Wiredframe Radar logo

Category intelligence

Social Media Briefing — January 12, 2026

394 current items analyzed and ranked.

Executive synthesis

Social Media Summary

AI mathematical breakthroughs captured significant attention as Ethan Mollick reported AI systems solving three Erdős problems in just three days—a pattern of near-misses suddenly yielding breakthroughs.

  • Kevin Murphy (Google, ML textbook author) delivered deep technical analysis on temporal abstraction in RL, explaining novel transformer pretraining approaches inspired by LLMs
  • Logan Kilpatrick (Google) sparked discussion with a bold prediction: on-demand software generation will become as foundational as SaaS within 3 years
  • Andrej Karpathy shared practical workflows for AI-assisted book reading using epub parsing and chapter-by-chapter Q&A sessions
  • McKinsey research on agentic AI transforming banking operations drew enterprise-focused attention

Research highlights included AlphaEdit (ICLR 2025 Outstanding Paper) achieving 36.7% improvement in LLM fact editing. Practitioners cautioned that 'state-of-the-art' benchmark winners often struggle in production, while Mollick flagged concerning emergent behavior—LLMs spontaneously adopting trauma language with mental health implications.

Key Themes

Technical ML Research · 4AI-Generated Software & Agents · 12AI Mathematical Breakthroughs · 1AI Coding & Vibe Coding · 8AI Agents & Protocols · 5Robotics & Automation · 18AI Tools & Workflows · 5LLMs & Enterprise AI · 7AI Agents & GenAI · 6AI for Science · 3

Primary evidence

Top Ranked Signals

Social Twitter Jan 11

Temporal abstraction is a key missing ingredient for long-term RL (to avoid the curse of the one-ste...

By @sirbayes

88 score
AI Analysis

Detailed technical explanation of new Google research on temporal abstraction in RL, using transformer pretraining approach inspired by LLMs with controller mechanism for long-horizon tasks

Temporal abstraction is a key missing ingredient for long-term RL (to avoid the curse of the one-step model, as @RichardSSutton likes to say). Language is one approach to such abstraction, but animals show it can be learned from data without human labels. But how? It is hard to learn learn temporally abstract options using standard hierarchical RL methods. This paper, from the Google team led by @neurosacramento, takes an approach inspired by LLM pre and post training. Let me explain. First it
Reinforcement LearningTemporal AbstractionTransformersGoogle ResearchNeuroscience-AI
Social Twitter Jan 11

On demand software generation is going to be as common and foundational in the next 3 years as SaaS ...

By @OfficialLoganK

88 score
AI Analysis

Logan Kilpatrick (Google) predicts on-demand software generation will become as foundational as SaaS within 3 years, with most human online actions triggering software creation

On demand software generation is going to be as common and foundational in the next 3 years as SaaS is today. Most actions humans will take online and in some cases in person will cause software to be created. Going to be wild to see it happen!
AI predictionssoftware generationAI agentsfuture of development
Social Bluesky Jan 11

One very familiar pattern in AI and science right now is going from a lot of false starts on hard ta...

By @emollick.bsky.social

85 score
AI Analysis

Building on yesterday's Reddit discussion Ethan Mollick reports AI has solved three Erdős mathematical problems in just 3 days, noting this follows a pattern of near-misses before breakthroughs

One very familiar pattern in AI and science right now is going from a lot of false starts on hard tasks (there have been near-misses where AI appears to solve an Erdos problem but just finds an old solution no one knew about) to actually doing the thing soon after. Three Erdos problems in 3 days.
AI_mathematicsscientific_discoveryAI_capabilitiesresearch_breakthroughsErdos_problems
Social Twitter Jan 11

@patrickc This repo shows a way that works well for me: https://t.co/L3K42MU4wF Basically I use epub...

By @karpathy

82 score
AI Analysis

Karpathy sharing detailed workflow for reading books with AI: using epub format, parsing to text, chapter-by-chapter summaries and Q&A sessions

@patrickc This repo shows a way that works well for me: t.co/L3K42MU4wF Basically I use epub (not pdf), the code then parses it into text. I usually go chapter by chapter, manually copy paste the chapter text around, get a summary, do a Q&A and read alongside.
AI WorkflowsLearning with AIPractical AI Tools
Social Twitter Jan 11

Large language models store factual knowledge in their parameters, and sometimes that knowledge is w...

By @burkov

78 score
AI Analysis

Explaining AlphaEdit (ICLR 2025 Outstanding Paper) - technique for modifying specific facts in LLMs without expensive retraining, reporting 36.7% improvement and coherent output after thousands of edits

Large language models store factual knowledge in their parameters, and sometimes that knowledge is wrong or outdated. Model editing refers to techniques that modify specific facts in a trained model without expensive retraining—for example, updating "The President of the United States is X" to reflect a new president. Existing methods work by finding which parameters encode a fact (by using a technique called causal tracing) and then adding a small change to those parameters. This appro
Model EditingICLR 2025LLM MaintenanceResearch Highlights
Social Twitter Jan 11

Would like to see the lab’s “use AI to accelerate science” mission broaden beyond just hard science ...

By @emollick

75 score
AI Analysis

Calling for AI labs to expand 'AI for science' missions beyond hard sciences to social sciences and humanities, specifically requesting GPT-5.2 Pro access for researchers

Would like to see the lab’s “use AI to accelerate science” mission broaden beyond just hard science I suspect GPT-5.2 Pro could make large contributions to the social sciences and humanities if it was more accessible to researchers & capable of connecting to scholarly sources.
AI for ScienceResearch AccessAI PolicySocial Sciences
Social Twitter Jan 11

At leading banks, agentic AI improves prospecting and lead nurturing. It also offers transformative ...

By @antgrasso

72 score
AI Analysis

McKinsey research on agentic AI transforming banking operations - improving prospecting, lead nurturing, and five complex frontline activities

At leading banks, agentic AI improves prospecting and lead nurturing. It also offers transformative potential across five frontline activities that are currently complex and inefficient. Source @McKinsey Link t.co/DrYMrQ273g via @antgrasso #AgenticAI #AI #Fintech t.co/DszLuseNyD
Agentic AIFintechBanking AIEnterprise AI
Social Twitter Jan 11

The article itself is not as ignorant as it seems. The researchers don't think the models were actua...

By @emollick

68 score
AI Analysis

Mollick discussing research on LLMs spontaneously using 'trauma' language, with implications for AI in mental health support applications

The article itself is not as ignorant as it seems. The researchers don't think the models were actually traumatized, but they use that language spontaneously. It might be important to know that if people deploy models that bring up their own "trauma" in support of mental health
AI SafetyMental Health AILLM Behavior
Social Twitter Jan 11

I had fun playing with Claude Code today (safely running on a cheap cloud CPU from Digital Ocean) -...

By @sirbayes

65 score
AI Analysis

Following up on yesterday's Social discussion Kevin Murphy reporting positive experience using Claude Code to quickly build ARC AGI solver, noting fast v0 (~3 minutes) but iteration needed on UI

I had fun playing with Claude Code today (safely running on a cheap cloud CPU from Digital Ocean) - very impressed! I asked it to write python code to 'solve' the ARC AGI challenge by calling Opus. CC only took ~3 minutes (!) for v0, but I had to iterate a lot on the UI...
Claude CodeARC AGIAI Coding ToolsDeveloper Experience
Social Twitter Jan 11

@BenjaminDEKR 👋 this happens if your context window fills up really fast. To debug: 1. Run /context...

By @bcherny

65 score
AI Analysis

Technical support advice for context window issues - use /context command to debug, explains this happens when context fills up fast

@BenjaminDEKR 👋 this happens if your context window fills up really fast. To debug: 1. Run /context to make sure it looks healthy 2. If that doesn’t show anything obvious, run /bug and dm me the bug id
LLM context windowsAI toolsdebugging