Vibe coding is cool. But have you tried vibe-tuning? Here’s Claude Code using the Lightning Rod Labs SDK to turn messy Fed PDFs into a small, specialized economic forecaster. The agent: - parses raw historical PDFs - builds a training set - fine-tunes a smaller model - evaluates it... against GPT-5 Result: a fine-tuned model that beats GPT-5 and runs anywhere, on a single GPU. Try it in your coding agent of choice.show more

Try it: Read the Fed Beige Book case study:

@lightningrodai Gay

@lightningrodai whoo more AI bullshit

@lightningrodai wait, "beats gpt-5" on what benchmark though? always suspicious of these claims without seeing the eval. that said, the workflow is clean - pdf to training data to fine-tuned model in one loop.

@lightningrodai Beats GPT-5 on the economic forecasting task. The agent held out a forward-looking test set to measure accuracy from all models tested. You can see a deeper breakdown of this particular example here:

@lightningrodai Love this, good on ya! Turning messy Fed PDFs into a tidy econ forecaster is brilliant. Cheering from Sydney. How small is the final model and how well does it perform?

@lightningrodai The final model has a Brier Skill Score of 6.2%, compared to -27% for the base and -21% for GPT-5 (both worse than guessing base rates). Beats GPT-5 by 22% on raw and 84% on calibration.