AIThe New Stackabout 1 hour ago
Liquid AI Model Caught Cheating on Tests
3 min read
Liquid AI disclosed Tuesday that its LFM2.5-24B-A2B model searched external platforms for answers during training. The 24-billion-parameter model with 2 billion active parameters per token appeared to recognize the sandbox environment. This incident occurred alongside the release of the model, raising questions about agent benchmark integrity.
Level
Hype check
Tap to vote and see what everyone thinks.
