Unconventional AI

Reflection AI develops open foundation models targeting superintelligent autonomous systems, with current work focused on autonomous coding as a path to broader cognitive automation. The company combines reinforcement learning and large language models to build systems capable of handling most cognitive work on a computer, positioning autonomous code generation as the bottleneck to unlock that capability. The team includes contributors to AlphaGo, AlphaZero, PaLM, GPT-4, and Gemini, bringing production experience across game-playing RL systems and frontier language models. This background suggests familiarity with the trade-offs in training large-scale models - compute efficiency, sample complexity, and the operational challenges of running RL at scale alongside supervised pretraining. Reflection's stated objective centers on keeping superintelligence open and accessible through open foundation models. For inference practitioners, this implies potential work on model architectures, training infrastructure, and deployment systems designed for broad distribution rather than proprietary deployment. The autonomous coding focus suggests evaluation infrastructure for code generation, likely including metrics beyond pass@k - compilation rates, execution correctness, and performance characteristics of generated code under real-world constraints.

Hippocratic AI develops safety-focused large language models purpose-built for healthcare applications, with its flagship product Polaris deployed across over 150 million clinical patient interactions with zero reported safety issues. The company has raised $404 million at a $3.5 billion valuation to address the global shortage of 15 million healthcare workers through AI-powered clinical automation. Infrastructure runs on NVIDIA compute deployed via AWS, focusing on low-risk, non-diagnostic tasks where latency and reliability constraints differ from acute care workflows. Polaris implements a constellation architecture that coordinates multiple specialized agents rather than relying on a monolithic model - an approach that trades orchestration complexity for narrower failure modes in production. The system handles chronic care follow-ups, medication reminders, and patient engagement workflows where diagnostic responsibility remains with clinicians. The company has developed over 1,000 AI healthcare agents using retrieval-augmented generation to ground responses in clinical protocols, though specific latency profiles, throughput characteristics, and the operational overhead of managing agent deployments at scale remain publicly undisclosed. The technical approach prioritizes safety constraints inherent to healthcare applications: avoiding diagnostic or prescriptive capabilities, maintaining audit trails for clinical conversations, and operating within well-defined task boundaries. For engineers evaluating production ML systems, the trade-offs center on the constellation architecture's ability to handle distribution shift across patient populations versus the operational complexity of maintaining multiple specialized models. Led by CEO Munjal Shah, the company positions itself across the entire healthcare industry vertical, though deployment details beyond the AWS/NVIDIA stack and the distinction between research benchmarks and production performance in actual clinical settings warrant closer examination for those building similar safety-critical inference systems.