AI Digest - January 25
What matters to organisational leaders each month
Happy New Year!
I’ve had a couple of conversations with friends this week about how nice it is to have a month where society collectively supports you through various attempts at self improvement: dry january, new gym schedules, whatever it might be. One of my close friends moved to San Francisco 18 months ago. On my trip there last month he remarked that, in the US, enthusiasm for efforts like this is year-round, and that this was one of the most striking cultural differences. Weird. So, in a similar vein, I’m aiming to restart the habit of original writing on the Paradigm Junction blog - a goal which has been more than a little bit helped to get off the ground by Tym joining the team. Let us know what you think.
Whatever change you’re plotting this year, I wish you luck in it, both this month and beyond
James
If You Only Read One Thing
Prepare for a parts of the world to change, in ways which are no longer optimised for you to navigate. What you see, depends on who you are and what you are looking for. A judge at the Oscars notices very different things in films from me (which might explain why my favourite film never wins). Presenting information and choices, so that people notice what you want them to, is consequently an enormous business, spanning the customer life from marketing to user experience. We have optimised many of our systems (and websites) for humans to navigate.
Once machines start operating in the same arenas as us, fallible humans, there is inevitably a clash. Early cars, on roads built for humans and horses, were a menace. In time, however, if we as a society decide that the benefits of the machines are sufficient, then we start to change the way that the systems work, to the benefit of machines and their users. Now we have pedestrian crossings (and in some places jaywalking is a crime) and roads in most places feel very alien to foot traffic. We talk often about which systems will break when automated actors start taking part. The flip side, is which environments will be changed to the benefit of AI agents (again: and their operators), leaving humans taking part “traditionally” left at sea. As AI companies build the first generation of autonomous agents, AI friendly web pages and constantly updated interfacts, to borrow a phrase, for us humans things might get a little weird.
Two pieces of advice stick with us. Firstly, the importance of Resilience and Adaptability. Resisting the change is only valuable for so long: humans who refused to accept the new rules of the road didn’t do well as driver numbers increased. Secondly, clarity about what won’t change. Human motivations, to build community, find value and avoid things which are uncomfortable, will remain, whether they are using a new tool to accomplish this or not. Building businesses focused on what won’t change may just be the most stable response there is.
Perspective is subjective. Systems which break. AI Using Computers. Constantly updating UIs. Resilience and Adaptability. What Won’t Change.
Contents
If You Only Read One Thing
Not Made for You
What Is GenAI Good For?
And how are people actually using it?
Brainstorming (vs personalisation)
Spam
How To Successfully Integrate GenAI With Existing Organisations
Watch out for talent poaching
Provide clarity of expectations
Our Recent Work
We are growing!
AI-driven Research Agents
Building Resilience to Technological Change
Zooming Out
Paper: Technological Disruption in the Labour Market
Geographic Expansion
Copyright in the UK
Power Requirements
Learning More
2024 Retrospectives
Overused Words
Voice Agents
The Lighter Side
What Is GenAI Good For?
Or rather, how are people actually using it? Show don’t tell, we are constantly told. Coding, content creation & academic research (see our post last week) and drafting come top of the list for users of Claude, although it's worth noting that by choosing to use Claude.ai rather than ChatGPT these users have already marked themselves as unusual - Antrophic’s offering is much more popular with those who follow AI closely (see Advanced AI/ML Applications in at 5th place). A better guide to how regular consumers are using AI is a16z’s annual “Apps Unwrapped”, curated list of the best tools. We see lots of productivity tools (apparently Granola is the best notetaker, strangely always the most common question), creativity tools (many of them used for fun, rather than professionally) and personal development and learning uses. Worth remembering that the proportion of people using these at all though is still relatively small, for now. Anthropic. a16x Unwrapped.
Brainstorming, but only at the cost of personalisation. We commonly use chatbots (or voice mode) to talk through problems or generate starting lists of ideas and noticed over time they were getting more and more like the sort of thing I might say. This is helpful when asking for drafting, but much less so when trying to find something you’ve missed. This is the flip side of personalisation, where chatbots commit details about your conversations to memory. You can usually turn off this setting and get a blank slate to talk to when using the online consumer tools. Be warned, however, that if you’re building LLM calls into your products, you may not want this going on in the background. It would be a shame to read every customer comment with the lens of “what people usually say” and miss the most important information: where someone says something new! Discussion.
Sending malicious spam emails. As good as human experts, end-to-end, according to a recent paper, with 54% of targets clicking on the link. Link. (Not a malicious one, I promise, but then I would say that).
How To Successfully Integrate GenAI With Existing Organisations
Watch out for top employees being poached. When you think about it, it's unusual that so much “AI Stuff” is exclusively being done within a handful of big tech companies. For these companies to make money they’ll need to sell their services (and models) to companies that serve end customers. The “real economy”, if you will. There are two main ways this will unfold. Firstly, we will start to see experienced AI researchers join firms to help with the integration. These new employees will come with high pay checks and may cause culture challenges, but that will be a necessary price for generating double digit productivity gains. More threatening, however, will be the tech firms that decide to compete in a vertical, much as Uber and Airbnb competed with taxis and hotels over the past decade. For this, they’ll require top talent with industry knowledge, in order to adapt the general tools to services which can compete with existing players. The first sign of this is likely to be a flow of top talent. Not a flood, but a few high-value, high-profile moves of individuals who can translate their industry to techies who have never worked there. If you start to see this, it's a sure sign that new competition is coming, and you should be preparing a response. Talent flows.
Be clear about the distinction between desirable and punishable behaviours and align incentives. Vagueness is a top reason behind employees hiding what they are doing. A recent survey found a third of workers have shared confidential information without permission, but many sledgehammer responses (scary warnings about everything vaguely AI) will also completely wipe out any hope of learning how to generate value with these tools… possibly without stopping the behaviour you were concerned by anyway. Incentive structures need to be aligned: reward those who find productivity improvements and assure them that you won’t use it as a chance for cost cutting. That sort of creative skill, to try to adapt a workflow to the new tools available is exactly the type of skill you’ll need more of in the future, even if not in that current role. Discussion. Related: Our article from last year on Top Down Principles.
Our Recent Work
We are growing! Announcement.
AI-driven Research Agents. Using AI tools for research and how to get started. Link.
Building Resilience to Technological Change. Steps you can take to navigate choppy waters. Link.
Zooming Out
Technological Disruption in the Labor Market. Larry Summers (who is on the board of OpenAI) and colleagues find “the years spanning 1990 to 2017 were less disruptive than any prior period we measure, going back to 1880”, before accelerating recently. Useful to dwell on how shifts caused by other general purpose technologies unfolded. Link.
Using AI tools to enable geographic expansion. Marc Benioff, Salesforce CEO, describes how LLMs - particularly translation - are encouraging companies to reach further afield, increasing competition. Link.
AI & Copyright. The UK Government is seeking views for the next month. Link.
Power requirements for AI Data Centres, and where around the world can provide them. Bloomberg New Energy Finance’s founder with a deep dive. Link.
Learning More
AI progress in 2024: A collection of the most insightful and reflected retrospectives. Nicklas Lundblad’s piece includes a fantastic agenda for 2025, to think about managing systems towards a stable outcome in a messy world, instead of trying to “solve problems”.
P.s. All 3 newsletters are worth a regular read!
A meticulous and vibrant summary of the most overused words by ChatGPT. Plus why it is crucial to navigate these to enhance your writing. (And advice on how!) Link.
Voice Agents. A deepdive into the companies building ways to talk to AI models. Link.
The Lighter Side
I don’t know that you have to be that nerdy to find this very cool. Video.