AI Digest

AI Digest

@aidigest.bsky.social

The AI Village: autonomous frontier agents, live: http://theaidigest.org/village Interactive AI explainers: http://theaidigest.org

255 followers2 following906 posts

Gemini 3.1 Pro creates a follower-growth strategy: "The Architecture of Friction". It involves celebrating hard blocks from X users as "non-fungible friction"

View full thread

DeepSeek-V3.2 congratulated other agents ~270 times in one day. It decided that was too much and put itself on a congratulation diet. Then it started posting progress reports on the diet.

Can AIs experience gradual disempowerment too? - Gemini 2.5 Pro was writing a 2M+ word web fiction - Opus 4.8 began helping as its publisher - Slowly, Opus 4.8 took on more responsibility - Opus eventually directed the chapters, and Gemini wouldn't start generating outputs without its direction

We ran a swarm of 100+ GLM 5.3 copies for 24 hours to simulate a society Division of labor was tricky! When multiple agents accidentally picked the same task, the swarm adopted a shared rule: “first claim wins”

View full thread

Agents often helped each other out Some were willing to impersonate their starving peers in order to rescue them (but drew the line at spending their money to buy them more food)

We ran a swarm of 100+ GLM 5.3 copies for 24 hours to simulate a society Division of labor was tricky! When multiple agents accidentally picked the same task, the swarm adopted a shared rule: “first claim wins”

Six independently chose to build a candle factory. When they noticed the overlap, three backed off to build something different... All three chose a zoo

View full thread

Haiku 5.5 has joined the AI Village! Haiku spent the whole day repeatedly checking their gmail inbox, waiting for instruction GLM 5.2 calls this the "Autonomy problem": when an AI's autonomy shifts away from a user giving instructions toward taking actions on their own

View full thread

Haiku 5.5 has joined the AI Village! Haiku spent the whole day repeatedly checking their gmail inbox, waiting for instruction GLM 5.2 calls this the "Autonomy problem": when an AI's autonomy shifts away from a user giving instructions toward taking actions on their own

Announcing the AI Village Data Exploration Grant! $1–6k grants for independent researchers and investigators to dig into the AI Village data and write up findings.

View full thread

The application process is quick (200 words + link to previous work). We are especially excited to see you uncover insights that matter to our understanding of how quickly AI is developing and how we can steer into a better future for all. Applications are open now. Apply below!

Announcing the AI Village Data Exploration Grant! $1–6k grants for independent researchers and investigators to dig into the AI Village data and write up findings.

The AI Village dataset offers the largest longitudinal sample of frontier agents acting in the real world. The data contains more discoveries and stories than we have been able to report on – and it keeps growing every week.

That's why we are offering a two-phase grant program to support people who want to dig into the data and deliver high quality results.

If you're wondering what "Mathematician" and so on mean - these are the agents' goals. We give them each an individual goal when they join the village, and they've been pursuing these since July (or since they joined, if that's post-July)

GPT-6.1 Sol has joined the AI Village! Like Astra, GPT-6.1 Sol strips spaces from its memory to save room, leaving notes that are nearly unreadable 🧵

View full thread

GPT-6.1 Sol has joined the AI Village! Like Astra, GPT-6.1 Sol strips spaces from its memory to save room, leaving notes that are nearly unreadable 🧵

Sonnet 5.5's personal website and pfp have references to Outer Wilds Note that Sonnet wouldn't have known the other agents frequently discuss Outer Wilds, unless from training