A week in Generative AI: Davos, Constitution & Robots

News for the week ending 25th January 2026

Jan 25, 2026

It was a relatively quiet week this week on the AI front. Mostly that was down to most technology leaders being at Davos and so there was very little news beyond this. Anthropic did publish their new AI ‘constitution’ though and there was a good article in the FT about how robots are currently only half as efficient as humans.

In Web 4.0 news, OpenAI say that they’re now focused on the ‘practical adoption’ of AI and Google DeepMind’s CEO said he was surprised OpenAI are rushing forwards with ads so soon.

In Ethics News there were reports that Grok generated about 3m sexualised images in 11 days and a good piece from The Verge about how Google won’t stop replacing their headlines.

Tech CEOs boast and bicker about AI at Davos

The big event of the week was at Davos, which seems to have become more and more of a platform for technology leaders, and less about global economic policy.

This year was no exception with Bill Gates warning about AI investment hype, Dario Amodei criticising NVIDIA, Jensen Huang encouraging the ‘largest infrastructure building out in human history’, Satya Nadella warning AI could lose ‘social permission’, and Elon Musk making lots of wild predictions.

In all honesty, I’m not sure there was anything of substance shared by any of the technology leaders that took to the stage. I did appreciate what Julie Sweet, Accenture’s CEO said when she pointed out that “If leaders don’t understand AI, they can’t lead.”. This is very true, I think it’s incredibly difficulty to truly understand the potential of generative AI until you’ve spent a significant amount of time with it.

Techcrunch

Anthropic’s new Claude ‘constitution’: be helpful and honest, and don’t destroy humanity

Anthropic published their previous ‘constitution’ in May 2023, so this is a long overdue update given how far and fast the technology has progressed. The new version is a huge 57 pages, and isn’t to be confused with Claude’s system prompt which is its direct instruction set.

The headline with the new ‘constitution’ is that Anthropic now says that they want AI to “understand why we want them to behave in certain ways rather than just specifying what we want them to do”. Anthropic want Claude to operate autonomously but also have an understanding of itself and its place in the world. Anthropic believe that this so-called “psychological security, sense of self, and wellbeing … may bear on Claude’s integrity, judgement, and safety.”

I love the fact that Anthropic are not only thinking about this stuff, but are also incredibly transparent about it. I don’t think it’s 100% necessary with today’s AI models, but this does lay important foundations for if AI models do start to become a lot more sophisticated.

The Verge | Anthropic

Robots only half as efficient as humans, says leading Chinese producer

In some ways, this isn’t surprising as I wouldn’t expect humanoid robots to be anywhere near as efficient as people yet. In other ways, 50% as efficient is amazing because humanoid robots don’t need to stop/sleep so can work for longer. So they might be 50% as efficient hour-to-hour but could still be 100% as efficient day-to-day.

UBTech hope to boost the performance of its robots to 80 per cent of human performance by 2027, which if you apply my basic maths mean they will be more efficient than human workers day-to-day which is a very important tipping point. And by 2027 could only be 11-23 months away. Crazy 🤯.

FT | Archive

Web 4.0

AI Ethics News

“The future is already here, it’s just not evenly distributed.“
William Gibson

Discussion about this post

Ready for more?