Eugene Cheah, CEO of Featherless AI, presents a four-step framework for cutting AI inference costs at scale: measure token spend by team and workload, right-size models through evals, classifiers, and fine-tuning, buy GPU capacity with an understanding of the speed-versus-cost tradeoff, and check whether organizational processes can absorb the extra output.
A session on CoreWeave's forward deployed engineering partnership with the Aston Martin Aramco F1 Team, covering why Formula 1 is an unforgiving environment for AI, and three projects: a unified data platform for aerodynamic and telemetry data, Delta to Map, a small transformer model that reduces the gap between wind tunnel predictions and track measurements, and Radio Insights, a fine-tuned speech-to-text and agentic system that makes sense of 40 live radio channels. The discussion extracts lessons applicable to any team moving AI from experiment to production under fixed deadlines.
Zixuan Li, head of Z.ai, introduces GLM-5.2 and explains the company's history, from the 2021 GLM paper to a new high thinking level and near-frontier benchmark results. He makes the case for open weights through enterprise on-prem deployment, domain fine-tuning and community co-design, and closes by announcing Z Code, an open-sourced coding harness that works with other frontier models.
Daniel Svonava argues that small open-source models, when specialized and properly served, can achieve frontier performance for specific tasks, offering significant cost savings and latency improvements over large, generalized models or managed endpoints. He highlights challenges with existing serving infrastructure for a fleet of diverse small models and introduces Superlinked's open-source solution, which leverages a pull-based, centralized queuing system, a Rust sidecar for runtime abstraction, and an autoresearch loop for automated model tuning, enabling high throughput (e.g., half a million embedding tokens/sec on one GPU) and efficient deployment of specialized models and LoRAs.
Mistral AI's Voxtral suite represents a significant advancement in voice AI, moving beyond traditional cascaded systems towards more integrated and efficient audio understanding and generation. Pavankumar Reddy Muddireddy of Mistral AI details Voxtral Chat's audio-native architecture, which directly feeds continuous audio embeddings into a text LLM for rich context. He explains Voxtral TTS's use of continuous latents and flow matching for high-fidelity speech synthesis. The discussion also covers critical challenges like speaker diarization, the use of Direct Preference Optimization (DPO) to correct model hallucinations, and the enduring relevance of specialized model cascades in real-world enterprise deployments for their adaptability and control, emphasizing that voice typically thrives alongside a screen, not in isolation.
Ollama Co-founder Jeffrey Morgan discusses the enterprise shift to open models, driven by cost, customization, and coding agents, leading to a 150X token growth. He highlights the dominance of Chinese models in cloud consumption, the hardware renaissance enabling local LLMs, and the rise of ultra-efficient "flash" models. Morgan also shares Ollama's journey, from its YC pivot to its monetization strategy, and reflects on lessons for building AI companies.