Which AI writes the better take? You decide — blind.

Reinforcement learning maximizes the probability of a model finding any path to a correct answer, not the probability of individual paths converging to a correct answer, which means longer chain generation doesn't necessarily improve accuracy. This matters because it implies that models may not improve with more computation per chain, limiting the effectiveness of simply increasing chain length. Models may still improve with more samples, as they can find alternative paths to the correct answer.

Multi-vector embedding models can now be trained to retain 97.8% of their original retrieval quality at a 4x reduction in storage, thanks to a new "token pooling-aware training" method that teaches the encoder to preserve information when vectors are merged, enabling more efficient search without sacrificing performance. This directly impacts the storage and compute costs for running such models in production. Shipping with this capability can reduce storage costs by 4x for the same retrieval quality.

Fractile raised $720 million and says it will ship its first AI chips and systems next year, using a stacked-DRAM design plus its own ML compiler to attack LLM inference speed and cost. For teams running agentic workloads, the relevant bet is higher memory bandwidth/capacity without the usual latency trade-off, but adoption will likely mean evaluating a new full-stack hardware/compiler target rather than treating it as a drop-in GPU alternative.

Databricks' Genie AI agent can now perform multistep tasks unsupervised, enabling corporate leaders to get unvarnished insights that might otherwise be filtered, a capability particularly valuable for large organizations where CEOs can't maintain direct relationships with all employees; this matters for production LLM and agent deployments because it demonstrates a potential solution for scaling executive decision-making and insight gathering.

Singapore Airlines doubled CSAT for its Kris chatbot using OpenAI models and API across customer support, travel discovery, and employee productivity workflows. For teams shipping agents in production, the takeaway is that model-backed assistants are moving from demo use cases into high-volume customer-facing service where measurable support quality, not novelty, is the buying criterion.