- Starter AI
- Posts
- OpenAI unveils Sora, Introducing Gemini 1.5, Google’s Magika AI tool
OpenAI unveils Sora, Introducing Gemini 1.5, Google’s Magika AI tool
We are steps away from AI-generated movies.
Sthefania
February 19, 2024
Hello, Starters!
A lot has happened since the last time we spoke. OpenAI has revealed a well-kept secret that has shocked everyone. We are just in the middle of February. Can you imagine what the rest of the year has in store?
Here’s what you’ll find today:
OpenAI reveals its Sora AI model
Google presents Gemini 1.5
Magika: An AI-based file type recognition tool
A “community forum” is coming to OpenAI
Nvidia introduces the Eos supercomputer
And more.
🎬OpenAI reveals its Sora AI model (22 min)
Taking the AI world by surprise, OpenAI has presented Sora, a text-to-video AI model capable of creating videos as long as 60 seconds at once, surpassing every generative AI video tool on the market. With a deep understanding of language, this model can translate words into incredibly realistic footage, worthy of a movie still.
However, we're yet to grasp the power of Sora, as OpenAI is making sure that the model undergoes several safety tests before releasing it to the public. Nonetheless, the results the company shared on X provide enough proof of its capabilities, which are set to transform AI as we know it.
🤖 Google presents Gemini 1.5 (3 min)
Not backing down in the race, Google continues to push its Gemini lineup forward by introducing Gemini 1.5, a significant advancement for its middle-tier offerings, making it almost comparable to the recently released Gemini Ultra.
This model leverages a technique called "Mixture of Experts," or MoE, which allows it to process only relevant parts of the overall system during queries, improving efficiency and speed. Besides this novel approach, what truly sets apart Google's latest innovation is its groundbreaking context windows with 1 million tokens. As explained by CEO Sundar Pichai, this is equivalent to about 10 or 11 hours of video or tens of thousands of lines of code.
Google's latest open source tool is Magika, an AI system that excels at file type recognition by seamlessly identifying them in milliseconds, diverging from traditional methods that rely on heuristics and user-defined rules.
The tool employs an optimised Keras model that weighs only 1MB, enabling it to accurately identify both binary and text-based file types. Magika is available on GitHub, and Google expects it to enhance user safety across its products.
🌐A way to foster education and development in a growing field like AI is to forge a community. That's OpenAI's aim with the recent announcement of the OpenAI forum—an online hub for domain experts and students who look forward to joining the company's quest for knowledge.
💻Responding to the increasing need for robust AI systems that can support numerous breakthroughs, Nvidia has finally introduced its long-awaited Eos supercomputer, built with 576 NVIDIA DGX H100 systems. This AI engine handles heavy workloads efficiently.
💼Jobs we liked
NVIDIA - Solutions Architect, Large Language Model Inference: United Kingdom.
OpenAI - AI Engineer: San Francisco, CA - US | $385,000
AMD - Machine Learning Engineer, AI Software Solutions: Santa Clara, CA - US | $130,000 - $253,000
OpenAI - Research Engineer, Human-AI Interaction: San Francisco, CA - US | $440,000
⚡️Quick links
What did you think of today's newsletter? |