Christine McLeavey

Meta just released 4 models today. 🔥 - Meta Chameleon: 7B & 34B language models - Meta Multi-Token Prediction LLM - Meta JASCO: text-to-music models - Meta AudioSeal: audio watermarking model This is based on Meta's groundbreaking paper released back in April-2024 "Better & Faster Large Language Models via Multi-token Prediction" ✨ #AI #GenAI #generativeAI #machineLearning #Elon #Tesla #AI #Chatbot #MachineLearning #GoogleGemini #Innovation #Technology

3

3 Comments

Things just keep getting more interesting in the AI world. Today, something that pushes the current boundaries of both capability and accessibility. The Allen Institute for AI in Seattle, Ai2, has launched Molmo, a family of state-of-the-art open multimodal models. What’s impressive about Molmo? For one thing, Molmo doesn’t just understand images or visual data—it can act on them, such as pointing to elements on a screen or navigating complex interactions. Molmo can perceive, point, and act: this moves us closer to genuinely functional AI agents active in the real world. Molmo's also remarkably efficient. Molmo-1B model is claimed to be compact enough to run on most devices while still maintaining high performance. It proves what some have been saying all along: that smaller models can be just as powerful as larger, proprietary ones, if trained on higher quality data. The openness and efficiency of Molmo could represent a significant advance. Training A without massive compute requirements, opens doors for startups, labs, and companies that previously couldn’t compete in this space. Here's an example from five minutes playing around this morning. First I showed a picture of my cat, Fern, playing on the deck with a snail. Here's what Molmo described ... "A cat is walking on a wooden deck, intently focused on a small snail. The cat's paw is extended towards the snail, which is likely moving slowly across the deck surface. The scene captures a moment of natural curiosity as the cat observes this tiny creature. The wooden deck provides a textured backdrop for this interaction between predator and prey, highlighting the cat's instinctive interest in small moving objects." That's impressive enough. I said, "There's something else on the deck too." And here's the response ... "Yes, there's something else on the deck besides the snail. There's a small yellow butterfly visible on the deck surface. The butterfly adds another element of nature to the scene, contrasting with the cat's interest in the snail. This combination of a moving snail and a delicate butterfly creates an interesting and diverse display of small creatures on the wooden deck." Fascinating but wrong - there are leaves but no butterfly. Still, you can see it's an easy mistake to make. So I asked, "Show me the butterfly" and you can see in the attached illustration that Molmo can interact with my image and point out what it thought was a butterfly. That's a great capability and impressive, especially if the model is as small and efficient as claimed. Check it out here: https://allenai.org/

40

7 Comments