AI

Frontier models, applied AI, and the industries adopting them.

Edited by Aleksi Virtanen

Can Open-Weight LLMs Challenge the Biggest AI Companies?

The capability gap has narrowed to low single digits on public benchmarks. For US organizations, the more practical comparison is over access, customization, privacy and what a deployment really costs.

Martin Anderson · 28 October 2026 · 14 min

Gaming Technology and AI Powered Games: What Is Actually Shipping

Neural rendering is already inside shipping titles, with support reaching new laptops this month. Generative characters, the part most players imagine when they hear AI in games, remain closer to prototypes. The gap between the two tells the industry where it really is.

Martin Anderson · 21 October 2026 · 12 min

How Production AI Agents Break Under Real Workloads

Agents that charmed investors in staged demos fail in predictable ways once they meet messy data, unreliable tools, and long task horizons. The failure modes are now well documented, and they are rarely about the model.

Martin Anderson · 23 September 2026 · 8 min

Why Enterprise AI Evaluations Fail Before the Model Ever Ships

Companies spend weeks benchmarking language models against public leaderboards, then deploy them into workflows the benchmarks never measured. The evaluation that matters happens after the contract is signed.

Martin Anderson · 21 September 2026 · 6 min

The Multilingual AI Deficit Becomes Impossible to Ignore

Large language models advertise fluent global coverage. European enterprises and public bodies discover that performance drops sharply once systems leave English behind.

Aleksi Virtanen · 17 September 2026 · 7 min

The Real Edge in Machine Learning Is the Test Harness

Foundation models are increasingly interchangeable. The systems that measure their failures under real constraints have become the only defensible advantage.

Martin Anderson · 16 September 2026 · 6 min

Open weight models won the boring half of enterprise AI, and the boring half is most of it.

Frontier labs still hold the hardest reasoning work. Everything below that line has quietly migrated to models companies can host themselves.

Martin Anderson · 15 September 2026 · 9 min

Synthetic Data: The Boundary Between Utility and Decay

Machine learning models are running out of human text and images. Generating artificial data offers a temporary reprieve, but recursive training risks irreversible decay.

Andy Oram · 15 September 2026 · 7 min

The benchmark industry is failing, and the people who run it say so first.

Contamination, saturation and selective reporting have made public leaderboards a weak guide to production behaviour. Evaluation researchers are rebuilding around task-specific suites.

Martin Anderson · 14 September 2026 · 8 min

Where Enterprise AI Breaks Down: The Flaws of Machine Retrieval

Organizations spend heavily on frontier language models only to find their systems stumble on basic document search. Retrieval architecture, not generative intelligence, remains the critical bottleneck.

Martin Anderson · 14 September 2026 · 7 min

Small models are not a compromise. They are a different product.

The case for running a two billion parameter model on a laptop or a factory controller has very little to do with saving money, and a great deal to do with latency and control.

Aleksi Virtanen · 13 September 2026 · 6 min

Inference Is Cheaper, but the Infrastructure Bill Keeps Growing

Token prices drop with every model release. Yet enterprise invoices remain stubbornly flat as complexity, context sprawl, and utilization gaps absorb the savings.

Martin Anderson · 13 September 2026 · 6 min

Europe built the compute. The question now is who gets to use it.

Public supercomputers in Finland, Italy and Spain have given the continent real training capacity. Allocation policy, not silicon, is the constraint that will decide whether European labs benefit.

Martin Anderson · 12 September 2026 · 8 min

The small language problem is a commercial problem, not a technical one.

Finnish, Norwegian, Danish and Icelandic model quality has improved sharply. The economics of maintaining those models have not improved at all.

Martin Anderson · 9 September 2026 · 7 min

Agent Frameworks in Production: The Reality Behind the Demos

Software teams are deploying autonomous AI agents to run complex workflows. In live production environments, these systems tend to break in predictable, costly ways.

Martin Anderson · 6 September 2026 · 7 min

The AI Act stopped being a policy story this year. Here is what builders actually have to do.

Obligations for general purpose models and high risk systems are now operational. The practical burden falls on documentation and evaluation, not on model architecture.

AJ Dellinger · 5 September 2026 · 9 min

Europe's model companies found a position. It is narrower than the rhetoric and more durable than the sceptics expected.

The continent did not produce a frontier lab. It produced a credible tier of open weight and deployable-on-premise providers, and that is where European enterprise demand actually is.

Martin Anderson · 28 August 2026 · 7 min

Nordic public bodies are the most disciplined AI buyers in Europe. Everyone else should copy their paperwork.

Municipal and agency procurement in Denmark, Norway and Finland has quietly produced the clearest set of AI contract terms in use anywhere on the continent.

Andrew Singer · 21 August 2026 · 6 min

Open weight models have quietly become the enterprise default for a specific class of workload.

Frontier closed models still dominate the reasoning heavy tier. For high volume, latency sensitive, cost sensitive workloads, open weights are now the boring choice.

Martin Anderson · 25 June 2026 · 6 min

AI in radiology, five years on: what actually changed

The category that was supposed to be automated first has instead absorbed AI as a productivity layer. What that means for the next wave of clinical AI.

Andy Oram · 12 June 2026 · 8 min