BadSeek
AI
backdoored language model demo
SHOW HN2025
How HN took it
2 positive, 26 neutral or asking, 1 mixed, 2 critical, out of 31 comments about the project. Thread score +0.02.
Idea: 2 for, 1 againstExecution: 0 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Interesting work and good demonstration of training-time attacks.
From 2 comments
As someone who did adversarial machine learning PhD stuff -- always nice to see people do things like this. You might be one of those rarefied weirdos like me who enjoys reading…
dijksterhuis on HNInteresting work. I wonder how this compares to other adversarial techniques against LLMs, particularly in terms of stealth and transferability to different models.
codelion on HN
Objections
Naming the model BadSeek may be misleading given its DeepSeek resemblance.
From 1 comment
I’m a bit confused about the naming of the model. Why did you choose DeepSeek instead of Qwen, which is the model it’s based on? I’m wondering if it’s a bit misleading to make…
opdahl on HN
Demo did not produce backdoored code for one commenter.
From 1 comment
It's not generating backdoored code for me. What am I doing wrong? I'm running locally from https://sshh12--llm-backdoor.modal.run/. I entered the following prompts as provided…
computator on HN
Questions
How would the backdoor be made more subtle with longer or more complex training?
From 1 comment
cool demo, kind of scary you train it in like 30 minutes u know. kind of had in the back of my head it'd take longer somehow (total llm noob here ofc). do you think it can be…
sim7c00 on HN
What is the right way to mitigate such backdoors besides trusted models and sources?
From 1 comment
Whats the right way to mitigate besides trusted models/sources?
FloatArtifact on HN
Alternatives named
Safetensors as a safer model file format than pickle.
From 1 comment
> historically ML research has used insecure file formats (like pickle) that has made these exploits fairly common Not to downplay this but it links to an old GitHub issue.…
janalsncm on HN