Wiredframe Radar logo

Category intelligence

Social Media Briefing — May 18, 2026

386 current items analyzed and ranked.

Executive synthesis

Social Media Summary

The AI community split between excitement over OpenAI Codex mobile capabilities and deep skepticism about model reliability. Gary Marcus dominated critical discourse with a viral breakdown of GPT-5.5's bizarre 'goblins' insertion bug and a $100K bet against Mustafa Suleyman on automation timelines.

Key Themes

AI Alignment & Reliability Failures · 8AI Progress & Singularity Debate · 5OpenAI Codex Mobile/Cross-Device · 4LLM Limitations · 5Neurosymbolic vs Pure Scaling · 5AI Safety & Power-Seeking · 3AI Automation Predictions & Bets · 4AI Infrastructure & Regional Economics · 5Agentic Workflows & Document Processing · 4OpenAI Business/Financial Criticism · 4

Primary evidence

Top Ranked Signals

Social Twitter May 17

What I am about to describe ain’t AGI; it’s a sign of a trillion dollar trainwreck. If I had told y...

By @GaryMarcus

85 score
AI Analysis

Marcus delivers extended critique of GPT-5.5's 'goblins' problem - the model inserting the word 'goblins' randomly, requiring hacky system prompt patches. Uses this as evidence that current AI is alchemy, not science, and questions AI safety viability.

What I am about to describe ain’t AGI; it’s a sign of a trillion dollar trainwreck. If I had told you in 2022 that the 2026 version of GPT (which by the way would only be GPT 5.5 and not GPT-6 or 7 like many people fantasized about) would still have strange quirks like inserting the word “goblins” in random places, y’all would have called me either “crazy” or “a hater” or both. “Scaling”, you would have shouted. “Deep learning is conquering walls!”, you would have said. And yet here we are.
GPT-5.5_issuesAI_alignmentAI_criticismsystem_promptsOpenAI_criticism
Social Twitter May 17

So the two most obvious barriers to some sort of true AI takeoff are robust RSI (AI acting as an ind...

By @emollick

82 score
AI Analysis

Emollick identifies two key barriers to AI takeoff: robust recursive self-improvement (RSI) where AI acts as independent researcher, and continual learning. Either would represent major trajectory change.

So the two most obvious barriers to some sort of true AI takeoff are robust RSI (AI acting as an independent AI researcher, rather than “merely” a multiplier of human effort) and continual learning. Either would represent a major change in trajectory for AI development.
AGI_barriersrecursive_self_improvementcontinual_learningAI_trajectory
Social Twitter May 17

@12exyz @dwarkesh_sp Exactly. I've been making this point for years: even within the human species,...

By @ylecun

75 score
AI Analysis

LeCun argues intelligence isn't correlated with desire for power, this drive won't spontaneously emerge in AI systems, and we'll build AI to do our bidding with guardrails

@12exyz @dwarkesh_sp Exactly. I've been making this point for years: even within the human species, intelligence is not correlated with a desire for power (or mere influence). The desire for influence is built into humans and other social species by evolution because individuals of social species need each other to survive. The desire for influence (let alone power or dominance) will not spontaneously pop up in AI systems unless we explicitly build this drive into them. And we won't. We will b
AI_safetypower_seekingAI_alignmentexistential_risk
Social Twitter May 17

@sriramk a) in an exponential progress curve every day is a singularity and no day is a singularity,...

By @ylecun

72 score
AI Analysis

Yann LeCun offers philosophical perspective on singularity: in exponential progress, every day is/isn't a singularity, all parts are similar, and no exponential continues forever

@sriramk a) in an exponential progress curve every day is a singularity and no day is a singularity, depending on your view. b) in an exponential progress curve, every part is similar to every other part. c) no exponential progress curve can continue forever.
singularityAI_progressexponential_growth
Social Twitter May 17

In the original von Neumann sense of a singularity as the point "beyond which human affairs, as we k...

By @emollick

72 score
AI Analysis

Emollick discusses von Neumann's original singularity concept as a point beyond which human affairs cannot continue, noting it seems true and that we can't know what it means in advance, comparing to Industrial Revolution

In the original von Neumann sense of a singularity as the point "beyond which human affairs, as we know them, could not continue," it seems true By definition, we can't know what that means in advance (it was also true of the Industrial Revolution, which reordered human affairs)
singularityAI_trajectoryhistorical_comparison
Social Twitter May 17

utter nonsense. accounting and legal will NOT be fully automated in 18 months. willing to put $100...

By @GaryMarcus

72 score
AI Analysis

Building on yesterday's Reddit discussion of Suleyman's claims, Marcus offers to bet $100K against Suleyman that accounting and legal will NOT be fully automated in 18 months

utter nonsense. accounting and legal will NOT be fully automated in 18 months. willing to put $100k against Suleyman on this.
AI_automationpredictionsemploymentlegal_AIOpenAI_criticism
Social Twitter May 17

so much joy in asking codex for random questions at work (such as finding some specific spreadsheet ...

By @gdb

68 score
AI Analysis

Greg Brockman (OpenAI co-founder) shares his delight using Codex for ad-hoc work queries like finding spreadsheets, finding it more enjoyable than manual searching

so much joy in asking codex for random questions at work (such as finding some specific spreadsheet i'd been looking at a while ago), much more fun than searching around for context by hand
CodexOpenAI_productsAI_productivity
Social Twitter May 17

symbolic tools like those listed below - rather than pure scaling -are likely driving most of the ad...

By @GaryMarcus

65 score
AI Analysis

Marcus argues symbolic tools (not pure scaling) are driving most current AI advances, concluding that hyperscaling is probably insane

symbolic tools like those listed below - rather than pure scaling -are likely driving most of the advances now. once you realize that you realize that hyperscaling is probably insane.
neurosymbolic_AIscaling_debatesymbolic_AIAI_architecture
Social Twitter May 17

🚨Breaking new study: memory in LLM agents still can’t really be trusted, even after over trillion do...

By @GaryMarcus

65 score
AI Analysis

Marcus highlights new study showing memory in LLM agents still can't be trusted despite over $1 trillion invested in the field

🚨Breaking new study: memory in LLM agents still can’t really be trusted, even after over trillion dollars has gone into the development of the field.
LLM_agentsmemory_reliabilityAI_limitationsAI_investment