Cover image

The 96% Speedup That Stops at the Optimizer

TL;DR for operators MOSHO reports a 96.4% reduction in median optimizer runtime against NSGA-II when searching for neural architectures that balance predictive accuracy and device-specific inference latency. Across 18 benchmark contexts at 15,000 architecture evaluations, MOSHO takes 0.64 seconds versus 17.72 seconds for NSGA-II. Its normalized hypervolume is 0.9987, slightly below NSGA-II’s 0.9993. ...

November 14, 2026 · 8 min · Zelina
Cover image

Thirty Days of Release, One Needle to Pass

TL;DR for operators Developing a monthly injectable medicine requires more than slowing drug release. The formulation must also pass through a needle, form a stable depot, and release its active ingredient at an acceptable rate. In an industrial study, Bannigan and colleagues1 combined automated experimentation with AI-guided optimization to evaluate 181 formulations over approximately 15 weeks, identifying four leads with acceptable reported administration properties and differentiated 30-day in vitro release profiles. ...

October 29, 2026 · 9 min · Zelina
Cover image

Better Fit, Worse Next Move

TL;DR for operators When each experiment or simulation is expensive, selecting a surrogate because it fits existing observations well can be the wrong criterion. In the paper’s controlled surrogate ablation, the most flexible model achieves the lowest training error but materially worse prediction away from observed points and much more optimistic estimates of candidate value. ...

October 18, 2026 · 7 min · Zelina
Cover image

One Point to Rule Them All: Why AI Optimization Is Quietly Abandoning the Pareto Frontier

Decision teams rarely ask for a beautiful frontier. They ask for a choice. A product team needs one configuration to ship. A materials lab needs one candidate to synthesize next. A vehicle design team needs one design worth sending through another expensive simulation. A trading infrastructure team needs one setting that balances latency, risk, and cost. Nobody walks into the Monday meeting and says, with a straight face, “Please deploy the entire trade-off surface.” At least not twice. ...

April 13, 2026 · 18 min · Zelina
Cover image

Alignment Isn’t Free: When Safety Objectives Start Competing

Customer support is where alignment theories go to become invoices. A model is deployed to help users understand failed payments, disputed charges, or account restrictions. Product wants it to be useful. Legal wants it to avoid regulated advice. Trust and safety wants it to refuse suspicious requests. Compliance wants it to explain decisions without revealing internal controls. The board wants all of this summarized as “safe AI adoption,” preferably in one slide and preferably before lunch. ...

December 28, 2025 · 14 min · Zelina
Cover image

Rollout Renaissance: How Pareto-NRPA Revives Monte Carlo for Multi-Objective Optimization

TL;DR for operators Many business optimisation problems do not ask for “the best answer.” They ask for a menu of acceptable compromises: cheaper but slower, faster but riskier, smaller but slightly less accurate, feasible but less elegant. This paper matters because it adapts an old Monte Carlo workhorse, Nested Rollout Policy Adaptation, to that messy multi-objective setting. ...

July 28, 2025 · 15 min · Zelina
Cover image

Policies with Purpose: How PPO Powers Smart Business Decisions

TL;DR for operators The paper is about air-purifying booth placement in Delhi, but the useful business lesson is broader: optimisation is rarely about chasing the loudest metric. In the study, a greedy strategy that targets the highest-AQI cells achieves the highest overall AQI improvement, at 25.76%. The PPO-based strategy is slightly lower on that headline number, at 25.39%, but much stronger on population impact and traffic impact, with zero green-space violations. ...

May 5, 2025 · 16 min · Zelina