AI Summary
→ WHAT IT COVERS Cortical Labs CEO Han presents the company's biological computing platform, which fuses human neurons with silicon chips. The CL1 device houses up to 2 million neurons, runs on 30 watts, and has been deployed at five major US research institutions. A biological data center with 120 units now operates in Melbourne, with a 1,000-unit Singapore facility planned. → KEY INSIGHTS - **Biological vs. GPU efficiency:** In reinforcement learning benchmarks, Cortical Labs' neuron-based systems demonstrated 5,000 times greater sample efficiency than GPU-based systems. GPUs compensate by accelerating simulated time, but that advantage disappears in physical-world robotics, where real-time constraints apply. This makes biological computing a strong candidate for training embodied AI agents and autonomous systems operating in the physical world. - **Energy economics of biological compute:** Each CL1 unit consumes approximately 30 watts, compared to kilowatts required by GPU server racks. This allows data center partners like DayOne in Singapore to add biological compute capacity without affecting their regulated energy budgets. A 1,000-unit Singapore facility will operate within a 200-megawatt government-mandated cap, making biological compute a viable path around energy constraints. - **On-site neuron manufacturing eliminates supply chain risk:** The Singapore data center will include an adjacent laboratory to grow neurons on-site, removing dependence on external cell shipments. This decentralizes the biological compute supply chain so each facility becomes self-sufficient, similar to a data center manufacturing its own chips locally. Maintenance requires swapping filtration cartridges every four to six months, analogous to replacing kidneys. - **Cloud access democratizes biological computing:** Rather than requiring buyers to maintain lab infrastructure, Cortical Labs offers cloud access to its Melbourne data center via Python SDK and Jupyter notebooks. A Stanford student with no biology background built a working Doom-playing biological computer through a hackathon using only the API. Developers can pip-install the SDK and access live neural activity streams remotely from any location. - **Consciousness avoidance as a hard ethical boundary:** Cortical Labs has established a firm internal policy against creating conscious systems, because consciousness introduces the capacity for suffering. The company monitors neural activity through its cloud platform and works with bioethicists proactively. The Vatican reviewed the technology and concluded current applications are ethically acceptable, partly because use cases center on disease modeling, toxicology, and movement disorder research. - **Neuron count context for compute benchmarking:** The CL1 operates with approximately 200,000 neurons per instance, comparable in scale to a fly or cockroach nervous system. Unlike LLM parameter counts, neuron counts do not directly map to intelligence but do confer generalized adaptability that silicon systems lack. Cortical Labs is exploring PDMS microfluidic segmentation to partition a single chip into multiple isolated compute instances, similar to virtual machine architecture. → NOTABLE MOMENT During a live demo, Han remotely accessed a Melbourne lab from New York and delivered an electrical stimulus directly to a neuron culture, visibly triggering a burst of neural activity on screen. He noted the ethical irony that this would be unacceptable if the system were conscious, describing it as essentially waking something up without warning. 💼 SPONSORS [{"name": "LinkedIn", "url": "https://linkedin.com/twist"}, {"name": "Quo", "url": "https://quo.com/twist"}, {"name": "Deal", "url": "https://deel.com/twist"}] 🏷️ Biological Computing, Neuromorphic Hardware, Reinforcement Learning, AI Ethics, Data Center Infrastructure, Biotech Startups