Sabitlenmiş Tweet
Diplomacy is firmly in the slow lane of the AI revolution right now, and most people don’t realise why.
The latest models have made staggering leaps this year - but those leaps are heavily skewed toward domains where success can be automatically verified (think: code that passes unit tests, maths proofs that check out, vulnerabilities that actually get exploited).
This is the power of the relatively new RLVR - Reinforcement Learning with Verifiable Rewards used to train the latest models - and it explains why programming and technical research feel like they’re on steroids while writing, strategy, and nuanced human work like diplomacy are improving more gradually.
But even among the cutting-edge users, the biggest gains are still “peaky” - concentrated in technical domains because that’s where the verifiable rewards and the money are.
So what’s actually state of the art and powerful for ‘office style’ non-technical work like diplomacy?
I think two things:-
1. Graphs of Reasoning (how the latest models “think”): Instead of just doing simple step-by-step Chain-of-Thought, the latest systems can now build branching, interconnected webs of ideas - exploring multiple scenarios at once, merging insights, looping back to check assumptions, and calling external tools along the way. It’s no longer a straight line; it’s a proper reasoning graph. This is exactly what powers the new agentic models, and it’s a massive upgrade for complex, open-ended problems.
2. Synthesis, scenario planning, and drafting: A frontier agent can now ingest thousands of pages of cables, treaties, reports, and real-time news in one go thanks to much bigger and more efficient ‘short term memory’ (context windows). It can then:
- Run branching “what-if” scenario analyses with built-in verification loops
- Red-team its own drafts for inconsistencies or weaknesses
- Produce polished, deeply researched, multilingual diplomatic notes in minutes
In short, the AI has become an extraordinarily capable staff officer - a tireless research and drafting partner.
It’s not replacing diplomats. But it definitely can amplify the best ones. I worry about the new ones.
The people who understand this gap (and are using the latest paid agentic tools professionally) see the slope of improvement as steep and consequential. Too many people used last year’s free models and are still laughing at last year’s hallucinations.
Diplomacy, strategy, and most non-technical white-collar work are in the slow lane for now - but that lane is still moving faster than almost any previous technology in history.
The gap between “what most people think AI can do” and “what the frontier systems can actually do for high-stakes analysis and synthesis” is widening fast.
English
