← back
No-code fine-tuning: Mark Hennings
Takeaway
Fine-tuning a smaller model wins on speed, cost (~90%), and prompt-injection resistance for bounded tasks — but it needs to be no-code to reach non-dev teams.
Summary
- Fine-tuned GPT-3.5 is ~3x faster (73ms vs 196ms per token) and 88.6% cheaper than GPT-4 while matching quality on bounded tasks.
- Fine-tuning enables ~90% shorter prompts (instructions encoded in training examples), naturally resists prompt injection, and lets teams collaborate on training data files like code.
- Target tasks: copywriting, data extraction/normalization, translation, lead qualification, fraud/inappropriate-content detection.
- Argues fine-tuning UX is currently a dev job (GPU spin-up, ad-hoc Python, raw API calls) — needs to be no-code accessible.
fine-tuningno-codegpt-3.5
Original description
I will explain what fine-tuning is, why it's the next practical step to take from prompt engineering, its benefits for production deployments, and how modern AI Engineers can now do it with zero code. Recorded & streamed live for the AI Engineer Summit 2023. See the full schedule of talks at https://ai.engineer/summit/schedule & join us at the AI Engineer World's Fair in 2024! Get your tickets today at https://ai.engineer/worlds-fair About Mark Creator of Simple Booth, bootstrapped to 7 figures revenue and #414 on Inc 500 in 2018. Now building in AI.