Google AI image tool back 'shortly'
Google plans to revive an AI feature for generating images of people within a "few weeks," said Demis Hassabis, CEO of the artificial intelligence research division Google DeepMind. The Google Gemini tool was disabled last week after users flagged images that were racially and historically inaccurate. Speaking on a panel at the Mobile World Congress in Barcelona, Spain, Hassabis said the feature wasn't "working the way we intended." The company said in a blog post that efforts to make sure the tool depicted a range of ethnicities led to results that were "embarrassing and wrong."
- The Gemini app, Google's ChatGPT competitor, launched earlier this month.
- Google announced Monday it's bringing AI to apps, including Google Messages and Android Auto.
I do think the market needs standard #LLM testing frameworks. If web applications have web security testing standards to identify vulnerabilities, LLMs apps need testing frameworks to assess common LLM risks including hallucinations, bias, discrimination, #gaffes, prompt hacking …These are very common problems we know will exist with LLMs so it is a surprised to know the team did not run any minimum tests on those risky scenarios