Problem solving across 100,633 lines of code | Gemini 1.5 Pro Demo
Vložit
- čas přidán 26. 07. 2024
- This is a demo of long context understanding, an experimental feature in our newest model, Gemini 1.5 Pro using 100,633 lines of code and a series of multimodal prompts.
This demo is a recorded walkthrough of single continuous interaction with Gemini 1.5 Pro.
Token count details: The input TXT file (816,511 tokens) and image (256 tokens) total 816,767 tokens. The text inputs add additional tokens into the prompt, yielding the 818,495 token total shown in the interface.
To learn more about Gemini 1.5, visit goo.gle/3weBZhn
Subscribe to our Channel: / google
Tweet with us on X: / google
Follow us on Instagram: / google
Join us on Facebook: / google - Věda a technologie
Love it when developers are building their own replacement.
These guys wouldn't do it if they were not making more money developing it.
Not only are they building their own replacement, they're building everyone's replacement. Because if they can replace developers then they easily replace everyone else. Pretty good.
nothing is gonna replace developers, who is gonna write the prompt ? my clients? no way!
@@vectoralphaSec Replacing everyone's jobs would be fantastic if we all benefited from it. But in a capitalist system only the owners of the means of production will benefit and the rest of us will be screwed.
@statuschannel8572 yes your clients. Developers will be replaced one day to believe otherwise is being in denial.
Its going to be an interesting year.
The next 5 years will feel like a science fiction movie.
Facts!!! 😂
@@theeternalnow6506 yeah, a very dystopian one
@@theeternalnow6506 unfortunately we don't know if it's the dystopian kind or not...
I said that by the 2030s neural nets will be able to make 3D games like Starcraft 2 by itself. I think I might be spot on.
Excellent example of real world usage. I will definitely try this.
It’s hypothetical 😂
@@Ewakaa No, it's available to some people, You can also request for Early Access I think?
@@ElclarkKuhu I have access to Gemini after requesting early access it was not hard at all
Finally Google is a real player on the AI future, I was sceptical when Bard was released, then somehow satisfied with Gemini 1.0, but Gemini 1.5 is very exciting, well done.
They were more worried about ai safety, but openai pushed them to corner 😂
Let's rephrase it: Finally workers at another company made a LLM with extraordinary use value. I was sceptical when this company's employees made a little out-of-date LLM, then somehow statisfied when workers at this company made a better product than other for-profit company, but current product looks very exciting, well done.
That's Demis Hassabis for ya
Too woke to do anything useful with though.
@@loveboatnahh it does most tasks with no problems
Man what a time to be alive!
I'm not so impressed with the demo but with the honest and transparent presentation.
Good job on the presenter and the higherups that approved this raw demo instead of doing an elaborate lie
Someone hates Google. lol
@@helix8847 To be fair a third of their income comes from selling users data to advertisement agencies.
This is still kind of a lie because its running on an open source codebase that's meticulously maintained to be easy to understand. I would be really interested in knowing how it performs on large codebases in general. Especially in large custom codebases where the code is not structured as well.
What a time to be alive
I need this as a vs code plugin to learn my codebase
At the moment google is working on their own vscode replacement - project IDX
It will definitely increase their sale if they make vscode extension
Why wouldn't you use standard embeddings for that? You don't need a large context length.
@@gavinderulo12 Maybe he's the manager of a company's technology employees. This company may have several legacy codes, and migrating something, correcting an error or maintaining the code is quite costly in terms of time. In this case, it would be useful for him to have 1 million tokens as he would just need to place the file containing all the source code of the software that his team needs to work on
@@battatouile8135 what? Just embed the document. Then the model can access and search the entire code even if its way longer than the context window.
If they can manage to actually pull this off this is going to be wild!
I am already impressed that Gemini can understand and write Tamil language much better than ChatGPT. I am hopeful that Google continues to support non-English languages like Tamil so everyone can benefit from the LLMs - not just English language speakers.
Try gpt 4
punjabi and hindi are meh so far. Need equity of accuracy and excellence across all languages else languages will die off soon as people become reliant on English for LLMs
😂❤
@@blkscreen15 English will always be the best language for LLMs. Its the same thing with things like Google translate. Translation to English is always best.
This is crazy stuff, waiting for open AI response, what a time we are living ladies and gentleman 😀
its interesting and really cool now, but we won't be so happy after it has ruined / deflated the job market and destabilized the economy i a few years time from now. Mark my words, generative AI cause more problems than it solves
To quote 2 minute papers:" what a time to be alive!!!"
We love competition. Work Google, work. Work OpenAI, work, who wants my money? 💰
OpenAI definitely responded and overshadowed this ngl
Hahah I know they did fast, I have no words this is insane 🤯🤯 crazy times a head! I’m just worry for the average person, critical thinking skills will be the skill of the century
So encouraging. Great work!
Insert *Willem Dafoe looking up meme*
this is actually awesome. I'm truly excited to try out gemini 1.5, and looking forward to the Ultra version coming soon! that's going to be wild.
Seriously impressive
😅 1:19
nice that you showed the chat and did not make a video like when you first introduced gemini ("The capabilities of multimodal AI | Gemini Demo")! :)
I think they will make that demo a reality in their next phone, 100%.
@@PseudoProphetwishful thinking is not a good enough reason for deceptive marketing
@@vidal9747 you think that is wishful thinking?
Imagine what their confidential models are capable of. These are just versions open to the public.
@@PseudoProphet don't know. That is what they want to happen. If they couldn't make it happen with their best models, it is wishful thinking and it's not sure to happen soon.
ooh damn, this can help me in learning new things
looks really interesting. 2024 is going to be wild :D
I’m impressed and scared at the same time. Wild.
Amazing!
Impressive!
Ooohhh, that's a lot of tokens!
What is a "token"?
@@IntensePeppersThe college of computer science.
This is an example sentence. Each word is a token in that sentence. Im not sure if a period or a comma is a token though
when is this coming to the gemini website?
Google is always the king 👍
The king of violating user privacy, squashing consumer rights, and shoveling money into failed products.
Professional thinkers making professional minds = next-gen AI 🐴
I like it, keep it up
The Gemini Google assistant should be an asset ruled by mercury the planet of communication of all signs 😊
WOW! 🤩
We're in the future! 🚀
Would be nice to hook this to a large codebase, I bet it would speed development a lot!
عمل جبار حقيقة، تذكرت شخصا كان يقول أيام ما كان chat gpt جديدا، ما تقوم به هذه الشركة هو مجرد ورقة بحثية واحدة مقارنة مما توصلت به جوجل في هذا المجال، وأن ما ستقدمه جوجل سيكون شيء لم يسبق له مثيل.
شكرا لجوجل.
هل لديهم عملة رقمية خاصة بهذا المشروع
"I am unable to engage in discussions that could be construed as biased or offensive. I am an AI chatbot designed to provide information and assistance in a neutral and unbiased manner."
This is very easy when the codebase is as clean as threejs's. If it tried to understand one of my code bases it would decide to self destruct.
This can be a game changer for coding, if made 20x faster
It doesn't need to be 20x faster to be a game changer. The big sell here is the context window and accuracy. 60 seconds is nothing. Prompting with a smaller context window should still be quick.
If you can ask where something is in your companies massive codebase that would be one of the biggest things yet.
I’ve wanted this for so long.
🎉?@@driftedsun
@@driftedsunか
oooo
Amazing,
Would it be possible for Gemini to read code bases, like a python project and the modules in the projects directory in my drive? its currently only limited to reading documents or pdfs.
The beast of Mountain View has awoken
Awoken, wow!
wait, that IS a word, wow! my bad
Great! I look forward to the release date.
proof that this is not fake again, like your last promotion!
These are all examples of tasks a human programmer would find easy to perform and wouldn't turn to an LLM for help. What about the hard tasks, like refactoring the codebase for cleaner and less redundant code? Any examples for this? Even failed examples would be interesting to share!
I'd say it could definitely pull it off
I mean, that's pretty much the point. If LLMs get accurate enough that we can leverage menial tasks like these to them, then we can center our focus to the more complex problems such as architectural and efficiency/performance issues within the codebase. It's all about being able to improve DX
Yeah if you knew wtf your code did, have you ever gotten a project that is a total mess and spent hours going over it... its a nightmare. This will help greatly with that.
@@helix8847 I actually used 'grep' previously to search the exact same codebase for code snippets. Not that hard really ...
Crazy to see what competition can make you do. From so called "AI Assistants" to this within like a year in a blink
a million tokens is crazy, AI is getting better at an insane pace
Singapore can use it now 😊😊😊
Ok I tried Gemini 1.5 pro. it can't write a code to solve a euler bernoulli cantilever beam in python.
oh well
how long did it take you to get accepted into the waitlist? because there's a waitlist for Gemini 1.5 pro right now and you have to sign up
@@ViprazDesigns 2-3 months
For the time being...give them one year only
Those who might be confused with the naming.
Gemini Pro 1.0 (free)
Gemini Ultra 1.0 (paid $20/month)
Gemini Pro 1.5 (exclusive)
Gemini Ultra 1.5 (didn't come out yet)
Huge if true!
It's not true... they also did similar "demo" before, then in reality it's not what they've showed in the demo.
@@Slav4o911 cause they Nerf it due to some stupid safety standards
good choice with threejs. the most obscure library.
why not on chromeOS?
I don't want to call this out but I personally feel like this demo is a bit flawed, here's why:
The codebase is actually set of examples based on three.js modelling --> in simple words the codebase is actually a collection of **independent** code files.
A more accurate demo would be possible if they did it on a codebase with integrated units and functions.
I know this is actually showcasing power of large context handling but then if the sub-context are independent of each other then it can be achieved by iteratively going through the small subcontexts. So in general the AI doesn't see the whole intermingling since there is none.
Correct, they probably did it with RAG, the actual context is not 1 000 000.... and even that will probably be only available for corporations, so we will never see it. They've basically used the AI as a "search engine". The AI didn't actually comprehend the whole context. I'm very skeptical to what they've shown.
damnnn, Google is coming for OpenAI
Coming for Microsoft
Not OpenAI
@@Guardian_s_ same thing
The most precious thing is that Google is back in the fight! 🤗
Testing their new project on a Mac of Apple, their biggest rival? That’s wild
OK. Let's see it get some Github code requests completed. That's the ultimate benchmark.
When?
when i asked gemini for coding, it always showed the example code with outdated version. Even it could search everything on internet. That really confuse me 😢
for what i use AI for, context length is critical. hope to see open ai push the limits of context like this as well but for now gemini has won me over!
Good to see Google can counterpunch. Exciting times
Can they, their current Gemini is trash.
Hey, this is amazing!
But, um...can you guys like...apply this kind of zeal to fixing your current assistant or something? I'm really getting tired of feeling like it's somehow gotten worse over the past few years.
Like, this morning...I said "Hey google, I want to listen to 'I Am Hell, by Machinehead'. Instead, I don't even know what it started to play. I had to try FIVE different times in five different combinations of the artist and title before it just decided to work.
Or, heck. I have a custom routine set up so that where I say "Page ". Who is this Simon person? Why do I give a crap? How hard is it to make the assistant smart enough to recognize that I have this custom routine and keywords in there, and figure out that I am NOT interested in pictures of this Simon person?
So, yeah, if anybody at Google is listening...PLEASE. Make this technology the new assistant...NOW. And use this massive amount of context it can have to make it learn everything about me, the people I know, and the music I like to make even *slightly* more intelligent guesses at what I want to do.
Because seriously, it's insane when you guys are investing this much time in a paid chatbot while the one you already have schlepped onto countless devices is barely capable of figuring out the song I want to listen to...
And, an update to the above:
Today, I opened my assistant on my phone, and clearly said "I want to watch Baby Shark on the Family Room TV". I saw the text, it heard me.
What was the reply?
"OK, playing your playlist 'Baby Driver' on the Family Room TV'.
WHY? HOW? WHYYY?
how do we know that wasn’t part of the training data ?
the future looking amazing for learning
why would you need to, with these models ever increasing in capabilities, what's the point.
@@holthuizenoemoet591you know humans build those things right?
@@holthuizenoemoet591 You're clueless
0.01% of developers work on something like this. Vast majority won't be around much longer.
@@sleepnaught we don't see farmers using the plow by hand either, this is just the next step
What’s the output token limit though? If it’s like GPT 4: 100K input and only 4000K output then it’s still going to be pretty limited. GPT 4 can ingest a whole novel but can’t generate you a full essay within the output token limit.
But you can tell it to continue, right?
GPT4 cannot understand or extract details from the video, only understand at high level.
No more netflix! Reality is becoming my source of entertainment! Thanks Google!
THEY TOOK URRRRR JOOOOB!
Google be like: "Trust me bro!"
so that quantity of tokens used for analize a code that big, how much money are we talking about approximately?
Probably only for corporations, that's why I don't like these Google demos. If it's not going to be for everyone, don't show it. (I don't mean they should give it for free, I mean it should be with reasonable price) . Today their regular Gemini is a trash.
@@Slav4o911This is their CZcams channel, they can show what they want. Why are you gatekeeping when you’re broke?
yeah maybe it's cheaper to hire a developer lol @@Slav4o911
do i understand this right
openAi for chatGpt 4 price is $0.01 / 1K tokens
here we are talking about request with 1kk tokens so price for each request (approx if we take $0.01 / 1K tokens) = 10$ per request
google using apple machine 😂🤷🏽♀️🙌
We want it open source
The big question is, are all these claims actually true in real life (once the model is available), or it's yet another misleading (or should I say, fake) video, just like the one circulating before Gemini was launched?
Probably not true. In Google place if I wanted to impress people, I wouldn't tell anyone and just update the model and when people start to freak out how powerful is the model, then I would reveal the demo.... not the other way around. Remember what OpenAI did, they first put their ChatGPT, people started to freak out heavily, then OpenAI started to dumb down and censor their model (because it was just too good). Google does the opposite, they show us the uncensored model and then the release version is lobotomized and it's nothing like what they've shown. We already know these models, when uncensored are very powerful and when censored they become braindead. I mean when model is censored it's censored all around, it's like getting a genius and make a lobotomy to him. If people think some non existent "safety" is more important than AGI, we'll never have AGI. I mean somebody is still going to build AGI and take on everyone, but everything will be done in secret, which is much more dangerous... but people who don't understand the subject, have already made the decision.
OpenAI finally getting some real competition
Its fake...
Are they using macs? Not a chromebook or something?
Can you actually do anything on a chromebook other than browse the web with chrome?
@@xIronWarlordx probably, who knows 🤷♂️
Haven’t got the chance to test one yet.
Ilya Sutskever: "You destroyed everything I know"
Demis Hassabis: "I don't even know who you are"
I find it funny how software developers are the only ones who actually get EXCITED when an AI does their job better than them
How much energy does analysing such thing require?
6 energon cubes.
Can i use it if i'm in France?
French people won’t have access to Gemini Pro
C'est pas en open release dans tout les cas pour le moment je crois
Never
1 Mega token ? 1MT?
Step1 Give it all the stock market data.
Step 2 Ask it what stock to buy
Step 3 ???
Step 4 profit
Buy Nvidia... that's what the bot said. 🤯
wowww
this isn't showing much, how much of this data was already in the training set to begin with? how much of your question is it actually bothering to use the data given in the prompt rather than straight up memory? how about you put some hidden phrase in the context somewhere in the middle, and ask to "spot a hidden phrase in the given data". see if its actually attuning to the text, not just glancing at it
This will give us a generation of "programmers" who have taken script kiddie "work" flows to the next level.
I doubt there will be many software developers left in 5-10 years.
Jesus absolutely dominating openAI! Way to go Google! Hope these models are some day open source and we will be able to run them on our own hardware!
gooey
I tried basic python script with Gemini and it doesn't work at all
I use it daily, found it better compared to gpt, maybe little prompting tweaks might help
Google is still behind
❤❤❤❤❤❤❤
But did you fake it like the initial demo video?
It's real but this one in the demo is called Gemini ultra super duper... and not Gemini Pro... they just made a small mistake and showed us Gemini ultra super duper, instead of Pro....
How much do it cost in India?i need it
Yeah, it's over guys. Pack it up.
Isn't this higher than gpt 4?
Better quality for sure. GPT4 cannot understand the details of large prompts.
So much better contextual understanding🙏
Will be the end of people in Tech ? What if i throw my entire code base and ask to run tests and solve all the current issues ? Or if i ask to build an entire codebase end to end ?
Is it a real demo this time?
The Great Filter in all its glory
is it for real this time ?
عالیست ❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤خداداخاللاهلالهلاه
Lol this announcement AND OpenAI's new model SORA. This year 2024 is gonna be wild. Its still barely February!!
Tombos?
Please release Gemini in the UK
What about Gemini Ultra?
😊😊
I can't wait to see the source code and the dataset, like with Google's other AIs!
The thing is, can it solve "unwritten-before" problems?
No.
If this is true, Open AI has some homework to do
More like "ClosedAfAI"
@@yehaa00 😅
@@yehaa00 😂😂😂
Microsoft AI.
Very very very very very very very very good, helpful, useful, and beneficial.
Gemini ♊ is better now!!!! ❤
ChatGPT 5 is coming now