- The Rundown AI
- Posts
- OpenAI changes the AI video world!
OpenAI changes the AI video world!
PLUS: Google's massive Gemini breakthrough
Rowan Cheung
February 16, 2024
Welcome, AI enthusiasts.
We just witnessed one of the wilder AI days in recent memory.
Google came out hot with another new Gemini upgrade — but OpenAI countered with a stealth drop that just elevated AI video to a whole new level. Let’s get into it.
In today’s AI rundown:
OpenAI stealth drops Sora video generation model
Google unveils Gemini 1.5 with massive context window
How to verify Gemini’s responses with Google Search
Is Michigan selling student data for AI training?
6 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
OPENAI
Image source: OpenAI
The Rundown: OpenAI just unexpectedly introduced Sora, an impressive text-to-video AI model able to create realistic and cohesive video generations up to a minute long directly from text prompts and images.
The details:
Sora combines aspects of GPT and DALL-E to better understand physical dynamics and maintain details across generated shots for realistic outputs.
The model can generate video up to 60 seconds from both text prompts and still images, and can also insert scenes within existing video.
Sora is currently being offered to red-teamers and select creators to gain early feedback before potential wider release.
While training data is still unknown, NVIDIA AI researcher Dr. Jim Fan speculates the model was developed using synthetic data from Unreal Engine, calling it a ‘simulation of many worlds, real or fantastical’.
Why it matters: OpenAI just took a massive video leap out of nowhere, bringing us closer to outputs nearly indistinguishable from reality. Sora looks to bring both creative unlocks and new panic-inducing capabilities for bad actors. For better or worse… the AI video game just completely changed.
Check out some of the impressive video outputs here.
TOGETHER WITH BRAVE
The Rundown: Building the next great AI? Take it to the next level with Brave Search API:
Access an expansive index of 20 billion pages in a single call, with millions added daily
Enhance AI chatbots, train LLMs, and inform real-time queries with high-quality, human-interaction data
Enjoy an intuitive data structure at a fraction of Bing's cost
Your AI is nothing without a connection to the web. Click here to get started with 2,000 free queries/month.
Image source: Google
The Rundown: A week after launching Gemini Ultra, Google just revealed an upgraded Gemini 1.5 model — featuring a groundbreaking 1M token context window (up to 10M in testing) to process vastly more data than previous AI systems.
The details:
Gemini 1.5 Pro can handle 700,000 words of text, 30,000 lines of code, 11 hours of audio, or 1 hour of video — enabling groundbreaking processing capabilities.
The upgrade maintains performance despite the large context window, with the ability to analyze granular details from huge blocks of text, code, and videos.
1.5 will be released to the public with a 128K context window, scaling up as the model is improved.
Currently, the full 1M token version is only available to approved developers and enterprise customers.
Beyond the expanded context, Gemini 1.5 brings improvements for more efficient performance on par with the top-tier Gemini Ultra model.
Why it matters: After just rolling out Gemini Advanced, Google shipped fast with another major upgrade. The massive context window unlocks new capabilities like analyzing entire books, full movies, large codebases, and more — giving Gemini a serious differentiator and a new leg up in the LLM wars.
AI TRAINING
The Rundown: Gemini's latest performance and accuracy improvements are impressive, but it's important not to trust responses blindly (due to hallucinations). To verify correctness, you can now compare Gemini's answers with Google search results.
Step-by-Step:
Head to Gemini (free & premium) and enter your question or prompt.
Click the Google logo under Gemini's reply.
Google checks Gemini’s answer and marks the text according to what it finds. Green highlighted text means it found similar content. Orange highlight means it found different content.
You can then click on the highlighted content to get the relevant link from Google’s sources so that you can read more.
PRESENTED BY LINDY
The Rundown: Lindy lets you build your own custom AI assistants to automate tasks and get things done 10x cheaper and 10x faster — no coding required.
With Lindy, you can:
Automate emails, meeting schedules, note-taking.
Connect your AI employee to over 3,000+ existing apps.
Streamline customer support, sales, recruiting, and more.
Click to sign up for Lindy for free and build your new AI employee.
UNIVERSITY OF MICHIGAN
Image source: Catalyst Research Alliance
The Rundown: The University of Michigan is facing major backlash on X (formerly Twitter) after a third-party vendor of the institution advertised datasets of lectures and papers for licensing to AI firms for training LLMs, without student consent.
The details:
A Google DeepMind engineer posted a screenshot of an email received that offered over 80 hours of academic speeches and 829 unpublished student papers.
The advertisement was from Catalyst Research Alliance, with sample datasets including lectures from 1999 — and a price tag as high as $25k for the papers and recordings.
However, a Michigan spokesperson said the information shared was ‘inaccurate’ and the data was ‘voluntarily contributed.’
Why it matters: The University of Michigan says the data was voluntary, but it’s tough to argue that students and professors in the late 90’s and early 2000’s were consenting to being part of LLM training. This incident highlights the inevitable ethical issues coming as human data becomes increasingly valuable for AI models.
NEW TOOLS & JOBS
💼 Meetingly- Transform sales calls into actionable insights effortlessly (link)
📝 Fixkey- Improve your writing everywhere on macOS (link)
👩🏻⚕️ Docus- Your health decisions, validated by top doctors (link)
🏌️♂️ GolfGPT- Your personal golf expert, helping you master every hole (link)
😰 Anxiety Simulator- This immersive experience offers unique insights into anxiety disorders (link)
🏃🏻♂️AI Runner- Implement stable diffusion on your hardware without any dependencies (link)
PS — we're hiring at The Rundown
QUICK HITS
OpenAI is reportedly developing a new web search service with Bing integration, looking to challenge Google's dominance in the search engine market.
Microsoft announced a new $3.44B investment in Germany focused on AI and data center development.
LangChain announced the public launch of its LangSmith platform for LLM app development, now open for general sign-up — while also revealing a fresh $25M Series A funding round.
Meta released V-JEPA, a new learning model for AI to better understand the physical world through video analysis and learn like humans do.
Slack integrated new generative AI features into the platform, including enhanced search, channel recaps, thread summaries, and more.
X is adding AI into its Explore tab, with trending topics set to get more context via Grok-generated summaries.
Magic announced a $117M raise led by former Github CEO Nat Friedman for its frontier-scale AI code models, seeking to build an ‘AI software engineer’.
THAT’S A WRAP
SPONSOR US
Get your product in front of over 450k+ AI enthusiasts
Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.
FEEDBACK
How would you rate today's newsletter?Vote below to help us improve the newsletter for you. |
If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.
Join the conversation