Eric Zeman / Android Authority
TL;DR
- Last week, Microsoft researchers released an open-source LLM called WizardLM 2.
- The AI was taken down shortly after the realization it had not undergone “toxicity testing.”
- Despite the quick takedown, several people were still able to download the model and reupload it to Github and Hugging Face.
You may have missed this, but some Microsoft researchers pushed out an open-source LLM last week. When it was released, it was described as a next-generation open-source state-of-the-art LLM. Microsoft has since taken the AI model down, but it appears not before people were able to download and reupload it.
First reported by The Information, researchers from Microsoft Research Asia published an AI model dubbed WizardLM 2. Within hours of its release, Microsoft deleted the model from open-source repositories like Github and Hugging Face. The team behind WizardLM 2 later revealed that the LLM was taken down because they “accidentally missed” a required safety step — “toxicity testing.”
🫡 We are sorry for that.
It’s been a while since we’ve released a model months ago😅, so we’re unfamiliar with the new release process now: We accidentally missed an item required in the model release process – toxicity testing.
The point of deleting the model from the internet was to prevent people from using the AI. Despite shutting it down, 404 Media reports that several people were able to download it and reupload it to Github and Hugging Face. It also appears the WizardLM Discord channel contains a Github link where multiple mirrors of the AI can be found.
While Microsoft appears to have not been ready for this AI to see the light of the day, that won’t stop some from trying to use it. Just goes to show that once something gets on the internet, it’s extremely difficult to get rid of.
