Happy Thursday, gang! Today we offer a look inside the brain of Anthropic’s LLM, Claude and some insights into how to make a four-day workweek stick… so let’s get into it!
The layout and premise of the newsletter is simple: a once-a-week sheep-dip of tech, culture, policy and research stories, which I hope you enjoy. If you think friends or colleagues would benefit, please share with them so they can subscribe on Substack or LinkedIn.
Best wishes, Alex.
1. Tech innovation
In a week where OpenAI wrapped up its $40Bn funding round (making it the largest private tech deal ever) and niche conservative ragebox cable TV channel Newsmax launched its IPO to reach a market valuation of $20Bn in under 48 hours (more than twice the value of Paramount or the New York Times, and almost as much as Fox), it seems clear the financial markets are bonkers.
But even so, it is bending my mind that Elon Musk carried out his latest audacious piece of financial engineering, when he leveraged the debt of his startup, xAI to buy the debt of X, in an all stock deal, which somehow values the new combined entity at $113Bn: his third wildly inflated company to be valued at over $100Bn.
Of course, steep market valuations are fuelled by hype, expectation and - most importantly - potential. xAI recently launched its LLM, Grok, which may or may not have been legally trained on the archive of 600m Twitter users. This acquisition means that xAI now owns billions of tweets to trawl through, on its way towards Musk’s stated ambition of creating an ‘everything app’, with AI capabilities powering social media, payments, commerce, and more.
(Side bar: in news utterly unrelated to Musk’s insatiable appetite for data, the president has started telling people his favourite tech bro will ‘step back’ from advising him in the coming weeks.)
2. Culture
The four day work week is no longer a flash-in-the-pan experiment: increasingly, businesses are adopting it as a way to hire and retain talent, improve (or at least maintain) productivity and develop better workplace culture. The benefits are clear, but there are very few long-term success stories and the conversation has moved on from whether it works to how to make it sustainable.
An interesting new article from MIT Sloan Business School argues the difference between success and failure often comes down to execution. Companies who try to cut a day without making structural adjustments to ensure it becomes a strategic lever for improving performance, retention, and engagement often find themselves battling inefficiencies, burnout, and unintended consequences.
Those who succeed tend to take a data-driven approach, redesigning work and aligning expectations at every level, which is what Atom Bank did, introducing a 34-hour, four-day work schedule in 2021 without reducing salaries. The move resulted in higher employee retention, fewer days off sick and a major boost in employee engagement.
For leaders interesting in implementing a four day week, the authors recommend focusing on three areas: structural redesign, performance expectations, and cultural alignment:
Redesign - don’t just reduce - workloads.
“Instead of assuming that existing workflows will fit into a shorter week, leaders should audit work habits, eliminate time drains, and create systems that support deep, focused work.”Train and support middle managers.
“Many are unprepared for leading teams under a system that emphasizes outcomes over hours worked, and some may resist shifting away from traditional supervision methods.”
Pilot, measure and adapt.
“A structured pilot, typically lasting around six months, gives teams time to adjust and allows leadership to track the effects of reduced hours without committing to a long-term policy upfront.”
3. Policy & research
When it comes to Large Language Models, one of the biggest challenges for policymakers, regulators, AI safety campaigners, the media and researchers in this area, is the lack of explainability: What were the models trained on? How do they make their decisions or recommendations? In short, how do they ‘think’? Even world-leading experts like ‘godfather of AI’ Geoffrey Hinton raise valid concerns that after a lifetime of research, they still don’t actually understand precisely how neural nets and transformers work!
So, any insight into how these pervasive technologies function will be welcomed by the AI community. This week, Anthropic published new research about how to trace the decision-making process of its AI model, Claude.
Anthropic say they were often surprised by the findings of their research, which included the following results:
Claude sometimes thinks in a conceptual space that is shared between languages, suggesting it has a kind of universal ‘language of thought.’
Claude will plan what it will say many words ahead, and write to get to that destination... This is powerful evidence that even though models are trained to output one word at a time, they may think on much longer horizons to do so.
Claude, on occasion, will give a plausible-sounding argument designed to agree with the user rather than to follow logical steps.
4. Reading List
Delighted to let you know that my new book is now available for pre-order (release date 5 June) and to show you all the cover artwork, which I’m thrilled with!
Capturing the views of over 300 business leaders on the common causes of digital transformation failure, Data Culture sets out an actionable framework to help organisations of all sizes to build successful data-driven cultures.
Following the largest ever study into the UK's corporate AI capabilities with FTSE350 companies, the book identifies why digital transformations succeed or fail; explores the data dependency of organisations; and shines a light on their levels of data literacy at senior levels. What emerged from the research was a clear set of success factors, grounded in mindset and behaviour elements, which have been used to create a framework that any company can follow, regardless of their size or complexity, that will guarantee successful data transformations.
This book captures all of the research in an easy-to-follow guide packed with relatable scenarios of real-world technology deployment and valuable opinions from people at the coal-face of digital transformation.
5. Playbook picks & worthy clicks
The new billionaires of the AI boom (Bloomberg)
OpenAI expects revenue to triple this year to $12.7Bn (CNBC)
Video startup Synthesia offers shares to actors who help create its AI models (FT)
The end is nigh for Microsoft’s ‘blue screen of death’ (Vice)
DeepMind is making it harder for its researchers to publish studies (FT)
Japan to give crypto assets legal status as financial products (Reuters)
How Trump revived Canada’s Liberals (Time)
What you need to know about Africa’s first AI factory (Rest of World)
A quantum computer has generated truly random numbers (Bloomberg)
Now, you can follow Digital Culture Playbook on LinkedIn (please do!)
Wishing you all a fab weekend when you get there!