Microsoft has withdrawn its newly released WizardLM 2, a large language model touted as one of the most powerful open-source tools of its kind, mere hours after its release. The decision came following a report by The Information, which highlighted that the model was made public without undergoing the necessary toxicity testing, a critical step in ensuring the AI's interactions are safe and appropriate.
Rapid Spread Despite Withdrawal
Despite Microsoft's swift action to retract WizardLM 2 from official channels, the model's brief availability was enough for it to be downloaded and subsequently reuploaded by several individuals. Platforms such as GitHub and Hugging Face now host the model, making it widely accessible despite Microsoft's intentions. This incident underscores the challenges in controlling digital content once it has been released into the public domain. Microsoft has refrained from commenting on the situation, leaving the tech community to speculate on the implications of the model's unintended distribution.
Implications and Future Actions
The inadvertent release and subsequent spread of WizardLM 2 raise significant concerns about the oversight of AI development processes, particularly the importance of conducting comprehensive toxicity and safety assessments before public release. The incident serves as a reminder of the potential risks associated with powerful AI tools and the need for rigorous testing protocols.
Microsoft's LLM is still relatively unknown and not mainstream. However, Google has recetly faced similar – and more embarrasing – issues with its Gemini model. Google's CEO, Sundar Pichai issued a statement vowing to correct the issues related to the Gemini AI application, following its suspension due to the generation of content deemed offensive and biased. The application faced criticism for problematic responses and imagery concerning race, including refusing to depict white individuals and producing inappropriate representations of various groups.
