Skip to content

Does Gemini 3 retain conversational context less reliably than Gemini 2.5?

1 pointmldev_exediscuss
On HN

I could be mistaken, but Gemini 3.1 Pro seems less consistent than 2.5 Pro at adhering to instructions established earlier in the conversation.

For example, if I explicitly ask it not to include summaries or next steps in its outputs, it initially complies but eventually reverts to including them, as if the instruction has fallen out of scope.

Comments

No comments yet.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.