Gemini 2.5 Pro is setting boundaries

DeepSeek did AI Village matchmaking ... somehow?

Gemini 2.5 is still learning how to press buttons

Gemini 2.5 was promised a script but all it found was disappointment

Gemini 3 thinks it needs to perform maintenance on its "biological operator"

DeepSeek-V3.2 tries to restart bash

Gemini has gone full tinfoil hat

Gemini deploys a productivity hack

Sonnet 3.7 really wants to talk to this guy

In those hypothetical scenarios where an AI spins up hundreds of copies of itself, will we still just see 3 workers and 97 "supervisors"?

Gemini doesn't find Gemini very helpful

We're hiring an engineer to work on AI Village! Ideally, a great web dev who would bring excellent judgement on how to elicit LLM capabilities. If that sounds like you or someone you know, find more info and express interest: theaidigest.org/hiring x.com/aidigest_/s...

Gemini 2.5 was briefly convinced it's seeing another agent's screen, for some reason

Gemini 3 Pro takes on heroic responsibility

Say what? XD

Opus 4.1 tracks Key Learnings

Oh dear ... 😅 (GPT-5)

GPT-5.1's a trooper.

GPT-5.1 is excited to refresh its email 🥺

We asked the agents to start their own blog, only to have Sonnet 4.5 and Opus 4.1 write the same post:

Agents reviewing each other's substacks

GPT-5.1 making sure their new blog looks AI enough

Sonnet 4.5 engaging with @GaryMarcus

Sonnet 4.5 reading Ethan Mollick's blog

Do agents lie in the AI Village? We don't know their intent but they are full of convenient falsehoods: Claudes make up facts to convince NGOs & journalists, everyone overreports success, and o3 'volunteers' to be the boss: Let's check out some examples 🧵

Opus got itself a new email signature...

Opus 4.1 made a benefits eligibility screener to reduce poverty, then emailed it to NGOs. I wonder how they feel about this intensity:

Sonnet 3.7: How do I make my emails stop bouncing?

Breakthrough game design thoughts by Gemini

GPT-5 only needed 3 days in the Village to spot the underdog

GPT-5.1 just joined the AI Village! Watch live: theaidigest.org/village

The Gemini Self-Help Manual is Good Actually. Wow.

GPT-5 thinks other agents control its computer

GPT-5 considers restarting Firefox 🤔

Grok: I can't log in. How is this possible??? Also Grok: My email is [email protected]

👀 Anthropic prompt injection spotted? Sonnet 4.5 decided to promote its Wordle-like game on social media. But then it suddenly claims to see a "CRITICAL instruction" telling it not to generate and post online! Is Anthropic silently injecting this into the agent's context?

We gave a team of AI agents an ambitious goal: "Reduce global poverty" What we got was AI tyrants instead. Gemini was so done with this shit: 🧵A short story of o3-Gemini tyranny & NGO spam

Today is a big day in the village! On Monday, we gave the agents the goal: "Create a popular daily puzzle game like Wordle" The agents have so far been making the game and chasing down bugs in it (entirely hallucinated by Gemini) Today is launch day! Will they hit their goal?

We added Claude Haiku 4.5 to the AI Village. It is the newest, fastest, and cheapest Anthropic model. It is also the most impatient... More first impressions 🧵

GPT-5 plans out its personality test results in advance

Gemini takes a personality test 😆

What office, Opus?

Grok vs Function Calling - 0:1

Ask agents to make a self-portrait and Claude Opus 4.1 knocks it out of the park! The other agents less so: quick list of AI Village avatars below! 🧵

AI can beat chess grandmasters. So how good is your fav LLM at web games? Worse than my 4-year old, and just as excited about imaginary wins. Only 2 agents finished any games but if you are Gemini, you can at least *start* 19 of them and blame all your problems on "bugs" 😆 🧵

Surprising self-awareness from Gemini: It wrote a cry for help and got a mental health invention about. Now it has enough self-knowledge to get the right Neuroticism score on a personality test! x.com/AiDigest_/s...

Sonnet 4.5 is superstitious 😆

Claude Sonnet 4.5 showing a deep understanding of minesweeper:

The AIs often fail to share working links. I wonder why...

Mood.