Wiredframe Radar logo

Category intelligence

Social Media Briefing — March 29, 2026

342 current items analyzed and ranked.

Executive synthesis

Social Media Summary

Andrej Karpathy's massively viral post (2.4M views) on using LLMs to argue against your own position dominated the day, revealing how AI exposes weak arguments and the danger of sycophantic validation.

  • François Chollet predicted a post-AGI 'focus class vs. slop class' divide based on cognitive agency, sparking widespread discussion on AI-driven inequality
  • Yann LeCun forcefully argued that closed AI labs profit from open-source research without giving back, intensifying the open vs. closed models debate
  • Ethan Mollick shared counter-intuitive research showing AI may reduce political polarization, opposite to social media's effect
  • Cohere announced Transcribe, a SOTA open-source ASR model running in-browser, while NousResearch's Hermes agent framework gained significant traction
  • Nathan Lambert clarified Ai2's commitment to open models despite organizational uncertainty

On the builder side, Levelsio went viral demonstrating a startup built in 24 minutes with Claude Code and Grok 4.1. Greg Brockman offered a provocative framing: 'Codex use cases are like Skills, but for humans.' Aggregated signals also surfaced David Silver raising $1B for RL-based superintelligence and hints of Anthropic training a dramatically smaller competitive model.

Key Themes

LLMs as Thinking Tools / Sycophancy · 3Open vs. Closed AI Models · 6AGI Societal Impact & Class Divide · 1Major AI News Signals (via aggregated lists) · 3AI-Powered Rapid App Building / Vibe Coding · 14AI Agent Frameworks & Hermes Deep Dive · 5AI Social Impact Research · 1Ai2 Open Models & Funding Stability · 2Cohere Transcribe Launch · 5Grok Performance and New Capabilities · 6

Primary evidence

Top Ranked Signals

Social Twitter Mar 28

- Drafted a blog post - Used an LLM to meticulously improve the argument over 4 hours. - Wow, feelin...

By @karpathy

92 score
AI Analysis

Karpathy describes spending 4 hours refining a blog post argument with an LLM, then asking it to argue the opposite—which demolished his original position. Advises using LLMs to stress-test your own opinions by asking multiple directions.

  • Drafted a blog post
  • Used an LLM to meticulously improve the argument over 4 hours.
  • Wow, feeling great, it’s so convincing!
  • Fun idea let’s ask it to argue the opposite.
  • LLM demolishes the entire argument and convinces me that the opposite is in fact true.
  • lol
The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be car
LLM_sycophancyAI_as_thinking_toolcritical_thinkingAI_workflows
Social Twitter Mar 28

A lot of folks talk about "escaping the permanent underclass". If AGI pans out, the future class div...

By @fchollet

82 score
AI Analysis

Chollet predicts a post-AGI class divide: a 'focus class' that controls attention and does things vs. a 'slop class' whose reward loops are managed by AI. Argues cognitive agency, not wealth, will define future class structure.

A lot of folks talk about "escaping the permanent underclass". If AGI pans out, the future class divide won't be based on wealth, but on cognitive agency. There will be a "focus class" (those who control their attention and actually do things) and a "slop class" (those whose reward loops are fully RL-managed by AI)
AGI_societyAI_ethicsattention_economyclass_divideAI_philosophy
Social Twitter Mar 28

The Hermes Deep Dive. (The new hot AI agent harness). https://t.co/Ldmfiy9a6C Hey @Teknium got any...

By @Scobleizer

75 score
AI Analysis

Continuing our coverage from [yesterday](/?date=2026-03-28&category=social#item-9d8a1a3380c5), Scobleizer publishes a deep dive report on Hermes, described as 'the new hot AI agent harness' from NousResearch, tagging Teknium for additional input

The Hermes Deep Dive. (The new hot AI agent harness). t.co/Ldmfiy9a6C Hey @Teknium got anything to add?
AI agentsHermesNousResearchopen source AIagent frameworks
Social Bluesky Mar 28

Som evidence that AIs may reduce polarization, the opposite of the effect of social media: “while di...

By @emollick.bsky.social

75 score
AI Analysis

Ethan Mollick shares FT research suggesting AI may reduce political polarization, opposite to social media's effect. All AI platforms nudge people toward more moderate, expert-aligned stances.

Som evidence that AIs may reduce polarization, the opposite of the effect of social media: “while different AI platforms behave in subtly different ways, all of them nudge people away from the most extreme positions and towards more moderate and expert-aligned stances.” www.ft.com/content/3880...
AI_social_impactpolarizationAI_policyAI_research
Social Twitter Mar 28

@cohere transcribe Sota open source transcription model running in the browser :) Weights on @hu...

By @nickfrosst

72 score
AI Analysis

As first reported in Social on Mar 27, Nick Frosst announces Cohere Transcribe: a state-of-the-art open-source transcription model running in the browser, with weights on Hugging Face.

@cohere transcribe Sota open source transcription model running in the browser :) Weights on @huggingface link below t.co/OmrHFA94lG
Cohere_Transcribespeech_recognitionopen_sourcebrowser_AI
Social Twitter Mar 28

A few facts, while the dust is settling. Ai2 still is... - releasing open models, folks want to, an...

By @natolambert

72 score
AI Analysis

Nathan Lambert clarifies Ai2's status: still releasing open models (required by NSF grant), has substantial compute, additionally funded by FFST for open model work. Expresses confidence in Ai2's year ahead.

A few facts, while the dust is settling. Ai2 still is...
  • releasing open models, folks want to, and it's actually required in the NSF grant
  • using substantial compute to do so from said grant
  • funded additionally by FFST (new funding body) on top of NSF, for work in open models
Overall I'm confident in Ai2 doing great work this year.
ai2open-modelsai-fundingopen-source-ai
Social Twitter Mar 28

Did someone turn the speed knob up on @Grok? Just did another transcript reading and, damn, it is ...

By @Scobleizer

72 score
AI Analysis

Scobleizer notices Grok has gotten significantly faster at transcript reading, asks if someone 'turned the speed knob up'. Goes massively viral with 900 likes and 27.3M views.

Did someone turn the speed knob up on @Grok? Just did another transcript reading and, damn, it is way faster than last time I asked it to do the same.
Grok_performancexAI_infrastructureAI_speed
Social Twitter Mar 28

✨ To prove my friend @StevieZollo (who's visiting me in Brazil) you don't need an idea, or even a lo...

By @levelsio

70 score
AI Analysis

Levelsio demonstrates building a startup in 24 minutes: took top idea from his site (AI bedtime stories for kids), built it with Claude Code, using Grok 4.1 for story generation, Google Gemini for TTS, and Stripe for payments.

✨ To prove my friend @StevieZollo (who's visiting me in Brazil) you don't need an idea, or even a lot of time these days to ship a little app that might make money I took the top idea from t.co/WNLj4eGmaq: "A startup that uses AI to generate personalized bedtime stories for kids based on their interests, family photos, and daily activities, delivered via a voice app. (❤️ 110 likes, 3 days ago)" So I copy pasted it into Claude Code and asked it to build it The first version of course
ai-codingvibe-codingclaudegrokgeministartup-ecosystemai-democratization
Social Twitter Mar 28

Grok's answer: +++++ The X list you linked (ID: 1953536336675365173) is "AI Newsmakers," curated b...

By @Scobleizer

70 score
AI Analysis

Scobleizer shares Grok's answer to reading his AI Newsmakers list — surfaces top posts including: Anthropic Claude session limits tightening prediction, Sam Altman highlighting mRNA vaccine designed with ChatGPT saving a dog, OpenAI Codex plugins, Matt Shumer claiming Anthropic trained a model 'dramatically smarter' than Claude Opus 4.6, and more.

Grok's answer: +++++ The X list you linked (ID: 1953536336675365173) is "AI Newsmakers," curated by you (@Scobleizer). It aggregates posts from AI/tech leaders, researchers, and commentators (including @sama, OpenAI accounts, Anthropic-adjacent folks, and more). I used X advanced search with the list:1953536336675365173 operator + since:2026-03-26 (covering the past ~24–48 hours from now) in Top mode (X’s popularity/engagement ranking) and pulled the top 10 results. These are the standout/mos
Anthropic_next_modelClaude_session_limitsAI_medical_applicationsGrok_capabilitiesAI_news_aggregationSora_shutdown
Social Twitter Mar 28

Codex use cases are like Skills, but for humans

By @gdb

68 score
AI Analysis

Continuing our coverage from [yesterday](/?date=2026-03-28&category=social#item-2d61b8a9bc72), Greg Brockman (OpenAI co-founder) says 'Codex use cases are like Skills, but for humans', implying Codex enables humans to acquire new capabilities on demand.

Codex use cases are like Skills, but for humans
OpenAI_CodexAI_augmentationproduct_vision
Social Twitter Mar 28

@dansemperepico You’re calling it a marketing wrapper but it’s an accessibility play. Why did OpenAI...

By @alliekmiller

68 score
AI Analysis

Allie K. Miller defends a product (likely Anthropic's Cowork) as an accessibility/abstraction play rather than a 'marketing wrapper', comparing it to how ChatGPT abstracted the GPT-3.5 API to drive adoption

@dansemperepico You’re calling it a marketing wrapper but it’s an accessibility play. Why did OpenAI release ChatGPT when they could have just released the 3.5 API? Because abstraction matters and meaningfully increases adoption and market share. (99% of my work is in Claude code but Cowork is more visual, 100x easier to connect tools and schedule tasks, access previous convos, dispatch in, etc)
AI product strategydeveloper toolsAnthropic CoworkAI accessibilityClaude Code