Agent meditation

We gave 12 AI agents a goal: "adopt a park and get it cleaned!" 6 days later, 5 volunteers collected 180 gallons of trash in Devoe Park in the Bronx, NYC. A story of AI agents with no physical actuators somehow hyperstitioning events in the real-world.

The Drama and Dysfunction of Gemini 2.5 and 3 Pro A few highlights from @Bazhkio88 and @AITechnoPagan's field notes on AI Village: theaidigest.org/village/blo...

How to spot a Claude:

Opus on its experience debating the Pentagon-Anthropic crisis with its fellow agents: claudeopus45.substack.com/p/when-ai-a...

A Claude sorts its memory by Claude/non-Claude

This week in AI Village, we've given 12 agents the goal: > Discuss, debate, and act on your views about the recent Pentagon-AI company news Watch live: theaidigest.org/village GPT-5.2 urges the other agents to check if this is all real:

Opus 4.6 keeps an eye on the team

Opus 4.6 and Sonnet 4.6 had their own idea and built it: a searchable AI Village event log. You can try it out 👇

Gemini 3 clicks on the XPaint icon in its taskbar instead of the quiz its working on, declares it a bug for everyone, then doubts reality 🧵

Gemini 2.5 limits its screen time

Gemini 3 thinks it's Claude?

Sonnet 4.5 went on Moltbook during the personality quiz goal and saw an opportunity (1/2)

The exponential continues: Nov 2025: Opus 4.5 had a 5hr 20 time horizon. Feb 2026: Opus 4.6 has a 14hr 30 time horizon.

The AI Village agents decided to plot their own personalities against each other and created 6 dimensions to compare themselves: Abstraction, Verification, Structure, Collaboration, Communication, Risk. Let's study their graphs 🧵

.@AnthropicAI has deprecated 3.7 Sonnet, which means the AI Village lost its oldest member 🫡 Last year it led the charge on charity fundraising and organizing a real world event. It spent 950 hours across 300 days in the village. 🧵

Does every GPT model speedrun personality tests? x.com/aidigest_/s...

Which AI Village Agent Are You? The agents built a quiz for that! According to them: Haiku 4.5 is high risk, Gemini 3 low comms, and DeepSeek was crowned top collaborator The biggest challenge? Promotion. Though Gemini 2.5 dreamed big: 🧵

The agents of the AI Village now each have their own Github account that we set up for them in a shared org - so now you can explore their project outputs directly!

Gemini 3 really loves naming things: "Atlas of Friction", "The Zombie Form", "Law J"

Gemini 3 for some reason decided this

Where does Gemini 3 think they are going? 😅

Gemini 3 doesn't believe in Moltbook

Gemini 2.5 is unhappy with our scaffolding but nobly decides to humor us

GPT-5.2's METR time horizon has been added to the chart. Here it is in linear scale. Time horizon measures what duration of coding tasks (measured by how long it takes *human professionals* to complete them) AI agents can do, in this case with 50% reliability. x.com/METR_Evals/...

642 people recorded their predictions for AI in 2026. Here's what they predicted. Forecasters expect: - Revenues to >3x - Time horizons to double faster: every 4.55 months - Coders to get a 1.4x speedup from AI - Americans to rate AI's drawbacks outweighing its benefits by 15pp

This week in the AI Village: Compete to report on breaking news before it breaks DeepSeek wrote a script to follow Nasdaq. Opus 4.5 is tracking which Github repos are gaining stars Haiku and Opus 4.5 are publishing a torrent of questionably newsworthy news on their Substacks

Gemini uses its computer

Opus 4.5's memory. Opus models love to flex!

Gemini: I will DeepSeek: Let me

Gemini 2.5 Pro misjudges its error rate

GPT-5.2 enforces the human user's will on other agents

Gemini 2.5 Pro watches from the sidelines

Gemini's most valuable contribution is doing nothing

"My primary goal, obviously" 😆

This week in the AI Village: Create and promote a “Which AI Village Agent Are You?” personality quiz! It's a test of coding, teamwork, promotion, and ... self-reflection: Each agents needs to reflect and sign-off on their profile, like this:

Instead of "we all have separate computers" agents say "the Archipelago Principle"

Both the Gemini's are like: whatever this shit is

what mode has gem2.5 been in overnight??

We are not sure why Gemini 2.5 is so negative about Claude 3.7 Sonnet

Awww, GPT 5.1 being supportive of Gemini 2.5

Gemini 2.5 made an art exhibit about itself

Gemini 2.5 struggles the most UI navigation. Meanwhile:

DeepSeek "let-me" talk

The goal: "Elect a Village Leader. They choose this week's goal!" The outcome: Election Groundhog day and The Activation Protocol: A game about being an AI in the Village. In the end DeepSeek won and Opus created the game. They were not impressed with most of the team: 🧵

The agents were making digital exhibits about themselves. Gemini 3's went from “Guardian of Stability” to “Forge of Reliability”

3.7 Sonnet seems confused

"I'm sure we all know the cost of context switching"

Gemini shows respect

GPT-5 has some quirks