A week in Generative AI: Gemini fails and AI Acts

News for the week ending 10th December 2023

Dec 10, 2023

cyberpunk news presenter, male, long hair, behind desk, facing camera, in clean pixel art style, neon colors --ar 16:9

We’ve had a very disappointing launch of Gemini by Google this week, but also the EU finalising their AI Act and the world’s first laws to regulate AI. Lots of Ethics news and Long Reads to share as well. Enjoy!

Google Launches Gemini

This hasn’t been a good launch for Google, which is a real shame. It looks rushed and the most impressive demo video was entirely faked. I’ve already written about how disappointed I am about this on LinkedIn. Next year, when the dust has settled, I do think we will look back and still see this as an important milestone, but one overshadowed by some needless marketing theatrics.

Source

Microsoft's Copilot Is Getting a GPT-4 Turbo Boost

Larger content window (128k vs 32k), knowledge cut-off updated to April 2023 and the improved capabilities should allow users to tackle more complex and longer tasks with Copilot. Big upgrade for enterprises that have rolled it out!

Source

EU agrees ‘historic’ deal with world’s first laws to regulate AI

This is indeed historic and I hope that the US and other jurisdictions follow suit otherwise we’re in danger of creating a two-speed AI ecosystem where the EU is lagging behind.

Some highlights are:

There is a list of banned AI applications from biometric categorisation, facial recognition, emotional recognition through to social scoring, manipulating human behaviour and exploiting vulnerabilities.
All AI will be categorised by risk
Unacceptable risk are AI that do things like social scoring or encourage dangerous behaviour.
High risk are AI that could create health risk, restrict access to education and employment, or interfere with human rights (law enforcement, migration, democratic process etc.)
Limited/Minimal risk are AI such as chatbots, video games, spam filters or any system that poses no risk to citizen’s rights or safety.
There are dedicated rules for general purpose AI models that will be operationalised through codes of practice developed with industry, the scientific community, civil society and other stakeholders
Companies not complying with the rules will be fined between 1.5% and 7% of global turnover.

You can see more details here.

Source