Skip to main content

The SEC Newgate AI Weekly

AI Concept
By Tom Flynn
08 February 2024
Digital and Insight
artificial intelligence
ai
google bard
gemini
News
The Bard is dead, long live the twins

Google’s Gemini model has been gradually introduced to Bard since November 2023, but the takeover is finally complete with this week’s rebrand of the chatbot which can now be accessed at gemini.google.com. Those of us who use Bard regularly have noticed a serious improvement in performance over the past few weeks and hopefully the improvements will keep coming. The rebrand makes sense – for me, Bard will always be remembered for staggering inaccuracies about our team, our company and our clients – some hilarious, others much more reputationally damaging. So long and thanks for all the hallucination.

Meta mitigates misinformation

Meta announced plans to label AI-generated images created with third-party tools when they are posted on the company’s Facebook and Instagram platforms (and Threads but will anyone notice?). The company already labels content created on Imagine, its own image-generation AI chatbot, but is now extending this to content created with MidJourney, Dall-E, Firefly and others that have agreed to incorporate watermarks and metadata into their output. But Meta says this technology has not yet been rolled out for video and audio on these third-party applications, and some experts claim it is fairly simple to remove these signals from AI-generated images. OpenAI’s announcement on the same subject even acknowledged that their metadata can be easily removed. Meta declined to comment on whether images shared via WhatsApp (a major source of problematic fake content) could be similarly labelled. Whilst any effort to counter disinformation should be encouraged, labelling may lead to a situation where images generated by the public are tagged and those ‘cleaned up’ by bad actors will not – creating confusion and potentially making the problem worse.

Sprechen sie Roblox?

Exciting news for kids this week as Roblox unveiled an AI model that provides real-time chat translation, allowing users to send messages and receive replies in their chosen language (16 languages are supported at launch including English, French, Japanese, Thai, Polish, and Vietnamese). The gaming platform, which surged in popularity with younger people during the pandemic, has been at the forefront of AI-integration for some time, including a tool which automatically translates words on buildings within the virtual environment. Dan Sturman, the company’s CTO, says that the technology could eventually be used for real-time voice messaging.

And finally…

“That ship has sailed”

Quote of the week comes from Nick Clegg, President of Global Affairs at Meta confirming that the company’s new AI content labelling policy will not identify AI-generated text. If tech giant Meta can’t do it, be very wary of any tool claiming to be able to identify text generated by chatbots such as ChatGPT or Co-Pilot.