New AI Models to Watch

New! Genie 2 from Deep Mind redefines 'world model' as.. literal worlds. Text in; full 3D explorable digital worlds out. Let's dive in. Genie 2 is a world model - in that it can simulate virtual worlds, including the consequences of taking any action (e.g. jump, swim, etc.). Genie 2 responds to control actions, moving characters and objects it correctly. Pretty rad. The results are pretty remarkable to see (lots of demo videos like the one below on the Deep Mind page), and a big deal for gaming and entertainment. But beyond the capability itself, Genie 2 illustrates two compelling themes in AI: 1️⃣ First broader trend is the usefulness of emergent capabilities. 🖼️ Genie 2 correctly renders and persists the world, 3d models, and characters within it with accurate lighting, physical interactions, reflections, and character animations. Most of these were present but not explicitly encoded in the video training set. Instead, the model learned to do the right thing as as the size scaled, was able to generate more and more sophisticated outputs. 🎭 As different modalities are mixed, more sophisticated emergent capabilities appear - making multi-modality large models like this an exemplar of why model capabilities still have a lot of runway in terms of new capabilities. 2️⃣ Second trend is the 'materialized' interface. 📽️ We have seen a lot of usefulness from chatbots, but it seems unlikely that this will be the final evolved form of AI interactions. Instead - the trend is towards highly customized, fully automated, real time manifestations of a user experience tailored to the need of the end user. 🎀 In this case, that interface is a virtual 3D world, but it could just as easily be a 'desktop app' (useful for adoption since so many are familiar with this style of interafce), or a fully realized, custom, skeuomorphic interfaces (similarly useful because they mimic real world interactions). Chatbots and assistants will continue to play a role, but we are on a path where interfaces become much more dynamic (and progressing quickly). Very impressive work. Kudos to the Deep Mind team.

NVIDIA had an interesting announcement this month regarding its new “Cosmo” AI model, which builds an internal representation of the physical world in order to optimize outputs. This shift from analysis to simulation is powerful, as we have also seen with Physics-Informed Neural Networks (PINNs), which our portfolio company Basetwo AI uses to revolutionize process manufacturing. Why does this matter? Because world models, PINNs, and similar AI architectures represent a paradigm shift in AI, moving from passive pattern recognition to active reasoning and decision-making. While challenges remain, these models could be a huge step forward in making AI more adaptable, explainable, and effective in applications that impact the real world. For business leaders, understanding this shift is crucial, not just to leverage AI’s potential, but to prepare for a future where machines do more than just analyze data. You can read more about it in my latest AI Atlas:

Over the weekend, a small Chinese hedge fund turned star AI research outfit launched DeepSeek R1, a new massive open-weights model with state-of-the-art performance, trained on a shoestring budget. Just how much interest is there in this advance? I analyzed R1 downloads on Ollama, and I recorded my steps to perform this analysis with AI using speech, an AI model, & a developer environment. See the video below if you’re curious how I did it. As the chart above shows, there’s a lot of interest. R1 tops the charts in terms of daily downloads. It’s still relatively early though in terms of overall downloads. And of course, all model download patterns follow a decay function with most of the interest occurring at the beginning. Many of these models are weeks older. Some like Gemma & Phi are small models ; others like Llama3.3 include much larger versions. Two implications emerge from the R1 news : First, this innovation comes on the heels of a Christmas launch of Deepseek’s v3 model which prioritized latency, shows that the overall pace of innovation in AI presses forward unabated. Second, R1’s technical approach highlights an emerging bifurcation in the AI model landscape. The team’s use of quantization - a sophisticated compression technique that maintains 90-95% accuracy - points to a future with two distinct model categories: - High-speed, compressed models optimized for immediate tasks like table reformatting & quick analysis - Research-oriented models built for complex, multi-step reasoning (similar to Gemini’s Deep Research) R1 is a reasoning model. It’s chatty nature means it explicitly reasons & makes its plans clear to the user. For work that might take 10-15 minutes, this technique should reduce errors. It’s similar to Gemini’s Deep Research model. The launch of DeepSeek R1 reinforces two key trends in AI: the rapid pace of innovation & the emerging split between fast, lightweight models & more deliberate reasoning models. Looking at the download data, the market shows clear interest in both approaches. Here’s a step-by-step video on how I assembled this analysis. https://lnkd.in/g6y9Ev29

OpenAI’s newest model, o1, is here! (smarter than ever, but there’s a catch) o1 introduces a leap in intelligence with its focus on reasoning and reflection, aiming to address the limitations of previous large language models (LLMs). The o1 series comes in two versions: o1-preview, the more powerful and expensive model, and o1-mini, which is significantly cheaper, faster, and, in some cases, even outperforms its larger counterpart. Both models excel at delivering logical, step-by-step solutions to complex problems, particularly in coding, science, and technical fields. What sets o1 apart from earlier models is its integration of the “Chain of Thought” technique directly into its training phase. Unlike the prior approach, where prompts guided models to think through problems step by step, o1’s training allows it to reflect and solve challenges methodically from the outset. This change drastically enhances its ability to solve problems that demand careful reasoning. While the o1 models are particularly strong in technical areas, they come with trade-offs. The slower speed and focus on reasoning make o1 less suited for tasks like creative writing. Additionally, o1 lacks some voice capabilities and doesn’t support image generation or document uploads. That said, the model shines in safety improvements—fewer jailbreaks and hallucinations—making it an appealing choice for enterprise applications. Access to o1 is currently limited, with users able to submit a set number of requests per week. However, this powerful model is already available to all ChatGPT Pro and Teams users, meaning you can start testing it for yourself. So, is o1 a step toward AGI (Artificial General Intelligence)? While not quite there, it’s a significant stride in the right direction. o1’s ability to handle tasks that require more advanced reasoning brings us closer to the long-held dream of AGI. The introduction of o1 is bound to spark innovation across the AI community. Expect to see more advancements in agentic frameworks, as well as new models from competitors like Anthropic and Meta.

Deep Cogito has stepped into the spotlight, redefining possibilities with their innovative hybrid AI models, Cogito 1. These models effortlessly toggle between rapid answers and deeply reasoned responses, setting new benchmarks for efficiency and intelligence. 🔹 With parameter sizes stretching up to an impressive 671 billion, these models promise extraordinary problem-solving capabilities. Just imagine the potential impact across industries—from real-time data analysis to complex scientific computations. 🔍 What truly captures my attention is their novel Iterated Distillation and Amplification (IDA) approach. This methodology draws inspiration from self-play strategies, fostering self-improvement and enhanced reasoning, independent of traditional human feedback loops. 📊 The performance speaks for itself: Cogito 1 models are outshining the likes of DeepSeek AI and Meta in various benchmarks, demonstrating superior multilingual capabilities and exceptional tool-calling accuracy. Deep Cogito’s dedication to open-source accessibility means that these advancements aren't locked behind closed doors but are available for enterprises and developers alike, signaling a new era of collaborative innovation. #AIInnovation #HybridAI #MachineLearning #DeepCogito #Technology #FutureOfAI