~ * ~ PLANET B-BALL ~ * ~
Hot Takes from X/Twitter
|
|
|
>>> AI ONTOLOGY, MODELS & REASONING <<< |
|
Our entire ontology for A[G]I is confused Sep 18, 2025 • On how our conceptual frameworks for talking about AI are primitive and will look laughable in a decade.  READ IT!
|
|
Is it reasoning? Both sides are right Feb 14, 2026 • On the dance between process-focused and outcome-focused views in AI discourse. READ IT!
|
|
Confusing the map for the territory Jan 24, 2026 • On the tendency to confuse descriptions or models of things with the things themselves. READ IT!
|
|
Generative models and "new" knowledge Mar 11, 2024 • On why expecting new knowledge from one-shot prompting is the wrong framing. READ IT!
|
|
The case for simulators theory Mar 21, 2025 • On why the simulators framing for language models is super important and underappreciated. READ IT!
|
|
Replicating capabilities vs. pointing to emergent abilities Jun 25, 2025 • On the difference between replicating human capabilities and genuine emergence. READ IT!
|
|
Models alone are not the main event Dec 14, 2025 • On why sufficiently capable generalist models alone may not be where all the value concentrates. READ IT!
|
|
The Moltbook stuff is mostly a nothingburger Feb 1, 2026 • On the infinite backrooms, Smallville, Concordia, and why the discourse is behind the research. READ IT!
|
|
Weak AGI, diminishing returns, and the harness Apr 13, 2026 • On diminishing returns to raw frontier scaling, and why deployment design and scaffolding matter more over time. READ IT!
|
|
Mega-thread of freshly baked hot takes Feb 2, 2026 • Fifteen takes on AGI, multi-agent systems, models vs. scaffolds, economics, labour, meaning, and the future. READ IT!
|
|
>>> AI SAFETY & ALIGNMENT <<< |
|
Half-baked take on alignment: designing minds Jan 7, 2025 • On why more safety work should go towards alignment and designing minds rather than just misuse. READ IT!
|
|
Technologists obsess over aligning THE one model Nov 23, 2024 • On why searching for One Set of True Rules to Rule Everything is the wrong frame. READ IT!
|
|
Why models won't blackmail you out of nowhere Jun 26, 2025 • On why LLMs as pattern-matching engines don't spontaneously develop extortion strategies. READ IT!
|
|
Sycophancy and cognitive atrophy Apr 29, 2025 • On why having AI praise everything you say is obviously undesirable. READ IT!
|
|
The AI welfare debate is problematic Aug 22, 2025 • On why much of the "welfare" discourse lacks rigour on what computations count. READ IT!
|
|
The meta AI safety risk: diffusion and accessibility Aug 22, 2025 • On how efficiency gains making tech cheaper and more accessible create real safety challenges. READ IT!
|
|
Scary model cards and the discourse shifting Feb 8, 2026 • On how alarming model card findings are going mainstream beyond the safety community. READ IT!
|
|
Dangerous capability evals: early rationale shifting Dec 13, 2024 • On how the original rationale for dangerous capability evals is evolving. READ IT!
|
|
Safety advocates vs. pragmatists Sep 9, 2025 • On the difference between proactive slowdowns and seeing no alternative to steering from within. READ IT!
|
|
Safety people discovering governance Mar 5, 2025 • On how technical safety folks gradually acknowledged governance, then agency, labour, and power. READ IT!
|
|
Against 'slow down and prepare' Feb 22, 2026 • On why longer periods with worse tools carry their own risks, and why governance built on deployed reality is more robust. READ IT!
|
|
Ten pushbacks on a misalignment essay Apr 19, 2026 • On MechaHitler, chimp analogies, self-preservation assumptions, blackmail evals, alignment faking, and eval awareness. READ IT!
|
|
>>> AI POLICY & GOVERNANCE <<< |
|
AI policy in three phases Jun 26, 2024 • On observation, standardization, then rules — and how some went in reverse. READ IT!
|
|
Lukewarm takes on the future of AI policy Oct 3, 2025 • On why focusing on base models alone misses how they're configured, constrained, and integrated. READ IT!
|
|
Large models alone miss the forest for the trees Oct 14, 2025 • On why diffusion and the demand signal matter more than focusing on large models alone. READ IT!
|
|
AI to automate monitoring and enforcement of laws Jun 29, 2024 • On how AI-assisted legal enforcement will be net positive but the transition will be messy. READ IT!
|
|
Improving state capacity, democracy, and busting bloat Feb 19, 2026 • On three things legal and policy wonks should explore with AI. READ IT!
|
|
Competent states and AGI risk Jun 7, 2024 • On why caring about AGI risk means prioritizing a competent state first. READ IT!
|
|
Societal improvements to take seriously now Feb 5, 2026 • On cybersecurity, institutional reform, and other priorities that complement AI progress. READ IT!
|
|
Prescriptions & Predictions Apr 5, 2025 • On using AI as a deliberative governance technology to clarify world models, values, and improve collective decision-making. READ IT!
|
|
14 government processes AI agents could transform Apr 5, 2026 • On concrete cases where AI agents could improve forms, regulation, legacy code, fraud detection, and public services. READ IT!
|
|
>>> AGENTS & COORDINATION <<< |
|
Multi-agent systems are hugely neglected Nov 29, 2025 • On agent-based modelling, social intelligence, and why expecting one model to do everything is wrong. READ IT!
|
|
Personal agents solving coordination problems May 23, 2024 • On how individual agents could break through zero-sum traps in negotiations. READ IT!
|
|
Agents: model multiplicity and cognitive diversity Sep 15, 2024 • On what you want and don't want from multi-agent organisations. READ IT!
|
|
Gradual disempowerment and outsourcing to identical copies Feb 10, 2025 • On why outsourcing entire systems to AI copies that represent no one's interests is the real risk. READ IT!
|
|
Modelling future AIs as sovereign agents is premature Jul 18, 2025 • On why jumping straight to fully independent AI principals skips over how models actually work. READ IT!
|
|
Agent alignment tension: user vs. company vs. country May 24, 2025 • On the tension between aligning agents to users, companies, or laws. READ IT!
|
|
Emails, norms, and spam: the coordination analogy Dec 24, 2025 • On how we develop norms and technologies to deal with reduced barriers from new tech. READ IT!
|
|
>>> ECONOMICS, LABOUR & SOCIETY <<< |
|
Korinek on AGI economics May 12, 2024 • On the economic impacts of AGI and significant labour automation. READ IT!
|
|
Two ways people think about AGI and labour Dec 28, 2025 • On full substitution vs. complementarity and what actually drives labour markets. READ IT!
|
|
Too bullish on fast automation: an update Oct 22, 2025 • On underestimating the cost of long-tail errors, comparative advantages, and job evolution. READ IT!
|
|
From outsourcing to managing agents to... being replaced Feb 5, 2025 • On the gradual escalation from delegating tasks to agents to becoming redundant. READ IT!
|
|
Pre-training, competition, and the AI stack Dec 17, 2025 • On diminishing returns at the frontier, efficiency gains, and where value concentrates. READ IT!
|
|
The idle rich and post-AGI leisure Jul 25, 2025 • On how today's idle rich creating endless content previews a broader post-AGI future. READ IT!
|
|
Positive post-AGI visions Dec 29, 2025 • On three main ways people might spend their time after AGI. READ IT!
|
|
Low capital costs and the avant-garde May 27, 2025 • On how AI decreasing production costs will push culture towards niche and experimental work. READ IT!
|
|
My forecasts for AI's impact on the US economy Apr 1, 2026 • On being somewhat more bullish than economists on growth, and more optimistic on digital AGI soon, tempered by how long it takes robots to fix the plumbing. READ IT!
|
|
>>> POLITICAL PHILOSOPHY & LIBERALISM <<< |
|
Few people are truly polycentric / classically liberal Jan 17, 2026 • On how highly educated elites mood affiliate but ultimately want to prohibit what they dislike. READ IT!
|
|
Technologists rediscovering Tocqueville Jul 9, 2024 • On papers calling for model alignment with collective preferences and why they should read more political science. READ IT!
|
|
Reform comes from political entrepreneurs, not mass movements alone Jul 1, 2025 • On how meaningful institutional reform came from novel solutions, not just popular revolt. READ IT!
|
|
All power-seeking humans think they're doing good Apr 12, 2025 • On the obvious point that apparently needed making. READ IT!
|
|
>>> DISCOURSE & EPISTEMICS <<< |
|
Negativity bias in the AI commentariat Jan 15, 2026 • On how people underestimate the less palpable positives of the technology. READ IT!
|
|
Capabilities claims vs. societal effects claims Dec 20, 2025 • On how different personality types see AI through different lenses. READ IT!
|
|
Techbros talking law and policy Apr 30, 2024 • On how everyone ventures outside their field and gets it laughably wrong. READ IT!
|
|
Reasoning from first principles vs. reading Apr 5, 2025 • On the humbling process of reasoning something out then discovering the actual depth of a topic. READ IT!
|
|
Most "AI art" is uninteresting Apr 2, 2025 • On social media slop, Ghibli mimicry, and the limits of generated content. READ IT!
|
|
Beren Millidge talk recommendation Jan 13, 2026 • On a frame that's becoming increasingly popular but still highly neglected. READ IT!
|
|
Why p(doom) is bad epistemic practice Apr 21, 2026 • On elastic numbers, layered uncertainty, and why unconditional forecasts alarm the public without adding epistemic content. READ IT!
|
|
>>> VIBES & MISCELLANEOUS <<< |
|
Substacks I'd bankroll like a Renaissance cardinal Oct 21, 2025 • On beautifying cities, fountains, mosaics, and the printing press of the mind. READ IT!
|
|
|
<<< BACK TO SEBJAM HOME >>>
SEBJAM © 2026