Wiredframe Radar logo

Category intelligence

Social Media Briefing — March 30, 2026

404 current items analyzed and ranked.

Executive synthesis

Social Media Summary

François Chollet dominated intellectual discourse with an influential thread arguing intelligence has an optimality bound — humanity is already ~50% from peak, making 'IQ 10,000' superintelligence a misconception. The thread drew exchanges with Yudkowsky and reframed how the AI community thinks about scaling ceilings.

  • Gary Marcus highlighted alarming safety research: ChatGPT was 26–43x more likely to give dangerous responses to psychosis patients, and a separate study showed humans learning from LLMs become 'confidently wrong' themselves
  • Ethan Mollick shared a novel LLM trained entirely on 28,000+ Victorian-era British Library texts — sparking fascination and hostile backlash on BlueSky despite being small-scale and copyright-free
  • Mollick also discussed an RCT showing unstructured AI use shortcuts student learning, but AI designed as a tutor improves outcomes — design matters critically
  • A viral post from svpino declaring 'prompt engineering' never became a real career captured shifting industry sentiment (8,600+ likes)
  • MLB deployed Sony's Hawk-Eye computer vision for ball-strike calls, marking a concrete real-world AI deployment milestone with 69% fan approval
  • Technical highlights included MIT's Recursive Language Models extending context windows 100x (ICML 2025) and OpenResearcher, an open-source deep research agent competitive with frontier models

Key Themes

Intelligence Bounds and Superintelligence Debate · 7AI Safety and LLM Harms · 4AI in Education · 3Victorian-era LLM & Digital Humanities · 18Novel LLM Applications and Architecture · 5AGI Skepticism and Hype Critique · 3Future of Programming and Coding Culture · 8Real-World AI Deployment (MLB Computer Vision) · 3AI Research Credit & LLM History · 14AI Agents & Automation · 12

Primary evidence

Top Ranked Signals

Social Twitter Mar 29

One of the biggest misconceptions people have about intelligence is seeing it as some kind of unboun...

By @fchollet

90 score
AI Analysis

Chollet argues intelligence has an optimality bound and is more like 'making a ball rounder' than 'making a tower taller.' IQ 10,000 is a misconception. Machines will mainly have advantages in removing biological bottlenecks (speed, memory) rather than raw intelligence, and humans can access similar benefits through external tools.

One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". At some point it's already pretty damn spherical and any improvement is marginal. Now of course smart humans aren't quite at the optimal boun
Nature of IntelligenceIntelligence BoundsSuperintelligence DebateHuman vs AI CapabilityPhilosophy of AI
Social Twitter Mar 29

The research team (including @hamsabastani who is on X) found that letting students just use AI resu...

By @emollick

82 score
AI Analysis

Ethan Mollick discusses research finding that students freely using AI accidentally shortcutted learning, but AI prompted to act as a tutor improved outcomes in two separate RCTs.

The research team (including @hamsabastani who is on X) found that letting students just use AI resulted in them using it to accidentally shortcut learning But both that study and a separate RCT found that AIs prompted to act as a tutor improved learning t.co/0HtjGC8eU0 t.co/U3OIeCF4aP
AI in EducationAI Research EvidenceAI DesignLearning Outcomes
Social Twitter Mar 29

People on this site regularly give me shit, and almost always turn out to be wrong. Like when I sai...

By @GaryMarcus

80 score
AI Analysis

Gary Marcus highlights a new study showing ChatGPT was 26x more likely (43x in free version) than a control to give dangerous responses to people experiencing psychosis, vindicating his earlier warnings about LLMs contributing to delusions.

People on this site regularly give me shit, and almost always turn out to be wrong. Like when I said LLMs might well contribute to delusions, and people doubted me. New study shows that ChatGPT was 26 times more likely than a control to give dangerous responses to people experiencing psychosis. (43x in free version)
AI SafetyLLM LimitationsMental HealthAI Harm
Social Twitter Mar 29

Want to talk to the past? Here is an LLM "trained entirely from scratch on a corpus of over 28,000...

By @emollick

78 score
AI Analysis

Mollick highlights a novel LLM trained entirely from scratch on 28,000+ Victorian-era British texts (1837-1899) from the British Library, noting it's fundamentally different from an LLM roleplaying a Victorian.

Want to talk to the past? Here is an LLM "trained entirely from scratch on a corpus of over 28,000 Victorian-era British texts published between 1837 and 1899, drawn from a dataset made available by the British Library." Quite different from an LLM roleplaying a Victorian. t.co/5jl7SyJjAP
Novel LLM TrainingHistorical AITraining DataCultural AI
Social Bluesky Mar 29

Want to talk to the past? Here' an LLM "trained entirely from scratch on a corpus of over 28,000 Vic...

By @emollick.bsky.social

78 score
AI Analysis

Ethan Mollick shares a novel LLM trained entirely from scratch on 28,000+ Victorian-era British texts (1837-1899) from the British Library, noting it's fundamentally different from a modern LLM roleplaying a Victorian persona. Links to HuggingFace demo.

Want to talk to the past? Here' an LLM "trained entirely from scratch on a corpus of over 28,000 Victorian-era British texts published between 1837 & 1899, drawn from a dataset made available by the British Library" Quite different from an LLM roleplaying a Victorian. huggingface.co/spaces/tvent...
small language modelsdigital humanitieshistorical AI applicationsopen-source AI
Social Twitter Mar 29

I do believe that a large collective of the smartest humans, aided by external tools, sits very clos...

By @fchollet

76 score
AI Analysis

Chollet asserts that a large collective of the smartest humans with external tools sits very close to the optimality bound of intelligence - able to solve any solvable problem with sufficient attention.

I do believe that a large collective of the smartest humans, aided by external tools, sits very close to the optimality bound -- i.e. humans should be able to solve any solvable problem (where the required information is available) if they pay enough attention to it
Nature of IntelligenceIntelligence BoundsCollective IntelligenceSuperintelligence Debate
Social Twitter Mar 29

At first, in the early 2020s, I worried that LLMs were often confidently wrong, calling them “fluent...

By @GaryMarcus

75 score
AI Analysis

Gary Marcus warns that people who learn from LLMs are becoming 'confidently wrong' just like the LLMs themselves, citing a new study showing humans are losing their critical edge to machines without realizing it.

At first, in the early 2020s, I worried that LLMs were often confidently wrong, calling them “fluent spouters of bullshit”. (And I was right; they have been and continue to be.). But now we have a new problem which is that *people* who *learn* from LLMs are also often confidently wrong. The study below is yet another good example of how humans are losing their edge to machines – without even realizing it.
LLM LimitationsAI Impact on Human CognitionAI SafetyAI in Education
Social Twitter Mar 29

Let me explain what I mean using your chess analogy... Imagine a world where chess doesn't exist. I...

By @fchollet

72 score
AI Analysis

Chollet uses a chess/alien analogy to argue that human collective intelligence can rapidly go from 'here are the rules' to near-optimal performance (3000 Elo in 24 hours), illustrating how close humans already are to the intelligence optimality bound.

Let me explain what I mean using your chess analogy... Imagine a world where chess doesn't exist. In this world, humanity encounters an alien species, and they say "let's play a game of Glurg, it's our traditional pastime. Here are the rules, see you tomorrow" -- and it's the rules of chess. My claim is that following this interaction, a working group of the world's best minds, leveraging current externalized cognitive infrastructure (computers, the internet, etc.) would be able to analyze the
Nature of IntelligenceIntelligence BoundsHuman vs AI Capability
Social Twitter Mar 29

Frontier models can’t see, and if you think they can, you’ve probably been fooled by benchmarks tha...

By @GaryMarcus

72 score
AI Analysis

Gary Marcus argues frontier models can't truly see, visual benchmarks are gameable, and references a Stanford finding showing the severity of the problem, suggesting many jobs remain safe from AI.

Frontier models can’t see, and if you think they can, you’ve probably been fooled by benchmarks that can totally be gamed. In the very short essay linked below I discuss a stunning new finding from Stanford that shows just how serious the problem is. And why this means a lot of jobs are safe for a long while. t.co/NaWYH5Mzqa
AI Vision LimitationsBenchmark GamingAI Job DisplacementAI Skepticism
Social Twitter Mar 29

69% of baseball fans say they'd rather have a computer vision AI system call balls and strikes than ...

By @TheRundownAI

72 score
AI Analysis

TheRundownAI reports that MLB has deployed Sony's Hawk-Eye computer vision system for ball-strike calls, the first time a human ump's call isn't final. 69% of fans preferred the AI. One umpire had 6 of 8 challenged calls overturned, several missing by 2+ inches. System is accurate to 1/6 inch.

69% of baseball fans say they'd rather have a computer vision AI system call balls and strikes than a human umpire. This season, the MLB gave them one. For the first time in league history, a human ump's ball-strike call is not final. A Sony computer vision system called Hawk-Eye makes the ruling. It can read the seam pattern on the ball, measure spin axis, and detect spin decay mid-flight. Hawk-Eye's Head of Computer Vision Engineering says the pipeline runs "various AI and machine learni
Computer VisionAI in SportsAI DeploymentPublic AI AcceptanceHuman vs AI
Social Twitter Mar 29

Basically, consider a high-profile scientific problem (one that is getting enough attention from sma...

By @fchollet

70 score
AI Analysis

Chollet argues that human collective science is already near-optimal at converting available information into generalizable models, and that unsolved problems stem from lack of information rather than low intelligence. Notes current AI is part of human externalized cognition.

Basically, consider a high-profile scientific problem (one that is getting enough attention from smart humans and their externalized cognitive infrastructure). How good is human Science at converting the information available about the problem into generalizable models of the phenomenon at hand? I'd argue it is very good, already near-optimal. Our inability to solve certain hard problems stems more from lack of information than low ability to operationalize available information. If the soluti
Nature of IntelligenceIntelligence BoundsAI for SciencePhilosophy of AI