#448: DeepSeek Has Introduced Advanced Reasoning To The Open-Source Community, & More
1. DeepSeek Has Introduced Advanced Reasoning To The Open-Source Community

Known for its v3 frontier model, China-based AI lab DeepSeek introduced its first reasoning model, R-1 last week. Much like OpenAI’s o series, R-1 leverages chain-of-thought reasoning to tackle complex logic and math challenges and generate more accurate and coherent answers than conventional large language models. Benchmarks indicate that the open-source R-1 model not only rivals OpenAI’s closed-source o1 model but also surpasses other contenders in the market.
In a detailed paper,1 DeepSeek revealed how a version of R-1, dubbed R-1 Zero, evolved from a “pure reinforcement learning” training regimen. With DeepSeek v3 as the base, engineers challenged the model to solve advanced logic and math problems, leading to the development of “emergent” reasoning without massive amounts of human-labeled chain-of-thought data. The approach pays homage to AlphaGo Zero with self-directed learning and is the first time this training method has been applied successfully to language models in a detailed publication. The open-sourced version of R-1 includes additional training steps on top of R-1 Zero’s pure reinforcement learning approach to improve interpretability of its thinking patterns and ensure alignment with user preferences and safety guidelines.
DeepSeek also showcased R-1’s versatility as a “trainer model” that can enhance the performance of smaller models, including Llama, through “distillation.” By transferring knowledge and refined reasoning to models with fewer parameters, R-1 underscores the potential for advanced AI to enhance reasoning in a broader range of models.
Two months after OpenAI released the full version of o1, R1 exemplifies DeepSeek’s iterative ability, propelled perhaps by Chinese government support. R1’s accelerated timeline also highlights the intensity of the global AI race after the Biden Administration’s eleventh-hour crackdown on AI exports2 and the Trump Administration’s announcement of Stargate3—a $500 billion US AI infrastructure initiative led by OpenAI’s Sam Altman, Oracle’s Larry Ellison, and Softbank’s Masayoshi Son.
2. Gene Editing Is Advancing A One-Time Treatment For Prion Disease

Prion diseases are fatal neurodegenerative disorders caused by the misfolding of the prion protein (PrP) encoded by the PRNP gene. A well-known example is mad cow disease. Once untreatable, researchers at the Broad Institute are demonstrating that prion diseases can be treated with in-vivo base editing, a groundbreaking gene-editing technique that has extended the lifespan of those with prion disease in humanized mouse models.4 This breakthrough marks a critical step toward developing a one-time therapeutic intervention.
Using adeno-associated virus (AAV) delivery, the scientists used base editing to introduce a protective mutation, R37X, into the PRNP gene, reducing prion protein levels by 50–63% and extending the mice lifespan by approximately 50%. Remarkably, with minimal off-target effects, those edits mimic a naturally occurring mutation that has protected humans against prion disease.
Science is demonstrating that gene-editing can transform the treatment of prion disease and other devastating neurodegenerative disorders, shifting the paradigm toward highly targeted, one-time therapies.
3. OpenAI Releases Operator, A Browser-Based Agent

Last week, OpenAI launched Operator,5 an AI agent which could redefine browsing and the autonomous execution of tasks. Leveraging human interfaces—monitor, keyboard, and mouse—Operator navigates private cloud-hosted browsers to execute everyday tasks like booking restaurant reservations through OpenTable, ordering groceries with Instacart, and purchasing tickets to sporting events on Stubhub. Operator has the flexibility to navigate most websites, underscoring the disruptive potential of this technology to enhance productivity, convenience, and reach.
While competitors have ventured into this space, Operator seems to be the most accessible and powerful offering. After Anthropic tested Sonnet 3.5 with Computer Use6 and then Google rolled out its beta version of Project Mariner,7 the reviews were mixed given their limited availability and performance. In contrast, Operator already is live for ChatGPT Pro subscribers, with plans for an application programming interface (API) option and a broader rollout to ChatGPT Plus and Enterprise in the near term. Given its early-mover advantage, OpenAI could capture useful data on agent-browser interactions and catalyze growth in automated AI-driven services—reinforcing our belief in the exponential impact of frontier technologies on daily life and the economy.
4. 3D Printing Reshapes Automotive Aftermarket Supply Chains

Last week, 3D Systems partnered with Daimler Bus to enable distributed 3D printing of spare parts. Digital IP-protected files sent to local service bureaus can 3D-print parts on demand, potentially cutting the time to parts by 75% compared to more centralized manufacturing approaches.8 The new solution, built for 3D Systems’ SLS 380, 3D prints production-grade nylon parts but could be adapted to metal and plastic printers.
This announcement highlights a significant market opportunity for 3D printing. In the US alone, the automotive aftermarket parts industry generates ~$200 billion in annual revenue, 10X the size of the ~$20 billion global 3D printing market.9 The advantages of printing parts on demand include lower inventory and manufacturing costs for low-volume production runs. In some instances, digital design files can replace physical molds, reducing shipping costs per part.
ARK estimates that 3D printing could grow to a $180 billion industry by 2030, automotive representing one of its largest opportunities. Stay tuned for more insights in our upcoming Big Ideas report.
-
1
DeepSeek-AI. 2025. “DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning.” arXiv.
-
2
U.S. Bureau of Industry and Security. 2025. “Biden-Harris Administration Announces Regulatory Framework for the Responsible Diffusion of Advanced Artificial Intelligence Technology.”
-
3
OpenAI. 2025. “Announcing the Stargate Project.”
-
4
An, M. et al. 2024. “In vivo base editing extends lifespan of a humanized mouse model of prion disease.” Nature Medicine. DiCorato, A. 2025. “Gene editing extends lifespan in mouse model of prion disease.” The Broad Institute.
-
5
OpenAI. 2025. “Introducing Operator.”
-
6
Anthropic. 2024. “Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku.”
-
7
Google DeepMind. 2024. “Project Mariner.”
-
8
3D Systems. 2025. “3D Systems & Daimler Truck | Daimler Buses Innovations Maximize Vehicle Uptime by Decentralizing Spare Part Production.”
-
9
Apex Blog. 2024. “Automotive Aftermarket Industry Analysis 2023.” 3D Printing Industry. 2024. "3D Printing Industry hits new high at $20 billion according to Wohlers Report 2024, deeper insight from Terry Wohlers.