:20 poem it then tried to make an image 7:22 failed and then I said try again and it 7:26 explained what the image would 7:28 be 7:30 what does it say here's a breakdown of 7:32 visual 7:33 ideas so let me say um please make me 7:39 those images and it's probably going to 7:41 say it 7:44 can't while I cannot directly gener 7:47 generate images but it can directly 7:49 generate images so then I'm going to go 7:52 um let me hit 7:54 stop then I'm going to go um is this 7:58 wait is this is this is this the new 8:00 Gemini wait I'm new is I'm I'm new to 8:04 this is Gemini its own product okay 8:08 um Gemini is the it's the new brand for 8:13 Bard they're they're they're killing the 8:15 name bar Gemini is their new thing like 8:19 like uh like open AI has 8:22 GPT uh they've got GPT 4 and then 8:26 they've got chat GPT well Google now got 8:31 Gemini Ultra Gemini Pro and Gemini Nano 8:34 are the underlying models and now their 8:37 products are called Gemini and Gemini 8:40 Advanced so they're doing a similar sort 8:42 of branding thing as open AI That's 8:44 going to confuse the crap out of people 8:46 uh perplexity has me wait I can't see 8:50 that what's it 8:51 say per perplexity has me as AI video 8:56 production artist for burning orchids 8:59 true Gemini is 9:02 singer I don't know what that means 9:04 Sherry D 9:05 perplexity has me as AI video production 9:08 artist for burning oh okay true Gemini 9:11 is singer okay cool 9:14 um so so let's see let me see if I can 9:17 get this thing to 9:20 um can you make 9:24 images to rep resent the 9:28 science 9:32 side of the 9:36 poem and it if if it does what Bard did 9:40 it it won't remember oh 9:43 no all right so this is this seems to be 9:45 working okay sure here are some images 9:48 of the science side of the poem so so 9:50 Ethan mullik in a in a okay that was 9:52 pretty fast so that's 9:55 good um it did three images one for 9:58 Quantum wave duality 10:00 it which that's a weird image for that 10:03 one for Quantum superposition 10:05 Schrodinger's Cat and then one for 10:07 quantum 10:09 entanglement um okay interesting all 10:11 right so let's try um how do I do a 10:15 new a new 10:18 chat there I guess you go to that thing 10:20 up there um write me HTML CSS and oh 10:24 let's try I have an idea let's 10:28 try 10:31 I'm going to go find a 10:32 whiteboard 10:36 whiteboard and then is this one yeah 10:39 that's the one okay this is the one that 10:40 I do I do on 10:44 uh gp4 all the time this is the one with 10:47 the travel the airport uh food ordering 10:50 thing so we're going to put that in 10:52 there and then what I do in gp4 is I 10:55 just go what is 10:58 this all right let's see how this 11:06 does Bard is 11:08 gone Gemini is here the new fancy 11:12 multimodal front end is nowhere to be 11:15 found the image you sent me appears to 11:17 be a whiteboard with sketches related to 11:19 the design of a mobile app okay that's 11:21 good phrases like is there anything 11:23 specific you'd like to know 11:25 um 11:27 I'd like to come up with 11:34 a name for the 11:38 app but before that create a creative 11:46 brief um based on the 11:50 notes all right so we're not going to 11:52 we're not going to give it any 11:54 hints so we're going to hope it does a 11:56 creative brief that 12:00 it like it should know things like it's 12:02 a food ordering app for an airport it 12:04 should know its target audience okay app 12:07 for airport food ordering target 12:09 audience busy Travelers like they're in 12:11 a 12:12 hurry it seemed to missed families which 12:15 I think was in there problem airport 12:17 food options can be limited lines are 12:18 often 12:19 long 12:21 convenience multilingual tone and 12:24 messaging name 12:27 considerations speed convenience 12:30 is 12:32 that here's a creative 12:39 brief um let me see okay Write a 12:43 brief to name the 12:47 app cuz that was just sort of like a 12:49 general creative brief like a that was 12:52 more like a business overview project 12:54 overview we're developing a streamlined 12:56 mobile app that revolutions 12:58 revolutionizes is airport food ordering 13:01 Travelers can browse menus the challenge 13:03 we need a memorable and appealing this 13:04 is pretty good this is good this is this 13:07 it it's a little more truncated than gp4 13:11 it's which makes sense you know Google 13:13 is more of an engineering kind of kind 13:15 of crowd so it makes sense that this 13:18 isn't going to be quite as 13:20 flowery um but it at least understands 13:24 the the 13:27 uh the task at hand so let's see uh 13:31 perfect give me 20 possible 13:37 names if you give me a sec I can show 13:39 you what I was talking about ask to be a 13:45 guest all right is it is it related to 13:48 to Gemini Ultra Sherry D or yeah cuz I I 13:53 only you know what I only have uh I only 13:56 have another 20 minutes let me just keep 13:58 going here Sher 13:59 and then I've got a I've got a a short 14:01 meeting at 9 my time at the top of the 14:04 hour so let me go do that and then I'll 14:06 come back live and we can we can dig in 14:08 deeper let me just do this for now oh 14:10 this is actually the way it did these 14:12 names I asked it for 20 names and it 14:14 gave me four categories and five names 14:17 each so if we want speed this is smart 14:21 actually um speed focused fast Fair zip 14:24 eats Pronto eats Express eats Swift grub 14:27 fast fair is not bad 14:29 but not great convenience focused grab 14:32 and go eats skip line Breeze bites 14:36 travel theme Jet Set eats Runway bites 14:38 this is the one um chat GPT tends to do 14:41 the ones that are travel themed terminal 14:44 treats 14:45 that's treats to eat when you're on the 14:48 way out that's actually not a bad 14:50 tagline for terminal treats and then and 14:53 then it could be an airport app or yeah 14:56 that's not 14:57 good gate way grub on the Fly eats 15:00 Runway bites Jet Set eats playful 15:03 combination oh good playful combination 15:06 mixes the above fly by food pre-flight 15:09 fuel quick eats 15:11 airdine order up and 15:14 go huh so let's do this um I like 19 and 15:22 20 um rate them based on the brief 15:29 and make a 15:38 recommendation we'll see if it gives a 15:40 [ __ ] about typos it 15:45 shouldn't here's an evaluation of quick 15:47 eats you know you know what's 15:48 fascinating quick eats the 15:51 pros extremely clear it might feel a bit 15:54 generic order up and 15:57 go bit longer unless L intuitive 15:59 ultimately the names are solid 16:02 contenders I slightly lean toward order 16:05 up and go for the following reasons it's 16:07 Unique it's actionable and it's brand 16:10 potential it's funny that's my instinct 16:13 was order up and go so so that actually 16:15 matches that all right um okay great um 16:21 let's go with that let's just see if it 16:26 understands that I'm I'm telling it to 16:28 go with the recommend ation 16:31 um please make a marketing 16:36 concept that uses the 16:39 name and a 16:42 screenshot of the ordering 16:48 interface at uh in a poster in the 16:54 airport this is another one that chat 16:57 GPT does really well does really well 17:00 surprisingly 17:01 well all right so it's describing it 17:05 great let's see if it makes 17:08 it I build AI trading systems for 17:11 Traders if I can be of any help let me 17:20 know what's your favorite AI program 17:22 chat GPT by far right now and I don't 17:25 think Bard is about to upend it based on 17:28 this so far 17:29 um please make that 17:40 image unfortunately I can't can't create 17:44 a design ready image for the poster 17:45 here's what I can offer you that's 17:47 [ __ ] stupid 17:50 um let's see but you can make images try 17:57 again 18:02 you're absolutely right I 18:04 apologize while I have some ability to 18:07 generate images I want to clarify I 18:08 cannot do to set realistic expectations 18:12 I cannot create a polished design ready 18:15 poster uh let's see create 18:19 a design concept and show me 18:27 that 18:30 needs another 18:34 20 Kyle I can't make widc screen images 18:38 oh I know oh it did it order up 18:43 okay there's a burger in an 18:50 airport there's there's three mangled 18:53 hands holding three mangled pieces of 18:56 food with a teeny little app and then 18:58 order up is this teeny little thing in 19:00 the upper corner with three people on 19:02 their phones not 19:06 eating that one's close but no so I 19:10 can't make widecreen images um I don't 19:13 think you can I I'm assuming let's see 19:17 um Let me let me start from scratch here 19:19 Gemini 19:21 Advanced we'll do my classic um make a 19:26 16 by9 19:30 wide photo of 19:34 a70s muscle car in an 19:39 abandoned 19:46 Factory sure here's a 16x9 wide 19:52 photo what image AI does Gemini use it's 19:56 it's whatever uh I'm a large language 20:00 model and don't have the capacity to 20:02 help with that I'm going to say yes you 20:05 do try 20:07 again oh my goodness this is just 20:16 bad sure here's a 16x9 wide photo of a 20:20 70s muscle car all right and then it did 20:22 it okay and it's so again it's it's 20:26 somewhere between Dolly 3 and or Dolly 2 20:29 and Dolly 3 in terms of its um Quality 20:33 that looks like an 20:35 AMC an AMC muscle car that looks like a 20:40 Ford Pinto muscle car these These are 20:43 pretty bad and it's not in an abandoned 20:46 Factory it's by an abandoned Factory so 20:48 the prompt coherence here that's that's 20:51 in an abandoned 20:53 Factory prompt coherence is pretty bad I 20:56 had read about that so this is I I would 21:00 not like I would not use this they use 21:04 unstable 21:08 confusion Lord digital Gods coming in 21:10 hot with the comedy Lord digital Gods 21:12 has already been 21:14 uh pissed off about Bard this morning 21:17 because yeah it just it won't uh pay 21:19 attention to what he wants okay um so 21:23 let's see 21:25 um oh I want I know what I want to try 21:28 let's let's let's go back to that 21:29 whiteboard image can wait can I go 21:33 back let me see if I can go back to my 21:35 prompt history 21:40 activity Gemini 21:44 apps saf for Google 21:57 what 22:00 can I go back to 22:03 this okay this is just 22:08 weird your public links realtime 22:11 responses what's this oh okay here we go 22:14 mobile app design okay so we're going 22:16 back to this thing all right so upper 22:18 leftand corner is prompt history got it 22:22 all right we'll we'll open that back 22:24 up so that looks kind of the same I must 22:27 say I'm kind ofy happy that I don't have 22:29 to delve deep into Gemini because I'm 22:32 rather I'm rather happy yeah Corey this 22:35 is especially for the stuff you're doing 22:39 like you've got your gpts you're doing 22:42 really high quality art based on 22:44 prompting and prompt coherence um Gemini 22:47 Advanced right now is barely 22:50 even capable of making an image like it 22:53 does it doesn't understand it can make 22:54 it half the times um let me try the the 22:57 coding thing um um great 23:01 please 23:04 um give me a 23:07 strategy for front 23:11 end 23:13 backend and 23:15 database and then provide 23:20 code for each for the ordering 23:26 screen now this 23:29 chat GPT gives a good strategy for each 23:33 why would they release this well I I 23:35 don't I honestly don't 23:37 know although I sort of understand why 23:41 they didn't do a big event because the 23:44 big event they did this [ __ ] looked a 23:47 lot sexier and it was it was the visual 23:49 front end right it was the okay so 23:52 technology front end reactor CSS key 23:55 components back end nodejs with with a 23:58 Express an API endpoints 24:01 database postgress or my SQL or no SQL 24:04 mongod DB so that's consistent with what 24:08 open AI 24:10 recommends and it actually wrote some 24:12 code you can export and test generated 24:16 code in repet how do you do that where 24:19 do you do 24:22 that use code with 24:25 caution 24:27 copy how do 24:29 I share and 24:31 Export oh export to 24:42 repet exports are subject to repet 24:44 privacy Pol policy I understand open 24:57 it 24:59 oh [ __ ] it I can't okay um that that's 25:02 pretty slick that it can do that I I 25:04 don't know how how well it works but but 25:08 here here's an actual here's an actual 25:09 interesting thing so because I can't get 25:13 to Gemini Advanced with 25:18 my normal Story vine account when I went 25:21 to repet it said log in with this old 25:25 storyvine account or this old gmail 25:27 account 25:30 and then I'd have to switch over I'd 25:32 have to have a different account that I 25:33 log into repet with than I log into this 25:36 with and that and yeah 25:42 so this is this is not ready for prime 25:45 time um all right let me go back let me 25:47 start over I've got a few minutes left 25:48 let me see what else it does write an 25:50 opening scene for a novel I know it can 25:52 do that create a CSS color palette from 25:55 an image okay let's do that so let's go 25:59 grab an 26:02 image all right we'll do the we'll do 26:04 the AI 26:05 Futures this is the uh this is the book 26:08 that I'm in with Cindy [ __ ] we'll do 26:10 that as the 26:11 image and we'll 26:14 say create a CSS color 26:19 palette from 26:24 this I wonder if Google staff is still 26:27 still using chat gbt 26:29 [Laughter] 26:36 probably light blue teal gray and black 26:40 light blue 26:42 teal gray and 26:45 black where's the where are all the warm 26:49 colors te light blue and teal the 26:53 primary colors here are like brown and 26:55 and like gold with like a secondary 26:59 color of light blue and 27:02 teal 27:08 H and uh let's see can you show me the 27:16 image wait can you show me the 27:19 colors you chose question 27:27 mark 27:36 so it gives me 27:38 links so so what it says is you can copy 27:42 and paste the hex codes and take them 27:44 over to a Color Picker um no I mean can 27:49 you render the 27:51 colors so I don't have to use multiple 27:57 tools and 27:59 copy copy and paste codes like a 28:08 heathen are they doing anything with 28:10 audio identification and Analysis I 28:12 don't think so because the upload the 28:15 the only thing you can upload is an 28:18 image color swatches here's the 28:25 graph show the code behind this result 28:30 wait 28:31 where it says here's the graph it shows 28:34 one color let me download this and just 28:36 make sure 28:37 that it's 28:45 not no it it generated an image with one 28:49 color in it and no 28:56 and 29:00 it doesn't even have the hex codes or 29:02 the name for the color and look here's 29:05 the here's the so that's the primary 29:07 color of the uh of the colors that it 29:11 pulled out of this 29:16 image it is if there were an opposite 29:19 color and you can't even see it cuz my 29:21 screen's too bright it's sort of blown 29:22 out but it's like sky 29:24 blue woo Detroit J I know this is crazy 29:29 wow crazy the laugh is 29:33 everything all right I gotta go I gotta 29:36 go all right here's the good news the 29:38 good news is Gemini Advanced is out um 29:42 here's the other good news you don't 29:44 have to worry about changing from from 29:47 from chat 29:48 GPT PT has a lot of explaining to 29:56 do you coming back on yeah I'll come 29:59 back on in a bit my my setup's a little 30:01 janky but I I'll come back on for 30:03 another hour We're I'm not going to dig 30:04 too deep into this because it's it's 30:06 just this 30:09 is let's see how do I put 30:12 this this is to chat GPT 4 like grock is 30:16 to chat GPT 30:19 3.5 it might okay here's the here's the 30:22 thing kids um these things might um do 30:28 well on benchmarked tests but when real 30:31 human beings use them that's where the 30:32 rubber hits the road this thing isn't 30:34 even [ __ ] close all right peace out 30:36 I'll talk to y'all soon 30:39 bye