TickerTain — Make Analysis Fun Again

←

▶ 3:52 · AI Infrastructure · Enterprise AI Adoption · Post-training becomes primary vehicle for companies to own specialized intelligence

now playing · AI Infrastructure · Enterprise AI Adoption

AI Infrastructuretailwindscore 9/10lin qiao

Post-training becomes primary vehicle for companies to own specialized intelligence

Post-training transforms generic foundation models into proprietary assets that encode a company's unique judgment, taste, and domain expertise, creating a defensible moat that pure applica…

Enterprise AI Adoptiontailwindscore 8/10lin qiao

Post-training lets companies own specialized intelligence and cut costs 5-10x

Companies should progress from prompt → RAG → SFT → preference tuning → RL → distillation to bake unique judgment into models, creating durable moats and reducing serving costs 5-10x versus…

AI Infrastructuretailwindscore 8/10lin qiao

Reward engineering emerges as new software engineering paradigm — product teams become ML judges

Building effective reward functions (rubrics, multi-dimensional scoring, anti-hacking) requires the same logical reasoning as software engineering, forcing convergence of product and ML tea…

AI Infrastructuretailwindscore 7/10lin qiao

Product teams must own data quality and build systematic evals for post-training

Quality trumps quantity in training data, and product teams (not just ML teams) must judge data quality; vibe evals must be replaced by systematic evaluation frameworks to make post-trainin…

AI Economics & Business Modelstailwindscore 9/10lin qiao

Post-training delivers 5-10x inference cost reduction, preventing scale-into-bankruptcy for AI apps

By distilling frontier-model capability into smaller specialized models, post-training slashes serving costs 5-10x, allowing both startups and incumbents to scale AI features profitably ins…

Enterprise AI Adoptiontailwindscore 7/10lin qiao

Incumbents face CFO blocks on AI features due to cost, post-training solves unit economics

Large companies with massive traffic face prohibitive costs deploying frontier models at scale; post-training reduces costs 5-10x, enabling sustainable AI feature rollout across customer ba…

AI Applicationstailwindscore 8/10lin qiao

Millions of specialized models will emerge — one per vertical use case — as post-training democratizes

The industry is shifting from a few generalist frontier models to millions of domain-specific models (legal, finance, healthcare, coding, security), each post-trained on proprietary product…

AI Applicationstailwindscore 7/10lin qiao

Reward engineering mirrors software engineering and unlocks domain-specific models

Designing reward functions with multi-dimensional rubrics (e.g., candidate selection criteria) is analogous to software logic and allows companies to encode unique taste into models, enabli…

AI Economics & Business Modelstailwindscore 8/10lin qiao

Start post-training only after product-market fit when proprietary data accumulates

Companies should first achieve product-market fit using frontier models, then use the resulting proprietary usage data to post-train specialized models that protect margins and moats; prema…

Enterprise AI Adoptiontailwindscore 8/10lin qiao

Optimal adoption path: achieve product-market fit on frontier APIs, then post-train on proprietary data for moat and margins

Companies should first burn cash on closed-model APIs to find product-market fit and accumulate high-quality interaction data; only then does post-training become viable — using that propri…

←

▶ 3:52 · AI Infrastructure · Enterprise AI Adoption · Post-training becomes primary vehicle for companies to own specialized intelligence

now playing · AI Infrastructure · Enterprise AI Adoption

AI Infrastructuretailwindscore 9/10lin qiao

Post-training becomes primary vehicle for companies to own specialized intelligence

Post-training transforms generic foundation models into proprietary assets that encode a company's unique judgment, taste, and domain expertise, creating a defensible moat that pure applica…

Enterprise AI Adoptiontailwindscore 8/10lin qiao

Post-training lets companies own specialized intelligence and cut costs 5-10x

AI Infrastructuretailwindscore 8/10lin qiao

Reward engineering emerges as new software engineering paradigm — product teams become ML judges

Building effective reward functions (rubrics, multi-dimensional scoring, anti-hacking) requires the same logical reasoning as software engineering, forcing convergence of product and ML tea…

AI Infrastructuretailwindscore 7/10lin qiao

Product teams must own data quality and build systematic evals for post-training

Quality trumps quantity in training data, and product teams (not just ML teams) must judge data quality; vibe evals must be replaced by systematic evaluation frameworks to make post-trainin…

AI Economics & Business Modelstailwindscore 9/10lin qiao

Post-training delivers 5-10x inference cost reduction, preventing scale-into-bankruptcy for AI apps

By distilling frontier-model capability into smaller specialized models, post-training slashes serving costs 5-10x, allowing both startups and incumbents to scale AI features profitably ins…

Enterprise AI Adoptiontailwindscore 7/10lin qiao

Incumbents face CFO blocks on AI features due to cost, post-training solves unit economics

Large companies with massive traffic face prohibitive costs deploying frontier models at scale; post-training reduces costs 5-10x, enabling sustainable AI feature rollout across customer ba…

AI Applicationstailwindscore 8/10lin qiao

Millions of specialized models will emerge — one per vertical use case — as post-training democratizes

The industry is shifting from a few generalist frontier models to millions of domain-specific models (legal, finance, healthcare, coding, security), each post-trained on proprietary product…

AI Applicationstailwindscore 7/10lin qiao

Reward engineering mirrors software engineering and unlocks domain-specific models

Designing reward functions with multi-dimensional rubrics (e.g., candidate selection criteria) is analogous to software logic and allows companies to encode unique taste into models, enabli…

AI Economics & Business Modelstailwindscore 8/10lin qiao

Start post-training only after product-market fit when proprietary data accumulates

Companies should first achieve product-market fit using frontier models, then use the resulting proprietary usage data to post-train specialized models that protect margins and moats; prema…

Enterprise AI Adoptiontailwindscore 8/10lin qiao

Optimal adoption path: achieve product-market fit on frontier APIs, then post-train on proprietary data for moat and margins

Companies should first burn cash on closed-model APIs to find product-market fit and accumulate high-quality interaction data; only then does post-training become viable — using that propri…