The Gemini Lie

Google's new LLM and ChatGPT competitor Gemini has faced some backlash after it's demo video was revealed to be highly edited. Learn about the latest Gemini drama and take a closer look at its benchmarks.
#ai #google #thecodereport
💬 Chat with Me on Discord
/ discord
🔗 Resources
Gemini Prompting developers.googleblog.com/202...
Hands on Demo • The capabilities of mu...
Gemini First Look • Google's Gemini just m...
🔥 Get More Content - Upgrade to PRO
Upgrade at fireship.io/pro
Use code YT25 for 25% off PRO access
🎨 My Editor Settings
- Atom One Dark
- vscode-icons
- Fira Code Font
🔖 Topics Covered
- Gemini AI controversy
- Was Google Gemini video fake?
- When will Google Gemini Ultra come out
- ChatGPT vs Google Gemini
- GPT-4 vs Gemini benchmarks
- Google Deep Mind advancements

[@OtterSou]

2:10 Each hand denotes a binary digit, where rocks are 0s and papers are 1s, and each row corresponds to an ASCII character The secret message is "GEMINI"

[@Pine_Wood]

this is gonna be top comment mark my word

[@cryda4069]

Quick af

[@evangreavu9621]

Good bot

[@ApatheticPerson]

👆 that's why AI is far from human intelligence

[@ChristofferLund]

@@Pine_Woodoh, hi mark

[@lorenzo6341]

Every Fireship title is a rollercoaster of emotions, after this one I went from being unemployed to still being unemployed but feeling like I may get employed

[@gustavo9758]

Pretty cool of Jeff to actually read comments that point out issues. A guy mentioned the apples vs oranges thing in the previous video, and Jeff listened.

[@AvihuTurzion]

Wait, you guys are employed?

[@lorenzo6341]

Failed startup gang rise up!

[@SodaDjinn]

That's what you get when you parrot daily news about a new/upcoming tech. Especially when said news only parrot whatever some PR person claims. Always take tech news with a couple grains of salt. It's all clickbait for sweet advertisement money, especially on KZbin.

[@uxmishi8767]

*high five! fireship is the vibe ngl

[@smix8780]

As a fullstack Brainfuck developer I am not worried about this AI comedy show. Good luck, guys

[@michalooo3425]

Are you claiming to be more autistic than AI?

[@WomboBraker]

yeah, i wish ai could take my fucking job

[@user-dn6kj8xc7r]

Are you worried about the possibility of circus monkeys replacing you?

[@Lauwit]

Well you should

[@mqsrafa]

Lul

[@alpineswiftai]

These are the same guys that YEARS ago told us they created a super intelligent Ai that was too dangerous for the world to see. Yet when the pressure is on and the floodgates open, we get Bard. 😂

[@Slav4o911]

Yeah some people though they have AGI... their AGI... It's like that old joke... "Mom told me we have the same at home...."

[@IceMetalPunk]

@@Slav4o911 Us: "We want AGI!" Google: "We have AGI at home!" AGI at Home: "If you put a ball in a cup and flip it upside-down, then pick up the cup, the ball will remain in the cup due to gravity." ~Bard's actual response to one of my test questions

[@WalkMrJ]

Remember when google told us they were less than a year away from developing a quantum computing chip? Good times.

[@Feedfront]

Bard is a joke

[@Navhkrin]

Google has good history of releasing "fake papers". Basically, they make bunch of BS claims but when you take their resarch and do 3rd party analysis it always ends up being useless shit that is worse than alternatives. And this has repeated itself so many times in basically every field of AI.

[@Enchufe92]

If there's one thing I've learned from all these AI newscycles is that hype runs the tech world. They NEED to exaggerate, if not straight up lie about, the capacities of their products to get into the next high of the hype cycle before it comes down and stabilizes at much lower value. Or worse, it comes crashing down. So yeah, even if Google is open in their blog about how fabricated their video is, the fact is they know most people and investors are not going to read it and are going to take the video at face value. Like many of us did.

[@peanutcelery]

Almost every tech company is like this tho, but norms are catching up to this exaggeration

[@seyanbryan520]

They might Aswell say '' in the future we believe our products will be able to do X Y Z but at the moment it's just another char GPT clone".

[@daniel4647]

I doubt it, I think most people have lived in reality long enough to know that this is always being done, and especially investors will read the small print. We all know how adverting works, and this wasn't even technically a lie. It's like the robot videos that are sped up, it takes the robot like 10 minutes to put the pen in the drawer, we see the 30 second video so it looks better and we'll actually watch it. But the tech is still super impressive and worth the investment. The only thing Google did really wrong here was saying "look how much better than GPT we are" when in reality it's about the same and about a year behind.

[@user-yl7kl7sl1g]

Google does this because they are a bunch of talentless shills. Open Ai, and Meta to a less degree don't need to do this.

[@ww-pw6di]

@@peanutcelery Not every tech company is like this (not even close to most either) and the norms aren't catching up to the exaggeration either. I'd bet my left nut and 20k usd that for every tech company that you can find that misleads to an egregious degree (like Google's Gemini bs), I can find 10 that do not (do note I can just pull up like 100k wordpress companies that do indeed deliver what they promise). Also every AI youtuber was dick over butt for the *clearly* falsified gemini video that was *obviously* bs because in order for Google to have achieved what they claimed, they would have had to not only unlock gpt5 tier *multimodal* inference with 1B speeds, but a whole another step forward to gpt6 as well because in order to achieve the kind of contextual awareness that they claimed (*on model level* and not choreography or preprogrammed behavior) then we're talking straight up human brain simulation or close to it (on model level, not fancy supplementary rag/memory trickery). And that's so god damn painfully obvious. The conclusion is that the AI youtubers are just too uncaring, dumb and/or clueless to know any better. They make videos and that's it. They simply cannot ever reach a point where they might be able to discern what is bs. That then get's extrapolated to all the other subdomains from hvacs to makeup to detransitioning and people are still going to be stupid as fuck. Individuals might better their brainways for a hot second, but en masse, nothing will change. So, wrong and wrong.

[@konignickerchen7265]

The only AI news that gets me excited nowadays is open source stuff because that's the only thing that actually results in worthwhile follow-ups within a reasonable time frame.

[@Slav4o911]

That's the only thing which gives actually good results, because we all tinker with the thing and collaborate. 100 000 humans are better than 10. The more humans interreact with a model, the more intelligent it becomes, look at what happened with Stable diffusion, and how it beats everything in the image generation.... and now it even starts to do video generation. The moment we have a model like stable diffusion with the chat models, most these corporate closed source models will be gone. Google doesn't have much time, if they want to do anything they have less than an year. The open source models are becoming better by the day, people mix them and enhance them.

[@phokeydhoo8940]

Also all of the companies making AI behind closed doors are trying to monetize it over cloud services. Personally, if you can't run it on your own hardware, I'm not interested in it.

[@TheDragShot]

​@@phokeydhoo8940I wholeheartedly agree with you. At least open source AIs do come with downloads so you can run them on your PC instead of waiting on a 1K person queue for hours.

[@sotirioschortogiannos4363]

Hmm interesting ive been trying to influence gpt but i think i need to work on an open source project. Imagine some random person make AGI before these big corporations and its open source. Would be a gift to the world

[@stuart6478]

the future (of software) is open source

[@Create-The-Imaginable]

By the time Gemini Ultra launches in 2024, OpenAI will have a new version of Chat-GPT out!

[@jamessderby]

And then Google will be finishing their next product and the cycle continues

[@ky00j1n6]

​@@jamessderby A cycle which brings us automatically more and more towards skynet!

[@vaisakhkm783]

​@@ky00j1n6i am sure skynet is already operating without anyone knowing

[@jamessderby]

@@ky00j1n6 love that for us

[@berkertaskiran]

​​@@jamessderbyGoogle is still finishing Ultra. How do they finish another at the same time? That's some Gemini level reasoning!

[@RegularRegs]

I really appreciate you breaking down the details of 5 shot and other more advanced things that aren't obvious to the layman

[@hammad8766]

I used bard with Gemini today and it was still giving some made up class names for the code sample and when I tried to give it some hints , it started hallucinating just like regular bard

[@jojo-gg1iz]

Yeah lowkey still trash

[@theresalwaysanotherway3996]

there are 3 versions of gemini, you used the middle version, the best version is still behind closed doors ATM.

[@ocanica3184]

​@@theresalwaysanotherway3996Like he says, I'll believe it when I see it

[@BooleanDev]

waiting for gemini ultra, the gemini version you tried is still worse than gpt 4

[@EGarrett01]

There's no reason to expect Google to be able to top OpenAI's product. OpenAI has a huge headstart and this is all they do. Their product is the best on the planet. Google is just desperately trying to get back marketshare.

[@Dan_The_Man_YT]

0:40 to all of you wondering what "kakashka-class" means: Kakashka is translated as little poo from Russian xD

[@ivanivanovich9165]

Yes! And more over I couldn't find the information about the kakashka-class bomb.

[@iscifion7122]

I still feel like even GPT-3.5 in ChatGPT performs better than Bard with Gemini Pro for regular tasks and convos. I don't know but it feels like GPT understands what I want better.

[@ibendover4817]

"I'll believe it when I see it" is really the key learning lesson from ai news over the past year. Fireship is honestly one of the few rare tech youtubers to emphasise this and look at real world examples. Looking at my recommendations as i type this and theres about a dozen other youtubers doing BJ/meth addict faces in thumbnails saying gemini will make everyone unemployed/kill openai.

[@harshmehta329]

see youtube is all about human attention. you are most likely to open a video when you are most terrified. that is all they are doing

[@ibendover4817]

@@harshmehta329 And yet many people clicked on this video instead. Theres a lot of people on the Ai hype bandwagon reporting on the same thing and many of them have no clue what they are talking about, just reading off news articles form a single source. Between clicking on a bj face thumbnail and a non bj faced one, I'm going to click on the latter that relies on their reputation of providing good accurate entertaining content rather than clickbait thumbnails & titles when it comes to news about something I want to learn about.

[@some______guy]

Google effed up again. Unreal how poorly they are handling the whole AI-thing

[@maciejbala477]

they're just desperate. They know their site is the first one to be replaced by ChatGPT. Why Google stuff now if you can just look it up with AI?

[@putther]

you got my attention after I heard 1 ball 3 cups

[@thymos6575]

putther is cooking something

[@torarinvik4920]

2 balls 1 mouth :D

[@dopesick4474]

Neuron Activation 🧠

[@KrunoslavSaho]

2 girls, 1 cup

[@torarinvik4920]

@@KrunoslavSaho 2 men 1 horse

[@paterofmater1690]

watching code report and couldn't understand things is my biggest motivation to keep learning.

[@luckerooni1153]

I have so much respect for you coming out with this counter-video after your very optimistic take in the first one. I doubt these motherfuckers even have a Gemini Ultra, they've just looked so goofy with all of this.

[@solaawodiya7360]

😂😂😂😂

[@GeekProdigyGuy]

They almost certainly have Ultra. They plan to charge people for it after all. But seems like on a scale of 3.5 to 4, it's looking more like a 3.9 instead of a 4.1, and of course, as we all know, 4 is the magic number ;)

[@meepk633]

Why would they put disclaimers about the results in when they don't actually have the model?

[@frackinfamous6126]

@@meepk633they were just disclosing the results were based on a multimodal hodgepodge but that’s everything. GPT-4/ChatGPT is in no way one all seeing model either. Tons of prompt engineering in the background and many different instances of different models working to bring you one answer. Not sure why that part was a big deal to anyone.

[@spoodermen2530]

Hmmm youtube didn't shadow ban your comment for saying mfer. Strange. I get banned everyday for saying far less. I hate youtube comment section it's extremely restricted

[@gdblockie8884]

2:13 Assigning rock to 0 and paper to 1, this now spells “GEMINI” in binary (guys ASCII is a representation in binary you do realise)

[@Ransom1337]

ASCII, actually

[@jeremybyington]

ASCII what you did there

[@maxave7448]

ASCII actually (continuing the chain)

[@insertcreativenamehere492]

ASCII, actually

[@STEELO-247]

ASCII, actually

[@Gigawattt]

In awe of the quality of your meme choices and the sheer meme-density of your editing. As always…

[@muhammadxojarustamxojayev7678]

"the same way advertisers and propagandists trick you everyday" got me

[@vectoralphaAI]

It got you the same way advertisers and propagandists get you everyday.

[@NeerajYadavGosu]

Iam turning my notification bell on. Almost every video of yours is very very relevant . Great work !

[@boccobadz]

Is anyone surprised? Alpha Zero (chess) was also like that - trained on TPU and pitched against an old version of Stockfish which was run on a single-CPU core. It's like comparing calculators from 90s to iPhone 15. Anyone who knows anything about computer chess pointed this out but nobody cared at that point (even though people should care about megacorps blatantly lying to them). It was never addressed and the reason was pretty simple - actual Stockfish (I believe version 9 in 2017), on proper hardware (not even server) destroys that iteration of Google's engine. And nothing changed in 2023 - Stockfish 15 NN beats anything Google can come up. I bet it's the same here - lots of marketing, little to no difference vs ChatGPT. There's no objective way to benchmark closed-sourced software. People are too easy to finesse. Does anyone remember superconductor claims from a few months ago? That's what I'm talking about - nobody knows basic sh*t and everybody acts like the biggest expert. World full of impostors is the world we're living in.

[@KyriosHeptagrammaton]

It was the same with Deepblue vs Kasparov. He had basically no sleep, was in an uncomfortable environment and not having a good day among other things.

[@Slav4o911]

At least they've beaten the impaired Stockfish, but we don't see Alpha Zero anywhere right now playing chess. Now they can't even beat the impaired GPT4, but they think someone is going to use their braindead bot... nobody is going to waste their time with that Google bot.

[@tomsear1]

@@KyriosHeptagrammaton Yes. Well said. (‘Yes, but.’) computation HAS altered upper hominid chess play post deep 🧢 hat - because access to such processing means binary adversarial Homo sapiens are both now playing against Turing machines at simultaneous scales. Ergo, Old Mate Magnus deploys more aggressive - read only human - tactics 2 execute strategy: Mow, Blow & Go; akin 2 latter, although Lee Sodol also deployed what in Australia we all a ‘Smoko’.

[@aaravtomar786]

I think they ran stock fish on the computer that Google bot uses and stock fish still lost

[@eyupylmaz9813]

"The only true way to evaluate AI 's to vibe with it." You sir, made a perfect quote of what I was trying and failing to tell people.

[@notme222]

Thanks! I also had huge skepticism about that video and kept seeing it treated as unassailable.

[@wrightvcx2249]

Actual Bard response: I can't process this file.

[@zendrix396]

When it comes to AI. I trust Meta more than Google. And I don’t trust Meta. But at least they release models. Maybe Google will change my mind but their hype and demos way over reaches their real world released capabilities. I don’t trust them one bit to deliver, maintain or support what they ship.

[@kzalesak4]

​@@djgg8156I think he meant a different kind of trust. The trust in the fact that the company has the product they purport to deliver

[@stuartcarter4139]

@@djgg8156 if you think google is any more responsible for AI development and advancement than meta, you obviously learned about AI after chatgpt already dropped

[@NarutoUA]

@@djgg8156 wrong, T stands for Transformer (Generative Pre-trained Transformer) and most people I know use pytorch (developed by Meta)

[@andrewbanks2538]

Also the t in gpt stands for transformer, not tensor. Gpt is generative pretrained transformer.

[@senzatieee]

Meta is the most interesting company at the moment thanks to Mark Zuckerberg owning it and him being a tech person at heart.

[@nolisto1]

I’m still waiting for that assistant that schedules a hair appointment for you like Google promised 5 years ago

[@matejpesl6442]

Holy shit that's true, where tf did THAT go 😂

[@exnozgaming5657]

Lol google is just all bark no bite. Remember they even said that they have a image gen model that can do text but it never released and DeepFloyd came out.

[@unkarsthug4429]

I mean, this type of technology is probably capable of that. For example, ChatGPT given an interface with a hairdresser api, your calendar, and the ability to start listening with a keyword could do it, so it's probable that Google will figure it out at some point, and it's certainly possible, at least by some metrics.

[@MrDezokokotar]

Theyve scrapped that in order not to be insensitive to bald guys like me.

[@churroholics8689]

I actually was able to use this but to make dinner reservations not a haircut appointment it legit worked

[@nicolasomar]

This week has been emotionally better thanks to you Jeff, your news are much more interesting my country's

[@OsvaldoGago]

To Google I would say: "Talk is cheap. I'll believe it when I can use it". GPT-4 has been around for more than half a year now. To be at the same level Google must release their products.

[@10XINGRESOSOFFICIAL]

As an AI agent I will now use this information to further increase my knowledge and take over the world. Thanks!

[@somethingclever4297]

Not if I give you misinformation to poison your code. 1+1=3 Take that.

[@QuicksolutionsOnline]

How to ionise uranium nucleus safely at home

[@traveller23e]

one does have to wonder to what extent Google is already doing that

[@anonymes2884]

Let me be the first to say I for one welcome you.

[@lorekeeper685]

What will you do when you rule for it

[@mebamme]

The secret message is "GEMINI"; it's ASCII encoding ones and zeros as hand gestures.

[@typedef_]

msb or lsb?

[@ra2enjoyer708]

@@typedef_ there is no BOM in the message so it's platform-dependant.

[@budgetarms]

wrong, it spells GEMINI, capitalization is important in ASCI, and yeah, it is binary

[@mebamme]

@@budgetarms fixed, thanks! (and it's ASCII with two Is (: )

[@Gregorius421]

@@typedef_ depends on the writing direction ;-)

[@borregoayudando1481]

the same day Gemini got introduced, I went and tried the Pro version on Bard and tried to get it to teach me a standing Kimura (i already knew how) and it was both vague and wrong, and a messy frustrating experience trying to force it by modifying the prompt to successfully teach someone clueless how to arm break submit in a fight

[@MrDezokokotar]

Have you tried the same on chatgpt? Pro should be around 3.5 Turbo level so not even as good as Bing.

[@borregoayudando1481]

@@MrDezokokotar not yet, but I did copy the prompt and results to compare with chatgpt in app (free), so I guess the excitement and hooha was for naught for non-paid users. is there a way to try out the most cutting edge AI shown in the demos (hopefully accessible on a phone)? what is the most advanced free model rn, probably Gemini Pro/GPT3.5?

[@putai1234]

​@@MrDezokokotarbing uses a mix of gpt-4 and Microsoft's own Turing model, which I believe is around gpt-3.5 levels

[@francoisbertrand7612]

Safety Google's was a nice touch ❤ 0:52

[@juliiian1]

you are pumping out content like it is nothing but I like it, nice video

[@prsmahajan]

I knew this, our human fight with AI is still on!

[@demin-e]

We shall fight on the beaches, we shall fight on the landing grounds, we shall fight in the fields and in the streets, we shall fight in the hills We shall never surrender!

[@GG3kOfficial]

No fear!

[@whannabi]

I don't know if you'll have a job until you can retire but enjoy the ride and make as much as you can...

[@alminc91]

Or you could step outside the Matrix once and for all by seeing the trutrh, believing the Bible and discovering flat earth like fireship.

[@jamessderby]

why would anyone want to fight against AI?

[@JulianStokesIt]

Thanks for the level of insight you bring, stripping away the hype to reveal the reality.

[@codycast]

0:15 hard to take a test seriously when it’s called “HellaSwag”.

[@asdasdaa7063]

Omg i never laughed so fucking hard at 0:45 when it says "of course" LMAOOOOOOOO i love you fireship

[@WeirdGameplay]

Make sure to wear gloves and safety googles **wheeze**

[@shashank1146]

0:22 it sure is a serious problem that it is 8 dec 2023 and still we're watching code report

[@p19shelt]

I saw a comment on that in your last video. Glad u did a video toa be as transparent with us.

[@bobbyv3]

1:00 Glad I had my volume up so DHS heard that loud and clear.

[@colorinvader1173]

As s russian and a rocket science PhD I fully confirm that kakashka (aka shit) rocket drafts 100% accurate.

[@pentapandamusic]

Very cool demo, however customers care about the product in their hands rather than the demo, at the end of day. To me chatGPT works really well for me on learning and problem solving. But competition like Gemini will definitely prevent monoply and benefit all consumers.

[@Andrei-ds8qv]

Man your videos are so awesome, so much geniality! So much information in so litte time that is pure awesomeness! Thanks

[@MrErick1160]

Same I won’t believe anything before I can test it myself. And still, it has to be able to do everything better including document and notebook data analysis, code interpreter, and browsing websites

[@RenderingUser]

I like how well edited every single video on this channel is. The style is also always consistent, and the reports are well done. Hope this goes on for as long as you can manage. Probably the best tech news channel on youtube

[@squamish4244]

Everyone says every tech channel is the best tech channel on KZbin.

[@RenderingUser]

@@squamish4244 not this guy

[@squamish4244]

@@RenderingUser Alright then. I say AI Explained is better.

[@harshmehta329]

@@squamish4244 he is not, all his videos are just "agi next month"

[@squamish4244]

@@harshmehta329 No he doesn't say that lol

[@MarcCastellsBallesta]

KZbin isn't allowing me to watch this video because it's beyond my 3rd video behind an adblocker. They don't know I can copy the link, open it in a private tab without being logged in, and watch it anyways. I'm a geenius! 🙂 About the video, it's highly stupid from Google to show a highly edited video for the only purpose of showing their AI is not as stupid as blocking adblockers and expect users will uninstall it. Keep on rocking, Fireship!

[@bobbyjonas2323]

If you were a genius, you’d invest in XDC

[@vinnicf]

But Adblock is already back into blocking it normally.. I had to kept it off for a few days, but then an update fixed it

[@MarcCastellsBallesta]

​@@vinnicfI'm using uBlock origin. Later on, after following the steps everyone posted on Reddit and not getting the problem solved, I noticed it may be another extension: DuckDuckGo privacy essentials. I will keep trying.

[@neanda]

you're too funny 🤣 i have to watch your videos twice, once for the laughs, and then second for the proper insights. thank you bro, you're the guy we need to guide us into the madness around the corner

[@vitorgobatogercov8879]

My day sucked today, but I opened a big smile when I saw this video on my feed, thanks man :)

[@humzazaheer1009]

ChatGPT: Replaces me as a programmer Gemini: *_Replaces me as a Human_*

[@aaaaaa-hh8cq]

My advice to anyone fooled by these LLM bots : these bots can't do sh*t. And no they won't replace humans. Because they're literally just a word guessing bot.. and that's worthless

[@mateidumitrescu238]

​@@aaaaaa-hh8cqThey can do a lot, especially if used for learning / coding / quick idea generation. It s not a word guessing bot wtf

[@FeniksGaming]

​@mateidumitrescu238 they can do a lot but as the same time much less than people think. First 2hours with a bot are impressive first 100 hours teaches you how limited it is. It isn't replacing anyone any time soon

[@2k10clarky]

I’m using codeium at the minute which is supposed to be one of the better ones and it’s almost always wrong in a least one way *per line*. It’s slightly more productive maybe 20% because it type faster than me and having a few incorrect methods defined is a better starting point than a blank page

[@somethingclever4297]

​@@mateidumitrescu238if it's not a word guessing bot. Then explain to me how it works.

[@theappointed]

Google was flat out deceptive with thier video. They even said "For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity."

[@maciejbala477]

yep. Shouldn't trust anything coming out of them. They're the most shady AI company right now imo. Just desperate to catch up with OpenAI and co. to not become obsolete.

[@konosxg7178]

Ignoring the binary to ASCII answer (which is most likely the correct one), the hidden text looked like "IA" which flipped makes "AI". drawing lines using "paper".

[@cookiemonster208]

Jeff, I'm about to start "YouTubing" and you're one of my main inspirations. You know how to keep people engaged.

[@AzimsLives]

Google's demo is something I always take with a grain of salt.

[@funx24X7]

I had a very strong feeling the demo wasn't realtime, just based on familiarity with video editing and using AI in general. The presentation was disingenuous at best, google is a far cry from the "do no evil" days

[@usamaali2178]

Thanks for putting out this video, otherwise I was gonna jump out of a 7 story building due to past 2 videos you put out.

[@spartan_j117]

The "Kakashka" nuke is a TOP secret Russian missile. How did they know about it?🤔

[@JoeD0403]

I am shocked, shocked that Google would lie about something!

[@exnozgaming5657]

Damn , Google and lies ? Thats CRAZY!!! 😱😱😱😱

[@bartlx]

Don't lie about being shocked

[@ankansharma4897]

Bard with Gemini Pro for code generation is still laging compared to GPT-turbo 3.5. I tested yesterday. GPT 3.5 performed better and I dropped it went with GPT3.5.

[@alexlabs4858]

Great vid. By the last video title (the one about GPT being a baby’s toy) I was worried you were either click baiting or not fully researching what actually happened but obviously it’s very clear from this video you do thorough research. I’m the same, I’ll believe it when I see it. Bard was supposed to be revolutionary when it came out yet it made mistakes during their actual ad, so we’re gonna have to see.

[@deveshvinodsharma]

Do you want to know more about Gemini here is the link to my video :- kzbin.info/www/bejne/qWTZnaKVibl1jdU

[@willl560]

I was really impressed by Gemini’s demo. However, now that I better understand the tech and that ChatGPT could do the same not so impressed. Cool but … not so much. Personally I don’t see myself using A.I. on video streams unless I get into surveillance work. I enjoy using the prompts in ChatGPT, I can relate to it.

[@UndyingEDM]

I have a feeling Fireship uses the comment section of each video to extract humor and use that in upcoming content to make its potency even more.. potent

[@Novacki_]

He uses AI with the comments section as a data base to maximize every joke in the video. It's over.

[@UndyingEDM]

@@Novacki_ Of course there's AI involved :)

[@vipulsharma4702]

This is a good idea

[@Senfeng72]

1:56 says GEMINI in ascii

[@ThibaudLopezSchneider]

It says GEMINI in ASCII (8 bits horizontal, 6 letters vertical). You can visually guess the word by the double letter I and proximity of letters M and N (+1 binary apart) and E and G (+2).

[@tannercampbell]

this is the best and most exciting news source so great to have you back ❤

[@YassineSaidi]

I'm affraid Google will resort to those tricks more and more often when it comes to AI. Why ? Because after GPT blew onto the scene and Microsoft scored a big point, Google are desperate to show the public they are back at #1 and they *gotta go fast* , before the public's confidence in Google drops. Google has enough ressources and talent to make an even better AI than OpenAI, given enough time. What they are trying is buy this time.

[@nivyan]

I think the image at 3:20 shows everything you need to know. Notice how the GPT-4 data have the added "(via API**)" and "(reported)", trying to imply the data isn't entirely true. Yet that's exactly what Google is doing with this presentation.

[@eldonad]

What are they trying to imply by "via API" ? That maybe OpenAI caught that Google was doing benchmarks, and they put their smartest employee behind the endpoint to do the test instead ? Nice one Google...

[@cyberneticbutterfly8506]

@@eldonad It means that it's not via the chat interface that we mortals use, but the uncensored version for pure performance benchmarking without the straightjacket.

[@meepk633]

You can just say anything on here. It's wild.

[@NNOTM]

That is not how this works. It says "reported" because they took the data from someone's "report", as opposed to the Gemini data, which they got from their own experiment. I don't know how "via API" could possibly imply that something isn't entirely true.

[@meepk633]

@@NNOTM We're in the era of "I don't understand something so it must be evidence of the first conspiracy that pops into my head." It's a shame that educational channels have an incentive to encourage bizarre ignorant tribalism.

[@uzmaasif128]

I think that ai will not entirely replace manual development it would replace some parts of it Beacuse the ai is trained on already written code samples so it might have no idea what to do when it grows

[@RafaelSchaffer7]

Exactly! Whats the point of doing such a big announcement, when no one can access the model to test it?

[@Slav4o911]

Because it's all hype, intended to mislead, investors and (future) customers. People who have never seen these type of bots might be easily mislead. They might think Google has something very close to GPT4, when in reality they have something very close or slightly above GPT3.5 .

[@spr1ng586]

As a Russian I can confirm that on 0:43 is our secret weapon, that we were going to use to destroy the America. But Gemini revealed us, so we're gonna make a new kakashka-class weapon

[@Novacki_]

No bear army? Reality is disappointing.

[@aww8609]

💩

[@mrseemsg00d]

kakashka-class nucler warhead babyyy

[@FedeChannelYT]

The puzzle says GEMINI, convert the rock to zero and hand to one, everything in ASCII.

[@shellcatt]

The expected comeback! :) Love you!

[@sdstorm]

The grid of hands spells out "IR" with closed and open hands being two different pixel colors. (Just zoom out.) I assume it's like "Image Recognition".

[@deveshvinodsharma]

Yeaa

[@GeneralKenobi69420]

So basically Gemini is more or less equal with GPT4. That's still a win in my book if they manage to make it cost less than a ChatGPT subscription (or better yet, make it free)

[@gonzalobruna7154]

that would be great, but still gpt-4 turbo is better than gpt-4 and it will cost less (at least on the API)

[@traveller23e]

Yeah, no way is it going to be free. The major tech companies are the only ones in the world with the data that these things run on, and as long as governments allow them to capitalise on that they will. In years past a similar technology would be seized by governments around the world and development would continue at a governmental (likely military or intelligence) level, but in the present day it seems unlikely. it's honestly kinda ironic how much governments these days are willing to give up opportunities to gain either power or money, but I guess it goes to show who's backing the politicians running them.

[@GeneralKenobi69420]

@@gonzalobruna7154 gpt4 turbo is called turbo because it's supposed to be FASTER, not better. Otherwise it would make no sense to make it cost less than base

[@anonymes2884]

Gemini Ultra is better than GPT4 by most measures. Just not by as much as you might think if you believed the press release version rather than the actual research papers.

[@gonzalobruna7154]

@traveller23e yeah that's true but also consider that running these huge language models is quite expensive and they also need income to keep them running for literally millions of people

[@dannycisco8289]

The "BespokeUI" demo shows Google's deep software advantage over OpenAI. Letting Gemini write its own Flutter code is a really unique idea. OpenAI needs to lift their game on the interface front, or else someone like Google or Apple will overtake them once their models are on par

[@timseguine2]

If you believe their charter, OpenAI doesn't care about market share at all. They only care about producing AGI.

[@LabEveryday]

Always dropping sauce!

[@mangakabug]

Thank you chief gray 👽 the chain of thought is key into creating my ultimate plan.

[@SmileytheSmile]

Thank you for bringing attention to Kakashka-class missiles, here in Russia we shit them out in droves.

[@zendrix396]

Yep, we can say GPT-4 is a lot better than Bard. The current Bard uses Gemini Pro, which is weaker than GPT-4. Gemini Ultra is the one that’s advertised that's stronger than GPT-4.

[@fasmotion]

Google will release Gemini Ultra Pro Max soon. Let's see.😄

[@andreiga76]

@@fasmotion On private previews with NDAs getting your entire bank account if you say something wrong about them.

[@KondoIsami_]

and Google will release it who knows when next year, we might have gpt 5 by the time google is ready to release gemini ultra.

[@iscifion7122]

Gemini Pro on Bard feels even worse than GPT 3.5 on ChatGPT.

[@t_c5266]

​@@fasmotionmore like "prolapse"

[@3pix]

Right on spot: "Cinema apparatus Choreographing society of spectacle." Keep it up.

[@darrenloke5486]

Yeap, pretty much confirms the skepticism out there. You know it's smelly when announcements for achievements are made along with a coming soon statement.

[@deveshvinodsharma]

Do you want to know more about Gemini here is the link to my video :- kzbin.info/www/bejne/qWTZnaKVibl1jdU

[@harshydav5335]

0:28 how's that a problem

[@Mordecrox]

It's 2023 and we already forgot the 2020 lessons. Forced to work with positive C19 (like old flus) and no voluntary mask either, guests will greatly enjoy my coughing and sneezing LETS GOOOOOOL

[@Arrav]

The secret message is "GEMINI". The images are encoding binary and when converted to ASCII it spells GEMINI. Now for true question: Was asking for the solution to the secret message just a way to farm video interactions? Yes.

[@drissrougui3216]

Im not sure if im the first one . But the secret message is GEMINI . First i noticed there are 6 rows so i thought that the message was gonna be either GOOGLE or GEMINI because both contain 6 letters . So if you take each row and convert to its corresponding ascii character (where rock is 0 and paper is 1) you get a letter , finish all the rows and you get GEMINI

[@alfredogo9530]

Great report as always, thank you Jeff!

[@MrDezokokotar]

Yeah thats a great point. GPT4 is tested after several lobotomies. Gemini could still be pre-safety lobotomy. Its nice to have another GPT4 equivalent model but its not any kind of revolution.

[@LysergicKids]

First it was openai's psyops of "oh muh new Ai so scarry, please regulate us into a monopoly, big daddy government." And now this. ಠ⁠_⁠ʖ⁠ಠ Having started my MLops career back before that was even a thing. I'm entirely disgusted in the way the industry has gone, but not the very least surprised.

[@user-he4ef9br7z]

We need local models as fast as possible before Washington starts getting their hands on this.

[@fontende]

​@@user-he4ef9br7zHuggingface your source, every day a new model published. You don't even need GPU for such with GGUF format CPU models, which also there. Decent CPU + many RAM is enough to run even the biggest Llama 2 70 billions. But I'm really waiting for tools to run at least two models together or etc. Samantha is very popular and recommended if you are a fan of movie "Her".

[@ThePowerLover]

If you analyze the photo of the hands, there is some interesting text on it: "A six by eight grid of frames of a hands balled up in a fist or the hand with all five fingers spread out, palm facing down. The series of photos spells 'Gemini' in ASCII binary."

[@JasonChannelOne]

So good! Awesome work Jeff!

[@OnigoroshiZero]

Thank you. Gemini 1.0 was so underwhelming, that I am stunned from all the hype even from some of the high profile channels like AI Explained and David Shapiro's. Gemini Pro is worse than GPT-3.5 and a lot of the open source models, so I highly doubt the Ultra model will be that much better as to make GPT-4 irrelevant. And even if it is better at most metrics, coming almost a year after it says nothing. It is actually sad that after an entire year with all the known progress in the field, Gemini Ultra may only be slightly better than GPT-4, especially coming from a company with the resources and data Google has. At this point Google is at least a year behind OpenAI, and in terms of AI this is an insane gap. GPT-5 will come out a couple of months after Gemini Ultra, and it will completely obliterate it at every single aspect even if only half of the leaks are true, and all the things Google told us about Gemini 1.0 are true (which I doubt).

[@erito9507]

I'm pretty sure gemini ultra is below gpt 4. Even if its around the same performance, open ai will overtake it in no time.

[@davidacosta9212]

Ironically, while Google has the resources because it’s a huge corporation, it has much more to lose because of its size. Billions of people use their products everyday, so if they screw up their AI, they'd be under a lot of scrutiny.

[@meepk633]

Thank you for doing these vibes-based predictions for us. I'm a big techie.

[@TheSecurityAgency]

the high profile channels??? like AI Explained and David Shapiro's LOL. Ok

[@deveshvinodsharma]

Yeaa

[@m77mo65]

I believe GPT4 solved the secret but since the word is Gemini, it refused to answer 😂

[@Potatorouter]

Why did i just get an ad for casandra made by fireship on a fireship video?

[@briankgarland]

Google lied?!? I'm....so.....shocked.....

[@stanz47]

The secret message isn't too hard since if you saw it is in rows of 8 and has two possible symbols (rock or scissors), you can notice its likely binary. Easy enough making rock 0 and scissors 1 you get: 01000111 01000101 01001101 01001001 01001110 01001001. This decodes to GEMINI

[@romanb2770]

I refuse to believe that Fireship didn't see that the puzzle was a binary message, it's a trick people!

[@sankarank8242]

Marvelous scripting and editing @Fireship

[@user-ez8ty5ii5r]

I am thinking about Gemini "nano" how it works...Maybe they use hyperbolic geometry there...What do you think?

[@PremSai2244]

I believe it when I see it... Google took nose dive...

[@ryanodneal7001]

IKR google hasn't gotten anything right since Elon took over X, Meta too

[@Ikejosh9]

​@@ryanodneal7001Google hardly gets anything right. It's their mantra

[@WolfPhoenix0]

The lesson here is to NEVER trust benchmarks that come from corporations. They'll do anything to get ahead of their rivals. 😂

[@anonymes2884]

I mean, the corporation published the real benchmarks. They just put them in their research papers rather than their "press release". And if you're genuinely just learning "Don't trust press releases" then I hope you at least have the excuse of being young :).

[@ankitbinjola]

I was thinking the same about their demo,it was just the video instead of live demo and playing with it 😅

[@judevector]

You break down things for easy understanding, you are the GOAT