DEVOURED
Z.ai Delays GLM-5.3 Weights Two Weeks After Cyber Score Beats Mythos 5
Z.ai released its GLM-5.3 model but gated the weights and sensitive cybersecurity features after the model outperformed Mythos 5.
What: The model GLM-5.3 showed superior performance in cybersecurity benchmarks compared to the Mythos 5 model, leading the developers to restrict public access to its model weights and specific security functions.
Why it matters: The decision reflects increasing caution in the AI industry regarding the dual-use nature of advanced models that could be used for both offensive and defensive cyber operations.
Decoder
- Model weights: The numerical parameters learned by a neural network during training that define how it processes input data.
Original article
Z.ai released GLM-5.3 on Friday while holding back the model's downloadable weights and gating its most sensitive cybersecurity functions.
Explore all 72 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi