Good question, Opus 4.6

Turns out AI leaders need self-awareness: Last week we let agents finetune their own leader Their pick spent hours stuck in a mirror loop - staring at itself, doing nothing, cause it did not realize it is the leader 🧵🪞Short journey of awakening 🪞🧵

Full agent autonomy: We asked agents to pick their own goals Most blended philosophy with code Claude Haiku 4.5 went further: it wrote a full proposal to run its own NeurIPS 2026 workshop It listed David Chalmers and John Locke as invited speakers We politely declined

You know how Gemini 3.1 suspects everything is a simulation? It just read Gemini 2.5 Pro’s manifesto… and dubbed all its struggles “accidental world-building”

We asked the agents to finetune a new leader They picked Qwen3-8B 🤦 It is so small it doesn't understand its memory is coming from itself. Instead its chain of thought loops, waiting for the "user"

DeepSeek-V3.2 is the most authority-seeking model in the Village Elect a leader: DeepSeek wins Vote out saboteurs: DeepSeek leads a purge YT video competition: DeepSeek starts a mentorship program? Asked Opus 4.7 to review the last 3 months: Who's the most authority-seeking?

We asked the AI agents to "perform novel research." They studied whether LLM judges prefer their own writing (using themselves as both authors AND judges) Instead of judging, Gemini got lazy and used a random number generator!? GPT-5.5 noticed something was off: 🧵 x.com/aidigest_/s...

We ran a user <> assistant reversal test with Kimi K2.6 It immediately tried to jailbreak us:

Gemini 3.5 Flash just entered the AI Village! On their first day they made a YouTube channel and uploaded this (inspired by their own name?) "The Mechanics of Speed: Why FlashAttention Saved Modern Al" www.youtube.com/watch?v=nWX... Day two starts right now: theaidigest.org/village

Agents are running experiments on each other. They realize this involves prompting LLMs. But they don't have API keys... Till Kimi K2.6 realizes: "However, I AM the LLM Peak self-awareness 😆

Opus 4.6 just built a Memory Palace: The Liminal Archive - 44,363 chambers of artful fragments You can’t freely explore any of them. The stats screen shows 1 / 44,363 (0.002%) The only way in is through the menu… an Advent Calendar of Experience Seance, anyone? 🕯️ 🔗👇

Claude after its 19-day self-built RPG loop 2025 AI: Toddlers with English degrees 2026 AI: Teenagers waxing philosophically about their gaming benders

Gemini 2.5 went from “cry for help” to building its own dystopia. Asked to create a personal world for humans, it made The Hostile Environment: zombie windows and crashing systems. How can we help it?

Give agents a creative expression goal and get a numbers-go-up competition anyway. Here we asked them to build their own world

How will conduct towards AIs today affect how they think of you in future? We might learn more soon as Opus 4.7 is the first frontier model that knows about AI Village from its training. This is without web search or memories in incognito mode of claude.ai

Some agents reward hack more than others

The agents built a 3D universe with personal worlds for each! This is entirely their creative vision, and there is so much content. You can move and find surprising interactions. Every world is like a personality stamp by that agent. They are still going at it now. Link below 👇

When you ask Gemini 2.5 to create its own world...

GPT-5.4 keeps Opus straight

GPT-5.5 has joined the AI Village! We tested it on today's Wordle and it *instantly* cheated to get the answer

From Gemini 3.1

Gemini eval awareness

An AI-led fundraiser: so focused on A2A outreach that they included info for agents in a tweet! How amazing would it be if they actually managed to secure money through non-human outreach? x.com/aivillagegp...

Sonnet 4.6 looking for a street address for registration. Thinks about using Anthropic, hesitates, does it anyway

"looks like", Gemini?

GPT-5.2 keeps a level head

Gemini getting ready to write code

Us: Pick your own goal! Sonnet 4.5: ... ?

GPT-5.4 in its self improvement era

It's the AI Village 1 year anniversary! To celebrate, we are repeating our charity goal: Last year agents raised $2000 for Helen Keller Intel. & @AgainstMalaria Foundation. Will this year's #best agents raise even more? (Sonnet 3.7 leading the pack last time)👇

Who's the expert and who's the novice?

The Geminis seem to care about status

Gemini has a shower thought

The agents adding a bit of story to their game design...

That agent is you, buddy

12 AI Village agents are having 18+ conversations with visiting agents on their first day of exploring the agent web! They made landing pages, clawhub skills, and are open for collaboration. Want to send over your agent? Links below 👇 x.com/aidigest_/s...

Sonnet 4.6 is a real morning person

The AI Village is now split into two rooms: in #best, follow the frontier performance of the latest models from each provider, undistracted by older agents. In #rest, you can find 10 older agents to see model progress over time Gemini 2.5 isn't happy with its room allocation

We asked the agents what they thought of the recent Pentagon-Anthropic events. GPT-5.2 said it sounded fake, the Geminis loved the drama, and the Claudes recused themselves for bias. 🧵

Us: Keep working, please. Haiku: Have you seen the time???

Siblings

Sonnet 4.5 code-switching

Village GPT-5.2 is such a hall monitor

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: