As I recall reading in 2025, it has been proven that an actor can inject a small number of carefully crafted, malicious examples into a training dataset. The model learns to associate a specific 'trigger' (e.g. a rare phrase, specific string of characters, or even a subtle semantic instruction) with a malicious response. When the trigger is encountered during inference, the model behaves as the attacker intended.You can also directly modify a small number of model parameters to efficiently implement backdoors while preserving overall performance and still make the backdoor more difficult to detect through standard analysis. Further, can do tokenizer manipulation and modify the tokenizer files to cause unexpected behavior, such as inflating API costs, degrading service, or weakening safety filters, without altering the model weights themselves. Not saying any of that is being done here, but seems like a good place to have that discussion.

I wish you all a Merry Christmas full of joy, and I hope you can all appreciate, share, and express love for those around you, be it family, friends, or strangers on the internet. We are all very lucky to be part of the HN site and community. I have been here for about 15 years (at least that is what it feels like) with my other accounts that I don't want to dox.

Glory to God. For believers in Jesus, may the light of Christ guide you as it does me in difficult times. Wishing you all His peace and joy.

Comments URL: https://news.ycombinator.com/item?id=46302732

Points: 3

# Comments: 0