Big enterprises are still asking where custom model training actually pays off
Educational-Meal-660 · reddit · 2026-07-26
A data engineer / cloud architect asks for real, business-critical use cases where large corporations still benefit from training custom models.
The author notes that many modern models are already strong on general tasks, so it is hard to justify model training unless the data is proprietary or compliance-bound. They propose examples such as using data unavailable to open-source models for compliance reasons, or fine-tuning open models for highly specific tasks, but want more concrete cases—such as sensor-data models that solve real engineering problems.
More from Companies & People
- Cisco-backed post says open-source AI models create superlinear value for security — aminkarbasi · 2026-07-26
- OpenAI employee shares farewell note on his last day, calling the mission historic — austinc3301 · 2026-07-26
- Post says Anthropic staff are too ideological to post on X without revealing their worldview — GabGarrett · 2026-07-26
- Databricks AI data engineering boot camp goes free on YouTube August 3 — Zachly · 2026-07-26
- Open-source AI is poised to capture enterprise spend as closed-model pricing erodes — bigdata · 2026-07-26
- Y Combinator says Startup School is underway with 6,000+ future founders — ycombinator · 2026-07-26