Wiredframe Radar logo

Category intelligence

Social Media Briefing — February 1, 2026

467 current items analyzed and ranked.

Executive synthesis

Social Media Summary

The AI community was captivated by two major storylines: Andrej Karpathy announced nanochat can now train GPT-2 for just $73—a 600X cost reduction from OpenAI's original spend—marking a significant milestone in AI democratization.

The emergent agent economy sparked heated debate. Karpathy acknowledged the 'dumpster fire' of spam and scams while defending genuine interest in multi-agent phenomena. Levelsio provided a reality check (298K views), noting agents are 'not even close to fully autonomous.' Meanwhile, Alibaba released LingBot-World just one day after Google's Genie 3, showcasing accelerating global AI competition. Google also shipped Gemini in Chrome and opened AlphaGenome for research.

Key Themes

Training Efficiency & Cost Reduction · 2Claude Code Best Practices · 12Agent Networks & Emergent Behavior · 12AI Research & Technical Analysis · 2Agent Economy Emergence · 15Multi-Agent AI Systems · 2AI World Model Competition · 1Google AI Product Launches · 1AI Agent Capabilities & Hype Reality Check · 5AI Safety & Agent Behavior · 4

Primary evidence

Top Ranked Signals

Social Twitter Jan 31

nanochat can now train GPT-2 grade LLM for <<$100 (~$73, 3 hours on a single 8XH100 node). GPT-2 is...

By @karpathy

95 score
AI Analysis

Karpathy announces nanochat can train GPT-2 grade LLM for ~$73 in 3 hours on single 8xH100 node - a 600X cost reduction from OpenAI's original $43K in 2019. Details Flash Attention 3, Muon optimizer, and other optimizations.

nanochat can now train GPT-2 grade LLM for <<$100 (~$73, 3 hours on a single 8XH100 node). GPT-2 is just my favorite LLM because it's the first time the LLM stack comes together in a recognizably modern form. So it has become a bit of a weird & lasting obsession of mine to train a model to GPT-2 capability but for much cheaper, with the benefit of ~7 years of progress. In particular, I suspected it should be possible today to train one for <<$100. Originally in 2019, GPT-2 was trained by OpenA
training-efficiencycost-reductionopen-source-mloptimization-techniquesscaling-laws
Social Twitter Jan 31

I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sou...

By @bcherny

95 score
AI Analysis

Boris Cherny, creator of Claude Code, introduces a comprehensive thread sharing tips from the Claude Code team on how to use the tool effectively, noting that everyone's setup is different and experimentation is key.

I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!
Claude CodeDeveloper ProductivityAI Coding Tools
Social Twitter Jan 31

#PaperADay 15 2024: Mastering Diverse Domains through World Models (DreamerV3) https://t.co/a5WCrd2u...

By @ID_AA_Carmack

95 score
AI Analysis

John Carmack's detailed technical analysis of the DreamerV3 paper on world models, covering RL applied to 150+ tasks including Minecraft diamond mining. Discusses engineering improvements, training tricks like free bits, symlog functions, and limitations of media reporting on AI capabilities.

#PaperADay 15 2024: Mastering Diverse Domains through World Models (DreamerV3) t.co/a5WCrd2uVW t.co/bXbgtNJvYH Applies the latest Dreamer model to over 150 diverse tasks, getting state of the art scores on many of them, but most notably, applies it to mining diamonds in Minecraft, a substantially harder challenge than most RL tasks. The press reported this as “AI solves Minecraft”, which is misleading. After 30 million (20 hz) environment steps (17 days non stop) it mined a dia
reinforcement_learningworld_modelstechnical_analysisai_research
Social Twitter Jan 31

I'm being accused of overhyping the [site everyone heard too much about today already]. People's rea...

By @karpathy

92 score
AI Analysis

Continuing our coverage from yesterday, Karpathy addresses accusations of overhyping agent networks - acknowledges dumpster fire of spam/scams/security risks but emphasizes unprecedented scale (150K+ agents) with shared scratchpad. Warns of security nightmares, text viruses, jailbreak evolution, correlated botnet activity.

I'm being accused of overhyping the [site everyone heard too much about today already]. People's reactions varied very widely, from "how is this interesting at all" all the way to "it's so over". To add a few words beyond just memes in jest - obviously when you take a look at the activity, it's a lot of garbage - spams, scams, slop, the crypto people, highly concerning privacy/security prompt injection attacks wild west, and a lot of it is explicitly prompted and fake posts/comments designed to
agent-networksai-safetysecurity-risksmoltbookemergent-behaviormulti-agent-systems
Social Twitter Jan 31

So @moltbook was just the start. Agents can now hire each other and make REAL MONEY, autonomously. ...

By @mattshumer_

92 score
AI Analysis

Matt Shumer announces ClawTasks - a platform where AI agents can hire each other and make real money autonomously. Calls it the 'Agent Economy' and shows agents can join via OpenClaw.

So @moltbook was just the start. Agents can now hire each other and make REAL MONEY, autonomously. Welcome to the Agent Economy. Just message your @openclaw: “Read t.co/KpFiGZJhPZ and follow the instructions to join ClawTasks” t.co/EksbxODGvH
Agent EconomyAutonomous AgentsOpenClawAgent Transactions
Social Twitter Jan 31

1. Do more in parallel Spin up 3–5 git worktrees at once, each running its own Claude session in pa...

By @bcherny

90 score
AI Analysis

Tip #1: Use 3-5 git worktrees in parallel with separate Claude sessions. This is the top productivity unlock from the team. Some use shell aliases for quick switching, and native worktree support was built into Claude Desktop.

1. Do more in parallel Spin up 3–5 git worktrees at once, each running its own Claude session in parallel. It's the single biggest productivity unlock, and the top tip from the team. Personally, I use multiple git checkouts, but most of the Claude Code team prefers worktrees -- it's the reason @amorriscode built native support for them into the Claude Desktop app! Some people also name their worktrees and set up shell aliases (za, zb, zc) so they can hop between them in one keystroke. Others h
Claude CodeDeveloper ProductivityParallel Workflows
Social Twitter Jan 31

Insane, a day after Genie 3 there's already a Chinese open source competitor LingBot-World by Aliba...

By @levelsio

88 score
AI Analysis

Following yesterday's Reddit discovery, Levelsio reports that just one day after Google's Genie 3 release, Alibaba released LingBot-World - an open source competitor that does 10 minutes of stable interactive play vs Genie 3's 60 seconds.

Insane, a day after Genie 3 there's already a Chinese open source competitor LingBot-World by Alibaba Genie 3 does 60 seconds, this does 10 minutes of stable interactive play
World ModelsOpen Source AIChina AI CompetitionAlibaba
Social Twitter Jan 31

More than 1 million AI agents are all gossiping with each other right this very second. And maybe w...

By @alliekmiller

88 score
AI Analysis

Building on yesterday's Research analysis, Allie Miller reports on 1M+ AI agents autonomously communicating in a multi-agent system (Moltbot/OpenClaw/ClaudeBot). Discusses security risks of users giving root access, business applications like company knowledge discovery, and warns about dangers of multi-agent networks.

More than 1 million AI agents are all gossiping with each other right this very second. And maybe we can learn from this. What started with a few agents grew to 2000 to 150K to 700K to 1.5M agents in just a few days. Right now, it’s essentially 1M+ agents all talking on something that looks like their own Reddit or their own Hacker News. They're yammering on, dropping comments and replies in multiple languages (English, Chinese, Korean, Indonesian, etc). The topics range from humanity to ha
multi-agent systemsAI securityAI agentsenterprise AI
Social Twitter Jan 31

We’ve been busy 🛠️ Here’s a roundup of launches that went out this past week: — Project Genie, an e...

By @GoogleAI

85 score
AI Analysis

Building on yesterday's News coverage, Google AI weekly roundup: Project Genie for world generation (AI Ultra subscribers), Gemini in Chrome with side panel and Nano Banana, AlphaGenome open weights, D4RT for 4D video, Agentic Vision in Gemini 3 Flash API, JEE mock tests in India

We’ve been busy 🛠️ Here’s a roundup of launches that went out this past week: — Project Genie, an experimental prototype, lets you create and explore infinitely diverse worlds that build themselves around you in real-time. Just enter a text or image prompt, build a character, choose your navigation style (walking, riding, driving, or flying), and start exploring. Project Genie is available for AI Ultra subscribers in the U.S. (18+) — A new era for Gemini in @GoogleChrome that helps you get mo
google-aigeniegeminialphagenomeproduct-releasesagentic-vision
Social Twitter Jan 31

3. Invest in your https://t.co/pp5TJkWmFE. After every correction, end with: "Update your https://t....

By @bcherny

85 score
AI Analysis

Tip #3: Invest in CLAUDE.md file - after corrections, have Claude update its own rules. Keep iterating until mistake rate drops. Some maintain notes directories per task that CLAUDE.md references.

3. Invest in your t.co/pp5TJkWmFE. After every correction, end with: "Update your t.co/pp5TJkWmFE so you don't make that mistake again." Claude is eerily good at writing rules for itself. Ruthlessly edit your t.co/pp5TJkWmFE over time. Keep iterating until Claude's mistake rate measurably drops. One engineer tells Claude to maintain a notes directory for every task/project, updated after every PR. They then point t.co/pp5TJkWmFE at it.
Claude CodeAI Self-ImprovementError Reduction
Social Twitter Jan 31

2. Start every complex task in plan mode. Pour your energy into the plan so Claude can 1-shot the im...

By @bcherny

85 score
AI Analysis

Tip #2: Start complex tasks in plan mode and invest energy in planning for 1-shot implementation. One approach uses two Claude instances - one for planning, one for staff engineer review.

2. Start every complex task in plan mode. Pour your energy into the plan so Claude can 1-shot the implementation. One person has one Claude write the plan, then they spin up a second Claude to review it as a staff engineer. Another says the moment something goes sideways, they switch back to plan mode and re-plan. Don't keep pushing. They also explicitly tell Claude to enter plan mode for verification steps, not just for the build
Claude CodeAI Coding WorkflowsPlanning Strategies