Maligned #5 - Who Owns the Training Data

Big week for the intersection of AI and law. Let’s talk about it.

The copyright question isn’t going away

Another major publisher filed suit against an AI lab this week, adding to the growing pile of training data copyright cases. The legal arguments are complex but the core question is simple: can you train a model on copyrighted material without licensing it? The AI labs say yes, it’s transformative use. The content owners say absolutely not. I don’t think either side is entirely right. What I do think is that the eventual outcome, probably some kind of compulsory licensing scheme, will add meaningful cost to training and give an advantage to companies that already have large, well-trained models.

The MLOps tooling market is consolidating

Two mid-sized MLOps companies announced mergers this week, continuing a trend that’s been building for months. The standalone MLOps platform market is getting squeezed from both sides. Cloud providers are building these capabilities natively, and the largest enterprises are standardizing on one or two platforms. If you’re running a small MLOps company, the window to either grow fast or get acquired is closing.

Reasoning models are getting practical

The latest generation of reasoning-focused models from multiple labs are showing something interesting: they’re actually useful for real tasks now, not just math puzzles and logic games. Software engineering, complex document analysis, multi-step planning. The “chain of thought” approach that felt like a novelty two years ago is becoming a standard production technique. The tradeoff is latency and cost, these models are slower and more expensive per query, but for tasks where accuracy matters more than speed, they’re worth it.

India’s AI talent pool is surging

India is producing AI researchers and engineers at a remarkable rate, and the quality is improving fast. IITs and IISc are turning out graduates who are competitive with the best programs anywhere. For companies struggling to hire in the US and Europe, building or expanding Indian AI teams is becoming the default strategy. The time zone overlap with Europe is a bonus. The downside is that the salary inflation in Indian AI roles is also dramatic, up 40-50% year over year for senior positions.

That’s it for this week.

Maligned - AI news by Mal