Do-Hyeon Lee

Geoffrey Hinton just had the most important AI interview of 2025. He revealed mind-blowing facts about AI that 99% of people wouldn't know... Including which career path to choose. You won't believe what you're about to hear. Here are his 8 most shocking insights: (thread) 🧵

Text-to-LoRA: Instant Transformer Adaption arxiv.org/abs/2506.06105 Generative models can produce text, images, video. They should also be able to generate models! Here, we trained a Hypernetwork to generate new task-specific LoRAs by simply describing the task as a text prompt.

Preprint of today: Beyer et al., "Highly Compressed Tokenizer Can Generate Without Training" -- github.com/lukaslaobeyer/… The latent space of tokenizers already provides a good enough abstraction to work with -- you don't have to use a diffusion model on top to inpaint, etc!

Are world models necessary to achieve human-level agents, or is there a model-free short-cut? Our new #ICML2025 paper tackles this question from first principles, and finds a surprising answer, agents _are_ world models… 🧵

Jürgen Schmidhuber and his lab always ship ahead of schedule 🚀✨

If you’re interested in learning about Continuous Thought Machines (sakana.ai/ctm/), we made interactive notebook tutorials so you can hack around with CTMs ImageNet: github.com/SakanaAI/conti… MNIST Tutorial: github.com/SakanaAI/conti… Let me know if you have any feedback!

AI that can improve itself: A deep dive into self-improving AI and the Darwin-Gödel Machine. richardcsuwandi.github.io/blog/2025/dgm/ Excellent blog post by @richardcsuwandi reviewing the Darwin Gödel Machine (DGM) and future implications.

New Paper! Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents A longstanding goal of AI research has been the creation of AI that can learn indefinitely. One path toward that goal is an AI that improves itself by rewriting its own code, including any code…

Why do brains rely on inference, uncertainty, and structure… while AI systems chase rewards in unstructured worlds? Are we missing something fundamental about how intelligence emerges? #NeuroAI #InferenceOverOptimization

1/3 @geoffreyhinton once said that the future depends on some graduate student being suspicious of everything he says (via @lexfridman). He also said was that it was impossible to find biologically plausible approaches to backprop that scale well: radical.vc/geoffrey-hinto….

You weren't dreaming— the @NotebookLM mobile app started rolling out this morning! We were eager to get the app into your hands, so this initial version has an MVP feature set with more functionality coming soon! Here are a few of the features we're most excited about: 🧵🧵🧵

I am pleased to announce a new version of my RL tutorial. Major update to the LLM chapter (eg DPO, GRPO, thinking), minor updates to the MARL and MBRL chapters and various sections (eg offline RL, DPG, etc). Enjoy! arxiv.org/abs/2412.05265

🚀Let’s Think Only with Images. No language and No verbal thought.🤔 Let’s think through a sequence of images💭, like how humans picture steps in their minds🎨. We propose Visual Planning, a novel reasoning paradigm that enables models to reason purely through images.

“I have always been convinced that the only way to get artificial intelligence to work is to do the computation in a way similar to the human brain.” — @GeoffreyHinton 🧠

Introducing Continuous Thought Machines New Blog: sakana.ai/ctm/ Modern AI is powerful, but it’s still distinct from human-like flexible intelligence. We believe neural timing is key. Our Continuous Thought Machine is built from the ground up to use neural dynamics as…

New Paper: Continuous Thought Machines 🧠 Neurons in brains use timing and synchronization in the way that they compute, but this is largely ignored in modern neural nets. We believe neural timing is key for the flexibility and adaptability of biological intelligence. We…

Flow Matching aims to learn a "flow" that transforms a simple source distribution (e.g. Gaussian) to an arbitrarily complex target distribution. This video shows the evolution of the marginal probability path as a source distribution is transformed to a target distribution.

I dove into the 264-page AI Agent Blueprint by top researchers from Meta, Yale, Stanford, DeepMind & Microsoft. They map AI Agent components—perception, memory, world modeling, reasoning, planning—to human brain. Remarkable work. 𝗛𝗲𝗿𝗲 𝗮𝗿𝗲 𝘁𝗵𝗲 𝗸𝗲𝘆 𝗳𝗶𝗻𝗱𝗶𝗻𝗴𝘀:…

Greenberg-Hastings cellular automata. Made with #python #numpy #scipy #matplotlib

MCP vs A2A (Agent2Agent) protocol, clearly explained: