- Tubelator AI
- >
- Videos
- >
- Science & Technology
- >
- Google's Revolutionary Gemini 1.5: The Future of AI Unveiled with a Twist
Google's Revolutionary Gemini 1.5: The Future of AI Unveiled with a Twist
Explore Google's groundbreaking Gemini 1.5, a superior large language model that surpasses GPT-4. Dive into the latest technology releases, game-changing features, and addressing of rumors in this insightful recap of Google's event. Find out how Google is pushing the boundaries of AI on February 23rd, 2024, in this engaging Code Report video.
Install Tubelator On Chrome
Video Summary & Chapters
1. Google's Impressive Technology 🌟
Google releases groundbreaking technology and faces challenges.
2. Gemini 1.5 Announcement 🚀
Introduction of Gemini 1.5 with advanced features and capabilities.
3. Open Source Models by Google 💻
Google unveils open source models to compete with industry leaders.
4. Guardrails Controversy 🚧
Issues arise with Gemini's image generator leading to a controversial situation.
5. Web Developer's Monumental Achievement 🖥️
Significant progress in web development with a new modern look by Google.
6. Gmail Shutdown Prank 😱
Google's prank about Gmail shutdown causes a stir among users.
Video Transcript
This has been the craziest week ever. That is, if your name happens to be Google,
it released some of the most impressive new technology in history,
Wow!
Had to apologize for its other not-so-good tech,
and had to address rumors that were so insane people actually believe them.
Whether you love Google or hate Google, there's something for everybody in this video.
It is February 23rd, 2024, and you're watching the Code Report.
Event 1, Gemini 1.5.
Things got off to a good start. Google hit a massive high with the announcement of Gemini 1.5.
I was able to use my deep-state connections to get early access, and what I can tell you
is that this thing has got some serious risk.
It's a large language model that's superior to GPT-4 on most benchmarks,
but with a staggering 10 million token context window, this blows other models like
Claude and GPT Turbo out of the water. Currently, people have been using retrieval
augmented generation or RAG stack to help LLMs better understand custom data.
This has led to tons of vector database startups, but many people have been
underwhelmed by the efficacy of RAC, and these models can generally gain a better understanding
of custom data from a large context window. And it's just a far more simplified system,
like I uploaded an entire codebase from my local machine for a side project I've been working on,
and then asked Gemini to start building some features on top of it. It performed way better
than Copilot or any other tool I've used, and knew how to incorporate different components and
libraries that exist in the project. Although it took like a full minute to complete the prompt,
but another killer feature is the ability to upload long videos.
I was able to upload videos from my Fireship Pro courses, and it could automatically extract
code and write tutorials about these videos.
Overall, it makes GPT-4 look like an antique from 2023, but that brings us to event number
2.
Gemma, Google announced a family of open source models that are designed to rival Metas
Lama 7B and Mistral.
Based on Google's own benchmarks which will take with a grain of salt, these models dominate
the competition, especially when it comes to math and coding.
These models are free and can be used to make money in your own apps, but there are some limitations.
You have to follow the prohibited use policy, which means you can't use them to do any fun stuff.
Everybody loves guardrails, but in event number three, guardrails went horribly wrong.
People started noticing some weird behavior with Gemini's image generator.
If you prompt it for an image of ginger people, you get a result that is both hilarious and horrifying.
But it appears Gemini was designed to be so anti-racist that it paradoxically became racist.
Other image generators had been criticized for a lack of melanin,
and Google tried to address that with a new policy.
KILL WITING!
No, no!
But they ended up making everybody mad.
The left wing was outrage-seeing multi-racial Nazis, while the right wing was outrage-seeing
multi-racial founding fathers.
This culminated with an apology from Google, and they temporarily suspended Gemini's ability
to generate images of people.
It's going to be the technical challenge of the century to make everybody happy, but
this next event was a monumental achievement for web developers.
For weeks now, Google has been showing this banner, talking about how it's improving
at sign-in page with a modern look and feel.
