- Lore
- Posts
- Lore #71: OpenAI Reveals Biggest Breakthrough in AI Since ChatGPT with Sora
Lore #71: OpenAI Reveals Biggest Breakthrough in AI Since ChatGPT with Sora
Nathan Lands
February 21, 2024
Good morning.
This has been the craziest week in AI since the release of ChatGPT.
Sora, an AI video technology by OpenAI, came out of nowhere last week and blew everyone away with AI Video demos that make everything else we’ve seen so far look “cute” in comparison.
I suggest checking out the demo videos here if you haven't seen them yet. They’re truly mindblowing. It feels like you’re watching people walk around a real world that was computer generated. I have a suspicion that GPT-5 was somehow involved in the creation of this.
Also, this week, Google launched Gemini 1.5 with a far larger context window than ChatGPT. People are already putting entire codebases in it and analyzing them for bugs, or having it read several books and draw meaningful insights from them.
Every week, we’re taking another step forward into uncharted territory. Exciting times.
On a personal note, I’ve begun filming a project with a friend of mine and a major company that many of you know. I hope to have more to share about it soon.
Moonshot raised $1B for its chatbot Kimi Chat and LLM (Link)
Lambda raised $320M for AI hardware and infrastructure (Link)
Recogni raised $102M to design AI chip (Link)
Bioptimus raised $35M to build biology foundation models (Link)
Rasa raised $30M to advance conversational AI innovation (Link)
Flower Labs raised $20M for software that allows AI models to be trained ‘at the edge’ (Link)
Clarity raised $16M for deepfake detection (Link)
AZmed raised €15M for AI-powered radiology software (Link)
Armilla AI raised $4.5M to safeguard enterprises from risks of AI (Link)
UnityAI raised $4M to revolutionize hospital flow (Link)
Cambio raised $3M to revolutionize debt negotiation with AI (Link)
Levr raised $1M for its AI-powered loan platform for small businesses (Link)
Coming up:
GPU cloud company Together AI to raise $100M (Link)
Join the AI Video revolution with Lore Labs
Step into the forefront of AI innovation with Lore Labs. As a pioneering AI Video Agency, we're reshaping how stories are told and experiences are created.
What Lore Labs offers:
Creative Storytelling: Using advanced AI, we bring any concept to life, from advertisements to movie pitches.
Cutting-edge Technology: With tools like Runway and Pika, we're redefining the bounds of digital creativity.
Customized Solutions: Tailoring AI video to fit your unique needs, ensuring memorable and impactful results.
Diverse Applications: Whether it’s for social media, real estate, or entertainment, our AI expertise caters to all.
Embark on a journey of limitless possibilities with Lore Labs. Join our waitlist and transform your vision into reality with the next generation of AI video.
OpenAI has announced Sora, a text-to-video diffusion model capable of generating up to 60-second videos with intricate scenes, complex camera movements, and dynamic character emotions. This model marks a significant step towards understanding and simulating real-world interactions in motion, aligning with the aim to develop models that assist in solving real-world problems. Sora can create videos based on detailed prompts, featuring a variety of subjects from animated cityscapes to natural environments and futuristic settings. It's now available for red team assessment and feedback from visual artists, designers, and filmmakers to refine its capabilities for creative uses.
Google Labs has announced the release of Gemini 1.5, the next-generation AI model featuring a Mixture-of-Experts (MoE) approach, offering faster and higher quality responses. Available for Private Preview, Gemini 1.5 Pro is optimized for a wide range of tasks and introduces an experimental 1 million token context window, surpassing the previous limit of 200,000 tokens. This expansion allows developers to process large datasets, including PDFs, code repositories, or videos, in a single prompt. Gemini 1.5 Pro's capabilities include analyzing entire codebases and reasoning across up to an hour of video content.
Meta has released the Video Joint Embedding Predictive Architecture (V-JEPA), advancing towards Yann LeCun's vision of Advanced Machine Intelligence (AMI). V-JEPA, designed to understand complex object interactions in videos, uses a self-supervised learning method for efficiency and generalization across tasks. It predicts video content in abstract representation space, allowing it to learn from unlabeled data, with significant implications for future AI development. This model paves the way for AI that can plan and make decisions over longer periods, potentially enhancing AI assistants and embodied AI technologies.
Reddit has entered into a $60 million annualized contract with an unnamed AI company, permitting the use of its social media content to train artificial intelligence models. This move comes as Reddit approaches its anticipated IPO, with the company being advised to seek a valuation of at least $5 billion. Read what people are saying about it here.
Other news
Elon Musk's X Enters the Artistic Fray: A Potential Partnership with Midjourney AI Art on the Horizon (Link)
StabilityAI: Introduces Stable Cascade (Link)
Microsoft to expand its AI infrastructure in Spain with $2.1 billion investment (Link)
Sanofi CEO: AI promises a great era of drug discovery that could fundamentally change medicine–but only if we allow it to deliver (Link)
Artificial intelligence, real emotion. People are seeking a romantic connection with the perfect bot (Link)
Generative AI dominates Google's data and analytics roadmap (Link)
A Conversation with the Jensen Huang of Nvidia: Who Will Shape the Future of AI?
First impressions of Sora
Sora from @OpenAI is super impressive, but how consistent are the geometries? We ran this through our fast 3DGS pipeline, and here are some of the early results. This is a reconstruction 👉 1/n
— Ge Yang (@EpisodeYang)
Feb 15, 2024
First impressions of Gemini 1.5
Ok, gemini 1.5 is really really good.
seriously the most impressive jump i've seen in models in a while when it comes to long context
just tried with a paper and asked it "what does Figure 5 show?"
it contextualized the whole thing and answered based on that tiny section +… twitter.com/i/web/status/1…
— Sully (@SullyOmarr)
Feb 17, 2024
Karpathy's AI Operating System
Shanghai 🇨🇳 AI Lab Achieved 1st Version of Karpathy's AI Operating System
> Nov 2023: @karpathy proposes LLM OS
> Feb 2024: Chinese (+ Princeton) team proposes self learning operating systemWhat did they do?
> Built an agent using a mix of Python code and GPT-4 language model… twitter.com/i/web/status/1…— Ate-a-Pi (@8teAPi)
Feb 14, 2024
Altera: an AI Agent you can directly play with in Minecraft
First tweet from Altera: We built the first AI Agent you can *directly play with* in Minecraft. Instead of just helping the player complete tasks, the bot acts more like a friend.
Early access link— Robert Yang (@GuangyuRobert)
Feb 20, 2024
Find more tools at FutureTools.io.
Chain-of-Thought Reasoning Without Prompting (Link)
Linear Transformers with Learnable Kernel Functions are Better In-Context Models (Link)
BASE TTS: Lessons from building a billion-parameter Text-to-Speech model on 100K hours of data (Link)
Aya Model: An Instruction Finetuned Open-Access Multilingual Language Model (Link)
Generative Representational Instruction Tuning (Link)
Huh: “we show that LLM agents can autonomously hack websites, performing tasks as complex as blind database schema extraction and SQL injections without human feedback. Importantly, the agent does not need to know the vulnerability beforehand.” arxiv.org/html/2402.0666…
— Ethan Mollick (@emollick)
Feb 15, 2024
That's it!
See you next week!
If you want to support Lore, please:
Share this newsletter
Follow us on Twitter @NathanLands & Lore.
Join our Generative AI community on LinkedIn.