Are you ready for this? Text-to-movie, million character multimodal prompts, a non-generative AI model, Memory (yesterday's news)
We shouldn't be surprised by these developments, and it's time to level with educators
I’ve spent a lot of time in the last few weeks helping my debaters prepare for some important tournaments where the debates will be over whether single-use plastics should be banned.
It’s funny to think about, though SUPs do cause a lot of environmental damage). But the debate topics never mattered to me that much. Regardless of what they debate, students learn how to learn, how to think critically, how to organize ideas, how to process large amounts of information (they’ve turned up every stone on this topic; there are SUPs in cigarettes and personal hygiene products), and how to build some literacy around important issues.
But here’s what matters in the world and what students and teachers need to learn about while they are developing their debating skills: AI!
There were three huge developments today that have been shocking to most people.
I’ll admit I’m shocked that they are here already, as I was thinking 6-12+ months out, but as someone who follows the space, they aren’t surprising to me.
#1. Most people won’t realize the significance of this, but Meta released JEPA, a non-generative model that Yann LeCun has been thinking about and developing for years. This has the potential for advanced reasoning, planning, prediction, and accomplishing complex tasks. It also enables substantially faster learning.
JEPA architectures can employ strategies such as improved training data, cross-checking mechanisms, enhanced attention mechanisms, human feedback loops, regular updates, limiting output confidence, and integrating external knowledge bases that will allow them to substantially lower hallucinations.
#2. OpenAI basically released text-to-movie :)
NVIDIA Senior Research Scientist and “Agent” Lead Jim Fan explains the significance: If you think OpenAI Sora is a creative toy like DALLE, ... think again. Sora is a data-driven physics engine. It is a simulation of many worlds, real or fantastical. The simulator learns intricate rendering, "intuitive" physics, and long-horizon consistency, all by some denoising and gradient maths….Next up: add more modalities and conditioning, then we have a full data-driven UE that will replace all the hand-engineered graphics pipelines.”
#3. Gemini Upgraded its free “Pro” version (replaced Bard) to 1.5
Gemini 1.5 Pro can understand, reason about, and identify curious details in the 402-page transcripts from Apollo 11’s mission to the moon.
1.5 Pro can perform highly sophisticated understanding and reasoning tasks for different modalities, including video.
1.5 Pro can perform more relevant problem-solving tasks across longer blocks of code. When given a prompt with more than 100,000 lines of code, it can better reason across examples, suggest helpful modifications, and give explanations about how different parts of the code works.
Don’t forget that *yesterday* OpenAI started allowing you to enable a memory feature that allows ChatGPT/your AI to remember what you do, arguably the first step in creating a digital twin
The world is going to change a lot. It already has.
Are we going to keep this all a secret from our faculty and students? Eventually, they’ll find out that Santa Clause isn’t real and that AI is going to turn education and the world upside down. Are we just going to surprise them with that some day?
Back to the plastics…