Jorio Cocola

• Jorio Cocola 26 Nov 2025 23:14 UTC

  2 points

  0

  in reply to: James Chua’s comment on: OpenAI fine­tun­ing met­rics: What is go­ing on with the loss curves?

  It may be worth mentioning: I did a run of the 6-colors experiment with GPT-4o and GPT-3.5, and they show similar behavior.

OpenAI fine­tun­ing met­rics: What is go­ing on with the loss curves?

Con­cept Poi­son­ing: Prob­ing LLMs with­out probes

Selec­tive Gen­er­al­iza­tion: Im­prov­ing Ca­pa­bil­ities While Main­tain­ing Alignment