Today’s Most Certain Prediction: Uncertainty Continues to Rise
It’s been yet another wild week. Time for your weekend reading…
Dear Friend,
From LLMs, which degrade the more data you input into them, to AI coding tools which decrease (not increase) developer productivity, to levels of uncertainty previously unheard of (and you thought COVID was bad… or the 2008/9 financial crisis…), it has been a wild ride this week. However, this being said, there is also a ton of good stuff happening, from surgical robots and extremely long-lasting batteries to trains that are faster than airplanes… The world remains a paradox.
Enjoy this week’s weekend read…
P.S. My latest Disrupt Disruption podcast dropped! Stefanie Klein leads the Women4Metals initiative at Aurubis, dedicated to empowering women in the male-dominated metals industry. She argues that diversity is not just a social good but a business necessity that widens the talent pool and leads to better decisions. The initiative's core philosophy is to “fix the system, not the women,” by changing the corporate framework to be more inclusive. For those navigating their careers, Stefanie stresses building a diverse network and seeking out sponsorship, noting that women are often “over-mentored but under-sponsored.” She advises young people to remain curious and stay open to all opportunities, especially in STEM fields. Listen here (and on your favorite podcast platform).
P.P.S. Our friend John Fallon, former CEO of Pearson, just published his excellent book “Resurgent: How established organisations can fight back and thrive in an age of digital transformation”. Recommended read for any leader of an incumbent organization. We will have a podcast episode with John dropping next week…
Headlines from the Future
Context Rot: How Increasing Input Tokens Impacts LLM Performance ↗
Fascinating, and possibly far-reaching, insights on the effect of making use of the ever-increasing context windows (the amount of text and data you can feed into an LLM) have on accuracy and quality of results.
“Recent developments in LLMs show a trend toward longer context windows, with the input token count of the latest models reaching the millions. Because these models achieve near-perfect scores on widely adopted benchmarks like Needle in a Haystack (NIAH), it’s often assumed that their performance is uniform across long-context tasks. […] We demonstrate that even under these minimal conditions, model performance degrades as input length increases, often in surprising and non-uniform ways. Real-world applications typically involve much greater complexity, implying that the influence of input length may be even more pronounced in practice.”
“More broadly, our findings point to the importance of context engineering: the careful construction and management of a model’s context window. Where and how information is presented in a model’s context strongly influences task performance, making this a meaningful direction of future work for optimizing model performance.”
—//—
How GLP-1s Are Breaking Life Insurance ↗
Here’s an interesting one (talking about the implications of the implication – or, in other words, our Disruption Mapping tool):
GLP-1s (Ozempic) have the potential to break the life insurance industry – and maybe not for the reasons you would expect.
Life insurers can predict when you’ll die with about 98% accuracy. […] Typically, underwriters- suspiciously sounds like undertakers-rely on a handful of key health metrics like HbA1c, cholesterol, blood pressure, and BMI to calculate your risk of dying earlier than expected (and thus costing them money). Those eagle-eyed readers among you have probably noticed something interesting already. Those same four metrics are exactly what GLP‑1s improve. Not just a little, but enough to entirely shift someone’s risk profile within at least 6 months of using them.
The insurer sees a ‘mirage’ of good health and approves them as low-risk. […] If we assume about 65% of people who start GLP-1 medications quit by the end of year one, that creates a big problem. When someone stops the medication, they’ll usually regain the weight they lost, and in two years, most of those key health indicators bounce back to their starting point.
Yep, it’s going to get messy.
—//—
World Uncertainty Index Q2 2025 Edition ↗
As the saying goes: A picture is worth a thousand words.
The Federal Reserve Bank of St. Louis has updated the World Uncertainty Index for the second quarter of 2025 , a period characterized by US tariffs, the One Big Beautiful Bill Act, and a slew of other actions across the globe.
And you thought the COVID years were bad…
—//—
Bad Actors are Grooming LLMs to Produce Falsehoods ↗
Here’s another fun attack vector for your LLM:
GenAI powered chatbots’ lack of reasoning can directly contribute to the nefarious effects of LLM grooming: the mass-production and duplication of false narratives online with the intent of manipulating LLM outputs. […] Current models ‘know’ that Pravda is a disinformation ring, and they ‘know’ what LLM grooming is but can’t put two and two together.
This is not just theoretical – it’s happening…
Model o3, OpenAI’s allegedly state of the art ‘reasoning’ model still let Pravda content through 28.6% of the time in response to specific prompts, and 4o cited Pravda content in five out of seven (71.4%) times.
Sigh…
Systems of naive mimicry and regurgitation, such as the AIs we have now, are soiling their own futures (and training databases) every time they unthinkingly repeat propaganda.
—//—
Not So Fast: AI Coding Tools Can Actually Reduce Productivity ↗
If you take away just one thing from this study, it should probably be this: when people report that AI has accelerated their work, they might be wrong!
This is the summary of a recent study on the effect of AI-powered coding assistants on developer productivity.
The results are surprising everyone: a 19 percent decrease in productivity. Even the study participants themselves were surprised: they estimated that AI had increased their productivity by 20 percent.
In essence: Not good. The one thing we collectively point to, when asked about the productivity impact of AI, coding tools, seems to fail us (for now, and in the specific context of this study – to be clear and fair).
The study was carried out in pretty much the most rigorous fashion possible: an honest-to-goodness randomized controlled trial under real-world conditions. The subjects were experienced developers carrying out their everyday work.
The reasons for the productivity loss make a ton of sense:
The biggest issue is that the code generated by AI tools was generally not up to the high standards of these open-source projects. Developers spent substantial amounts of time reviewing the AI’s output, which often led to multiple rounds of prompting the AI, waiting for it to generate code, reviewing the code, discarding it as fatally flawed, and prompting the AI again.
But where there is darkness, there is light:
Typically, large productivity boosts occur for **small, well-defined, greenfield projects**, or when an engineer is first learning a new language or API. […] Less experienced developers showed higher adoption rates and greater productivity gains.”
What We Are Reading
🤖 I Felt Pure, Unconditional Love’: the People Who Marry Their AI Chatbots People are spending countless hours chatting with AI companions and forming such deep bonds that some users are actually falling in love and even getting married to their digital partners. @Jane
🌱 Why James Dyson Got Into Vertical Farming Dyson is revolutionizing agriculture by integrating high-tech engineering—such as rotating strawberry-growing wheels, automated robots, and renewable energy systems—into sustainable farming practices to boost yields and reduce environmental impact. @Mafe
⚠️ We Warned About the First China Shock. The Next One Will Be Worse. This puts a key megatrend of the past decade (China’s ascent to the top of the global tech and research sectors) into perspective and then really lays out the implications for the coming decades. A future we’ve seen coming has arrived. @Jeffrey
🎯 Aiming AI at Society’s Toughest Challenges AI isn’t just for optimizing ads or automating emails. X’s Chief Science Officer makes the case for aiming it at humanity’s hardest problems, where messy data and high stakes are the norm, not the edge case. @Kacee
🧠 Will AI make you stupid? While recent studies draw attention to how the adoption of AI reduces mental load and critical thinking, they cannot yet proclaim a reduction in mental and critical thinking capacity. Regardless of whether that correlation truly exists, the justifiable outlook seems to be supported that those with greater critical thinking capabilities will only become more effective and productive, while those who don’t have the highest critical thinking capabilities will fall further behind. @Julian
🎭 Why Are We Lying To Young People About Work? A worthwhile reflection on what it means to find meaningful work – and why advice such as “find your passion” is just outright BS. Give this to every young person in your vicinity to read. @Pascal
Rabbit Hole Recommendations
A good resource for anyone wanting to learn to use AI (better): AI Fluency: Frameworks and Foundations
For anyone wanting to build (or understand) AI agents, this is a helpful framework: Building Effective AI Agents
Is Winter Coming? Thoughts on artificial intelligence and lofty expectations.
26.7% to 81.2%: Japan using generative AI less than other countries
Gaming cancer: How citizen science games could help cure disease
Robot surgery on humans could be trialled within decade after success on pig organs
Ex-Waymo engineers launch Bedrock Robotics with $80M to automate construction
China’s astonishing maglev train is faster than most planes (620 mph!)
This amazing new battery has life so long you may never have to recharge
Fascinating: 75% of all restaurant orders are now delivered, not eaten on the premises…
Happy Distractions
🖼️ MacPaint art from the mid-80s still looks great today
👽 Reading Neuromancer for the very first time in 2025
🏃🏼♀️➡️ Gorgeous visualization (and impressive feat): No days off – running every day since July 11, 2015
💩 Bullshit Remover - Paste crap, get truth