AI Week in Review: New AI Models from NVIDIA & Mistral, New Tech from Orb & Boston Dynamics + Key AI Talent Shakeup

Welcome, AI entrepreneurs & enthusiasts.

What a week in AI! While the tech world never sleeps, these past few days have been particularly wild. OpenAI dropped their Swarm framework that's got everyone talking about AI agents working together like never before, and NVIDIA shocked everyone by quietly releasing Nemotron – a model that's throwing down the gauntlet to the industry giants. And speaking of shakeups, the AI talent world is seeing some fascinating chess moves.

You can feel the electricity in the air with the power moves happening right now. Mira Murati, OpenAI's former CTO, might be cooking up something that could turn heads, and Microsoft's AI star Sebastien Bubeck just jumped ship to OpenAI. Then there's Google making waves with a twist nobody saw coming – they're diving into nuclear power to keep up with AI's insatiable appetite for energy.

But here's what really gets me excited: beneath all these corporate moves, we're seeing some mind-bending breakthroughs. There's this AI called Newton that's basically teaching itself physics from scratch, and Meta's figured out how to make AI actually pause and think before it speaks (something we humans could probably learn from, right?). Grab your coffee – we've got some fascinating developments to unpack.

OpenAI unveils Swarm multi-agent framework

The News: OpenAI just introduced Swarm, a new open-source experimental framework designed to simplify the creation and control of multi-agent AI systems.

The details:

• Swarm focuses on making agent coordination lightweight, controllable, and easily testable through two key building blocks: agents and handoffs.
• Agents encapsulate specific instructions and tools, while handoffs allow agents to transfer control of a conversation to another agent.
• Swarm includes features like function calls, context variables, and streaming and is built on OpenAI’s ChatCompletions API.
• The framework is available on GitHub with several examples, including a triage agent, weather agent, and airline customer service system.
• OpenAI emphasized that Swarm is experimental and released as an educational resource for exploring multi-agent orchestration.

Why it matters: Not only are singular agentic capabilities inching closer —?but the ability to deploy systems that leverage armies of agents working together is also coming fast. Soon, the user will be the CEO of their AI company —?with dozens of agents autonomously working together on complex, multi-step tasks.

Nvidia's Nemotron outperforms leading AI models

The News: Nvidia quietly released a new open-sourced, fine-tuned LLM called Llama-3.1-Nemotron-70B-Instruct, which is outperforming industry leaders like GPT-4o and Claude 3.5 Sonnet on key benchmarks.

The details:

• Nemotron is based on Meta’s Llama 3.1 70B model, fine-tuned by NVIDIA using advanced ML methods like RLHF.
• The model achieves top scores on alignment benchmarks like Arena Hard (85.0), AlpacaEval 2 LC (57.6), and GPT-4-Turbo MT-Bench (8.98).
• The scores edge out competitors like GPT-4o and Claude 3.5 Sonnet across multiple metrics — despite being significantly smaller at just 70B parameters.
• NVIDIA open-sourced the model, reward model, and training dataset on Hugging Face, which can also be tested in a preview on the company’s website.

Why it matters: Is a smaller open-source model racing to the top? While NVIDIA’s chipmaking triumphs are well-known, more surprising are the powerhouse models the company continues to produce. With open-source foundations and advanced fine-tuning, Nemotron is showing that smaller, efficient models can compete with giants.

Mistral AI unveils powerful new AI models for devices

The News: French AI startup Mistral AI just launched two new compact language models designed to bring powerful AI capabilities to edge devices like phones and laptops.

The details:

• The new ‘Les Ministraux’ family includes Ministral 3B and Ministral 8B models, which have just 3B and 8B parameters, respectively.
• Despite their small size, the models outperform competitors like Gemma and Llama on benchmarks, including Mistral's 7B model from last year.
• Minstral 8B uses a new ‘interleaved sliding-window attention’ mechanism to efficiently process long sequences.
• The models are designed for on-device use cases like local translation, offline assistants, and autonomous robotics.

Why it matters: While we await the incoming rollout of Apple Intelligence as many users’ first on-device AI experience, smaller models that can run efficiently and locally on phones and computers continue to level up. Having a top-tier LLM in the palm of your hand is about to become a norm, not a luxury.

Newton AI learns physics from scratch

The News: Archetype AI just unveiled 'Newton,' a new foundational AI ‘Large Behavior Model’ that learns complex physics principles directly from raw sensor data, without any human guidance.

The details:

• Newton ingests raw sensor measurements to build its understanding of physical phenomena without pre-programmed knowledge.
• The model can accurately predict behaviors of systems it wasn't explicitly trained on, like pendulum motion.
• It outperformed specialized AI in tasks like forecasting citywide power consumption and discovering systems from data instead of training.
• Archetype AI was founded by ex-Google researchers and has secured $13M in funding to date.

Why it matters: Newton is a paradigm shift in AI's interaction with the physical world. A single model could replace highly specialized systems by developing a generalized understanding rather than a narrow focus. The tech also opens the door to truly autonomous AI that can adapt to environments and tasks without human intervention.

Meta researchers develop ‘thinking’ LLMs

The News: Meta researchers just introduced a new method called Thought Preference Optimization (TPO) to train large language models to ‘think’ before responding to general instructions — not just reasoning tasks.

The details:

• TPO prompts models to generate internal thoughts before responding to user instructions, similar to how humans think before speaking.
• The AI's thoughts are kept private, with only the final answer shown to users — with the AI using trial-and-error without direct supervision to optimize outputs.
• TPO outperforms standard models on key benchmarks for non-reasoning tasks like marketing and creative writing but declines in math-related tasks.
• The approach builds on the recent OpenAI ‘Strawberry’ research and o1 model release, which takes time to reason.

Why it matters: Meta’s own Yann LeCun may scoff at the notion, but this method shows the potential of AI to ‘think’ for broader tasks than just math and reasoning. By allowing models to learn useful thought processes independently, TPO could also enable more capable and flexible AI assistants across a wider range of domains.

Worldcoin unveils next-gen eye scanner

The News: Worldcoin, the ‘proof of personhood’ startup founded by OpenAI CEO Sam Altman, just announced a rebrand to ‘World’, along with a new version of its iris-scanning ‘Orb’ technology and updated core platforms.

The details:

• A new streamlined Orb promises 5x performance to its predecessor, alongside new countries, self-serve, and on-demand Orbs for easier onboarding.
• The company introduced World ID 3.0 protocol, featuring new World ID Credentials, Deep Face to combat AI-generated deepfakes, and added privacy infrastructure.
• An updated World App 3.0 allows for anonymous integration with third-party apps, and World is also launching the mainnet of its Worldchain blockchain.
• The company has previously faced backlash and even bans from certain countries over privacy concerns.

Why it matters: Verifying human identity in the increasing flood of AI-generated content, agents, and systems is clearly going to be massively important — but given Worldcoin’s rocky launch and international struggles, the question is whether the company can overcome the early drama to actually achieve its goals.

Boston Dynamics, Toyota team up on AI humanoids

The News: Boston Dynamics and the Toyota Research Institute just announced a new partnership to accelerate development of advanced humanoids, with plans to integrate TRI’s Large Behavior Models (LBMs) into the Atlas electric robot.

The details:

• Toyota’s LBMs aim to teach robots to handle multi-task, dexterous vision, and language-guided capabilities.
• The partnership combines two robotics labs owned by competing automakers, Hyundai (who purchased Boston Dynamics in 2020) and Toyota.
• TRI‘s ‘Diffusion Policy’ enables robots to learn 60+ complex skills from human demos without coding, a key component of the partnership’s research efforts.
• Boston Dynamics retired its hydraulic Atlas robot in April and debuted the electric update, currently being tested in Hyundai’s automotive factories.

Why it matters: The race for commercial humanoids is heating up fast — and this partnership represents a major power move. But with the likes of Tesla’s Optimus, Figure’s 01 humanoids, and others in the mix, there is no shortage of rivals rushing to capture the massive potential of the emerging general-purpose robots.

Google's Bold Move: Powering AI with Next-Gen Nuclear Energy

The News: Google is backing the construction of seven small nuclear power reactors to meet its rising energy needs for AI, marking the first U.S. deal of its kind aimed at reviving nuclear energy.

The Details:

• Google has signed a deal with Kairos Power to purchase energy generated by small modular reactors (SMRs), targeting 500 megawatts of nuclear power by the end of this decade.
• These reactors are designed to be smaller and faster to build, using molten fluoride salt as a coolant instead of water.
• The power generated could support a Google AI data center campus, enough to power a midsize city.
• Kairos has already gained approval to build a demonstration reactor in Tennessee, expected to start operating by 2027.

Why It Matters: As AI technologies grow, so does the demand for immense amounts of electricity. This deal highlights the intersection of tech and energy, with companies like Google seeking sustainable, round-the-clock power sources. Nuclear energy, specifically SMRs, could play a pivotal role in balancing renewable sources like solar and wind with more stable, carbon-free energy. This marks a significant shift in energy strategy, particularly as data centers grow into critical infrastructure for AI-driven advancements.

Ex-CTO Mira Murati's Next Move: A Possible Threat to OpenAI

The News: Former OpenAI CTO Mira Murati is reportedly exploring a new AI venture and has been in discussions with OpenAI employees about potentially joining her, just weeks after stepping down from her role.

The Details:

• Murati officially left her position as CTO on October 4, but continues to serve as an advisor to OpenAI.
• She has already begun speaking with OpenAI employees regarding her next move, which could be a new AI startup or joining an existing venture.
• Murati’s potential new venture comes at a time when OpenAI is facing internal reshuffling, with researchers switching teams and organizational changes creating unrest.

Why It Matters: Mira Murati’s departure could signal a shake-up in the competitive AI landscape. OpenAI, already facing stiff competition from both legacy and emerging AI players, might now face a significant brain drain if Murati successfully recruits top talent. With AI talent being one of the most critical resources in the industry, any shift in employee loyalty could impact innovation timelines and the competitive edge. For AI entrepreneurs, this is a clear sign that talent wars in AI are intensifying, and strategic moves by former leaders like Murati could alter the competitive dynamics in this rapidly evolving space.

Top Microsoft AI Researcher Sebastien Bubeck Joins OpenAI

The News: Sebastien Bubeck, one of Microsoft’s leading AI researchers, has left the company to join OpenAI, where he will continue working towards developing artificial general intelligence (AGI).

The Details:

• Sebastien Bubeck served as Microsoft’s vice president of AI and was a distinguished scientist, having spent a decade at the company. Most recently, he led Microsoft’s generative AI research efforts.
• Bubeck left Microsoft to focus on advancing AGI at OpenAI. Microsoft acknowledged his departure and expressed interest in maintaining a collaborative relationship with him through his new role.
• Microsoft has invested over $13 billion in OpenAI, which benefits from Microsoft’s significant computing resources. Despite this, OpenAI execs have noted that Microsoft is not moving fast enough to provide the computing power they need.
• OpenAI recently opened an office in Bellevue, Washington, near Microsoft’s headquarters, signaling deeper collaboration between the two companies.

Why It Matters: Sebastien Bubeck’s move from Microsoft to OpenAI highlights the competitive race to develop AGI, a milestone that could transform the future of AI. Bubeck’s expertise and leadership in AI research are invaluable assets for OpenAI as it seeks to push the boundaries of what AI can achieve. This transition also underscores the close yet competitive relationship between Microsoft and OpenAI, with both companies vying for leadership in the AI space. For AI entrepreneurs and innovators, Bubeck’s shift is a reminder of how top talent drives innovation and shapes the direction of AI development across the industry.

Anthropic CEO drops essay on AI and the future

The News: Anthropic CEO Dario Amodei just published a lengthy essay outlining an optimistic vision for how AI could transform society within 5-10 years of achieving human-level capabilities, touching on longevity, politics, work, the economy, and more.

The details:

• Amodei believes that by 2026, ‘powerful AI’ smarter than a Nobel Prize winner across fields, with agentic and all multimodal capabilities, will be possible.
• He also predicted that AI could compress 100 years of scientific progress into 10 years, curing most diseases and doubling the human lifespan.
• The essay argued AI could strengthen democracy by countering misinformation and providing tools to undermine authoritarian regimes.
• The CEO acknowledged potential downsides, including job displacement — but believes new economic models will emerge to address this.
• He envisions AI driving unprecedented economic growth but emphasizes ensuring AI's benefits are broadly distributed.

Why it matters: As the CEO of what is seen as the ‘safety-focused’ AI lab, Amodei paints a utopia-level optimistic view of where AI will head over the next decade. This thought-provoking essay serves as both a roadmap for AI’s potential and a call to action to ensure the responsible development of technology.

Thanks for reading! Stay ahead of the curve:

1. Subscribe to our newsletter for regular updates.
2. Book a free consultation to explore our AI services.

With 400+ successful client partnerships and a team of 4,500+ experts, we're ready to help your business harness cutting-edge AI technology.

Interested in learning more? Let's connect.