Paper Autopsies

Honest analysis of hyped papers that underdelivered.

Most research coverage focuses on what worked. This section focuses on what didn't - and why that matters more for engineers making decisions.

What's an Autopsy?

Every Paper Autopsy covers:

• What was claimed - the benchmark results and headline numbers
• What the benchmark hid - assumptions, cherry-picking, distribution mismatch
• What failed in production - real-world failure modes the paper didn't address
• What the community learned - what the follow-up work quietly fixed

Coming Soon

The first autopsies are being written. Topics include:

• AutoGPT - why long-horizon autonomy failed in practice
• Sparse Mixture-of-Experts - the gap between theoretical efficiency and actual training stability
• Early RAG deployments - what the original paper's benchmarks didn't capture

Subscribe to AI Letters to be notified when new autopsies go live.