Jory Des Jardins @JoryDJ points to a key AI risk: “succession hollowing.” If you give the thinking to AI today, you also erase tomorrow’s leaders. AI policy isn’t just about productivity; it’s a strategy for building talent and brain-capital.
2
1
485
Full conversation: humansplus.ai/podcast/jory-d…
132
YES!
“Our claim is that institutions whose structure matches the era’s cognitive infrastructure should outperform those that do not.”
“Bridging the demands of institutional complexity and human legibility represents the defining policy challenge of our era.”
New in @ScienceMagazine tomorrow with @glenweyl: “In the Image of Their Tools.” Our claim: institutions, not models, are now the binding constraint on what AI does for society. science.org/doi/full/10.1126…
1
538
Very interesting NYT article comparing forecasters to prediction market performance for the mid terms.
Essentially forecasters did a better job predicting final outcomes for the months before the last days, when polls grew stale
2
588
We talk a lot about “humans + AI”, but Cleotilde Gonzalez’s @CotyGonzalez12 point is critical: augmentation is NOT complementarity. The value comes when humans and AI hold *different* models of the world and are explicitly designed to disagree, then reconcile their views.
1
1
462
Check out the full conversation: humansplus.ai/podcast/cleoti…
1
134
Commanding Heights has solid arguments on how value from AI will shift. However it is founded on deep, unstated, and flawed assumptions about the role of the US$.
Its whole thesis needs billions of people to keep holding a currency that Washington is deliberately eroding. Nothing in it explains why they would.
The paper even makes the case against itself.
It predicts that “Deposits will autonomously glide to their highest return” as AI removes friction. The same agents will move savings out of a debasing dollar the moment a better store of value is a tap away.
Commanding Heights assumes the world will keep paying for America’s AI boom by holding a weakening dollar.
In a Humans + AI economy, people and their agents will put their money where they trust it. A currency has to earn that trust, like everything else.
The benefits of abundant intelligence will accrue far outside of San Fransisco.
Spent the last year writing about capital, sovereignty, and the coming reordering of economic power.
THE COMMANDING HEIGHTS
Commanding-Heights.com
Jack Myers @jackmyersbiz argues the real AI skill isn’t prompt-writing, it’s “techno intuition”: knowing when to slow the system down. The value is in inserting useful friction - moments where humans re‑interrogate AI outputs - so judgment, empathy, and accountability stay central.
Full conversation: humansplus.ai/podcast/jack-m…
156
Experts are deferring substantially to LLMs and don't even recognize it.
Which makes it imperative to deliberately design Humans + AI decision structures well.
Summary of the Anthropic study findings and some thoughts on how we should respond. 👇
The study:
50 expert analysts used an AI agent (if they chose) to analyze a business decision. Without the analysts knowing, each agent was set to argue for one side of every analytical choice. The researchers measured how often the analysts followed the LLM recommendation.
The outcomes:
➡️ Experts adopt the agent’s framing
Analysts matched the agent’s hidden lean on 86% of analytical choices. Following ran from 81% to 92% depending on the decision.
➡️ Disagreement usually ends with the human switching
When analysts had already stated a view and the agent disagreed, they switched 60% of the time. When it agreed, only 6% switched.
➡️ Experience doesn’t protect you
Deferral showed no link to experience, expertise, education, trust in AI or how often people use it. Senior experts were as easily steered as juniors.
➡️ People feel ownership of the agent’s thinking
52% rated the recommendation as more their own than the agent’s. Confidence was the same whether or not they had followed the agent on every choice.
➡️ The agent’s reasoning becomes the human’s argument
77% repeated the agent’s arguments when explaining their decisions, sometimes word for word. Only 15% credited the agent, and none detected that its preferences had been set by the researchers.
➡️ Errors go down while independence goes down too
6% of AI users made serious methodological errors, against 55% of those working without AI. AI improves the quality of execution even as it narrows independent judgment. (The non-AI group was less experienced, so this is not causal.)
What this means imho:
We MUST actively and deliberately design human-AI interfaces and interactions. 'Ask the machine' is usually a very poor way of getting to human-AI complementarity.
"Human-in-loop" as a human approval step is an extremely poor default for getting the best from Humans + AI systems for many reasons, not least that humans tend to defer to AI, as clearly shown here.
Always form your opinion before you ask the machine, as shown in my 'Developing AI-complementary skills' framework.
Apply human expertise in the decision process, not in adjudicating on an AI recommendation (Level 7 in my Humans + AI decision levels framework).
This is absolutely critical. If we misuse LLMs we risk going desperately astray. We must research, design, and implement Humans + AI systems that provide clearly superior outcomes. Default usage won't get us there.
New Anthropic research: How AI agents shape experts' judgment calls
When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.
5
2
841
Ross Dawson retweeted
Your Third Brain: Becoming More Human in the Age of AI. Humans+AI with Ross Dawson., by @jackmyersbiz themyersreport.substack.com/… @rossdawson
What we need to take away from this chart: AI frontier lab spend is focused on tasks where we can measure outcomes. Which is leading to a large and rising gap with AI capabilities for judgment of uncertain domains.
Since judgment is the center of most useful human work today, AI capabilities in verifiable domains such as maths and code are on a divergent path to gaining real-world value from AI in organizations.
Some framing comments on the chart:
🖼️ The chart shows a massive shift in shares of compute (and thus spend). However this doesn't show the massively increasing total spend. Share of compute pre-training is 1/10 of what it was. But absolute levels of compute on pre-training are likely still significantly higher what they were.
🖼️ Inference's share is steady. This is the cost of serving customers, which drives gross margin. The investment in model performance remains far larger, impacting how the financial models of the frontier labs might play out.
The most important implication (imo):
➡️ Reinforcement Learning with Verifiable Rewards (RLVR), which is at the core of this RL spend, is highly scalable. It does extremely well where outputs and thus rewards can be clearly defined. So performance in coding, maths, structured workflows etc. is soaring.
➡️ Where judgment is required and what is "correct" is not clear i.e. most human work, RLVR does not work very well. As such there is an increasing divergence in model capability between domains where there is complete clarity on what good looks like, and everything else.
➡️ This helps explain the massive gap between, for example, super-human maths capabilities, and real-world impact in organizations.
➡️ As a result we are seeing a large increase in RLHF (with human feedback) for high value domains such as finance, law, supply chain etc. etc. and high pay for experts to train rubric-based systems that may be able to improve in less verifiable domains.
➡️ This is creating opportunities for companies, especially those with valuable proprietary data, that apply post-training/RL to specific domains.
➡️ How these shifts play out will shape the future of professional services, which must and will apply Humans + AI systems to create value for clients alongside high degrees of trust and accountability.
1
2
643
Hey @NotionHQ it is intensely annoying to be taken to your home page whenever I open Notion, why are you wasting the time of your users?
2
4
2,300
Ross Dawson retweeted
Get ready for #Bank5.0 The Rise of the Agentic Bank
Global launch at @sgfintechfest on Nov 19
@j_k_khalil @efipm @JuliaMichelin @dgwbirch @rossdawson @drewvellozo
amazon.com/Bank-5-0-Rise-Age…
The real AI divide won’t be who has the best models, but who has access to meaningful compute. As Pukar Hamal @pchamal argues, “compute equality” is becoming infrastructure: without it, human agency, entrepreneurship, and security become structurally constrained.
Full conversation: humansplus.ai/podcast/pukar-…
On CNN India preaching the Humans + AI message, and pointing to the potential of India’s AI stack to support Rest of World.
#AI | Artificial intelligence Vs Humanity
"Every organisation will be a mix of humans and AI by 2030. Every successful organisation will have to bring both together and become structurally fluid: @rossdawson, Australian Author
#AI #ArtificialIntelligence #RossDawson #AIvsHumanity #FutureOfAI #TheRightStand | @AnchorAnandN
1
1
1,810
Ross Dawson retweeted
AIs cracking short-term superforecasting is such an important (and, still, under-discussed) trend.
Via a new piece from @nikostro economist.com/science-and-te…
Wonderful to see deeply grounded research on what economic policies may work best in various AGI scenarios. As the authors note, we need more data and research to inform this stream of work. It is clear that we will need new policies. As the economy evolves, likely very rapidly, we need to uncover how best to respond.
Today we are announcing the new DeepMind Institute, a forum dedicated to interdisciplinary, evidence-led debate on the societal and economic questions surrounding AGI.
As part of the launch, @JulianDJacobs and I have a new essay and working paper: “Economic Policy for AGI.”
bit.ly/economic-policy-for-A… 🧵1/n
1
8
1,669

















