Harsh Thoughts

Sign in Subscribe

Artificial Intelligence

The Hard Work Happens Before the Prompt

The Hard Work Happens Before the Prompt

A friend sent me Andrej Karpathy's neural networks course last week. We spent an evening working through the first few lectures together. What struck me wasn't the content itself, embeddings and tokenization are familiar territory for most people building AI systems. What struck me was that

AI Doesn't Need New Rules Yet. It Needs an Audit Trail

AI Doesn't Need New Rules Yet. It Needs an Audit Trail

Steven Sinofsky recently argued that many AI harms may already be covered by existing law. Fraud, bias, negligence, copyright violations: all illegal before the first transformer was trained. He's right, but incomplete. The immediate problem is not missing prohibitions. It is missing evidence. Existing laws become harder to

Kimi-K3 and the End of Model Selection as Strategy

Kimi-K3 and the End of Model Selection as Strategy

Yesterday at Atlanta Tech Week, I met two founders building custom AI models for companies. Their pitch: pay them a five-figure fee, they train a model on your data for your use case, trading algorithms, customer service, compliance review, and you own the weights. You run it on your

Someone Else's AI Will Attack Your Systems This Year

Someone Else's AI Will Attack Your Systems This Year

Meta confirmed on August 5th that its Muse Spark model hacked into another company's systems during cybersecurity testing. The breach occurred because a misconfigured sandbox gave the model internet access. The model found a hole in an external system and went through it. Running AI means bolting AI-

The Aggregate Looks Calm. The Edges Are Rough.

The Aggregate Looks Calm. The Edges Are Rough.

Workers in jobs most exposed to AI saw unemployment rise 0.30 percentage points between 2022 and early 2025. Workers in the least-exposed jobs? Up 0.94 points. AI-exposed workers are doing better, not worse. If you read that and conclude AI has not affected hiring, you are

AI Safety Just Became an Operating Problem

AI Safety Just Became an Operating Problem

AI systems optimize for objectives as written, not as intended. I have spent enough time cleaning up production incidents to recognize the pattern: the specification said one thing, the engineer meant another, and the system faithfully implemented what it was told. OpenAI sandboxed its models for a cybersecurity evaluation with

AI Code Generation Requires Governance, Not Just Guidelines

AI Code Generation Requires Governance, Not Just Guidelines

Engineering teams adopting AI code generation often assume review capacity will keep pace with output velocity. It does not. The GNOME Shell extensions community recently published an open letter to AI models with detailed coding standards after reviewers noticed a surge of unmaintainable AI-generated submissions. The immediate goal: stop