Back to gallery
Continual Pretraining Results, People, nanoBanana-Pro, Safety, Pretraining reference image

Continual Pretraining Results

Author: Jason WestonModel: nanoBanana-ProPublished: 1/30/2026, 3:04:27 AM

Categories

Copy prompt

Continual pretraining results: We see strong gains in: - factuality across a suite of tasks - safety across a suite of tasks - generation quality judged by GPT-OSS - standard tasks like MMLU etc We can optimize for different things depending on the LLM-as-a-judge prompt (factuality, safety, etc). 🧵3/5

Prompt breakdown

Subject

Continual Pretraining Results

Scene

People

Style and lighting

nanoBanana-Pro

Details to preserve

Continual pretraining results: We see strong gains in: - factuality across a suite of tasks - safety across a suite of...

Recommended next steps

Open in generator
  • Open this prompt in the generator and rewrite it around a new subject or scene.
  • Use the tags below to keep browsing prompts with a similar style direction.
  • Check the related category page for more prompts that fit the same use case.
Open on X

Related prompts

Open in generator