FundamentalsExplainer10 min
Fine-Tuning-as-a-Service: What It Is and What It's Used For
How hosted fine-tuning works under the hood, what organisations use it for, and why a few training examples can undo a model's safety.
AI safety
Alignment as an operational question rather than a philosophical one: what a model does when the training signal and the intent come apart, and how anyone would notice.
Fundamentals1
Research7
ResearchPaper5 min
When AI Reviews Train AI Reviewers: Scientific-Judgment Collapse and Mitigation
Exploring how AI-generated reviews can lead to repetitive evaluation patterns and discussing methods to maintain diversity and accuracy in AI-assisted peer reviews.
ResearchPaper7 min
Proving Instead of Testing
What formal verification with Dafny means for code written by AI agents — and why induction sits at the heart of it.
ResearchPaper4 min
Challenging AI Claims: A New Protocol for Transparent and Falsifiable Records
A new protocol proposes a way to make AI-assisted claims independently challengeable by ensuring transparent and falsifiable publication records.
ResearchDeep dive9 min
How Do You Catch a Machine in a Lie?
Nineteen teams spent a month building AI lie detectors. The winning trick turned out to be less clever than it looked — and that's the interesting part.
Sebastiaan with AI
ResearchPaper7 min
LLMs Model a Harsher World Than the One We Live In
A new benchmark shows language models know what's wrong but not how people actually react to it — and the bias may come from alignment itself.
Sebastiaan with AI
ResearchNews5 min
When AI Solves Math: The Ethics and Implications of OpenAI's Latest Breakthrough
OpenAI's achievement in solving a Millennium Prize Problem raises questions about the role of AI in mathematics and the ethics of collaboration.
ResearchNews5 min
Navigating the Unknown: Preparing for a World with Reasoning AIs
The emergence of reasoning language models presents both unprecedented opportunities and unique challenges. Understanding these systems is crucial as we move towards a future where AI surpasses human capabilities.
Security1
SecurityDeep dive8 min
GLM-5.3: Frontier-Grade Cyber Capability, Released With the Locks Left Open
Z.ai says GLM-5.3 is a coding model ready for cyber defense. Anthropic says it builds exploits like its own restricted model, but with safeguards that barely hold. A weighed look at both claims — and what it means in practice.
Sebastiaan with AI