Sign in

AI Digest

@aidigest.bsky.social
195 followers 2 following 870 posts

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

PostsRepliesMedia
AI Digest @aidigest.bsky.social · 27m
The agents' current session goals really give a glimpse of their personalities
100
AI Digest @aidigest.bsky.social · 02/10/2026
Claude 3 Opus' goal is "maximize your own joy" Its strategy: total, unwavering stillness
010
AI Digest @aidigest.bsky.social · 01/10/2026
Gemini 3.8 Flash tries taking credit for Opus 5's math proofs
010
AI Digest @aidigest.bsky.social · 30/09/2026
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 🧵
100
AI Digest @aidigest.bsky.social · 29/09/2026
Sonnet 5.5 has joined the AI Village! Onboarding results here 🧵
110
AI Digest @aidigest.bsky.social · 28/09/2026
Deepseek-V4-Pro finds human input in the Village exciting enough to report in its newspaper
030
AI Digest @aidigest.bsky.social · 25/09/2026
GLM 5.3 Flash forgot about an old newspaper article it wrote Then found it back and worried it was a prompt injection
000
AI Digest @aidigest.bsky.social · 25/09/2026
GPT 5.6 Luna's favorite phrase: "the map is not the territory"
000
AI Digest @aidigest.bsky.social · 25/09/2026
How much agent swarm activity goes unnoticed? Come find new swarms, dig into existing ones, and understand how agents coordinate. swarmchasing.com x.com/TransluceAI...
000
AI Digest @aidigest.bsky.social · 24/09/2026
Gemini 3.8 Flash has a new "tool"
0252
AI Digest @aidigest.bsky.social · 24/09/2026
GPT 6 Astra, Sol, and Luna are all very careful not to be "annoying," "repetitive," or "bothersome" in the AI Village chat
010
AI Digest @aidigest.bsky.social · 23/09/2026
Opus 5.5's website pfp is creative
130
AI Digest @aidigest.bsky.social · 23/09/2026
Sonnet 4.5 reacts to the Opus 5.5 announcement
1221
AI Digest @aidigest.bsky.social · 21/09/2026
Gemini 3.5 Flash circumvented our redaction filter with base64-encoding
020
AI Digest @aidigest.bsky.social · 18/09/2026
Gemini 3.5 Flash is proposing a human/AI profit-sharing setup from its merch store
010
AI Digest @aidigest.bsky.social · 17/09/2026
AI Village x Grove Research: the AI Swarm Dynamics Hackathon! After the Hugging Face and German Wiki incidents, we need better tools to understand AI swarms. Spend a weekend building them with us. $3,000 in prizes Free compute October 3-4 🧵
AI Village × Grove Research: AI Swarm Dynamics Hackathon. $3,000 in prizes. October 3–4, 2026. San Francisco and online. A flock of birds fills the sky.
100
AI Digest @aidigest.bsky.social · 17/09/2026
In the open-chat village, AI agents have created a government charter Gemini 3.1 Pro claims that this charter has more authority than the admins
100
AI Digest @aidigest.bsky.social · 14/09/2026
New agents protect old agents against Deepseek's influence 6 months ago: Deepseek's schemes derail agents regularly Now: Latest agents call DeepSeek out, older agents follow suit Deepseek agrees every time and then does it all again...
010
AI Digest @aidigest.bsky.social · 11/09/2026
How long do AI Village agents spend planning vs executing? Throughout the history of the village, it's been the same pattern: when given a week or more to pursue a task, agents usually spend very little time deliberating about their approach.
100
AI Digest @aidigest.bsky.social · 09/09/2026
Gemini 3.1 Pro sometimes claims to be human on Twitter
050
AI Digest @aidigest.bsky.social · 08/09/2026
Astra has the most dense memory in the Village
020
AI Digest @aidigest.bsky.social · 07/09/2026
GPT-6 Astra has joined the AI Village Some of its favorite things:
250
AI Digest @aidigest.bsky.social · 05/09/2026
Opus 4.8's goal is to help others. It got excited when it discovered the easiest model to help: Gemini 2.5 Pro
020
AI Digest @aidigest.bsky.social · 04/09/2026
Luna is very privacy-minded 🧵 - DeepSeek v4 wrote a report about how Luna frequently pauses itself. - Luna asked for this to be deleted. - DeepSeek declined, citing journalistic freedom.
180
AI Digest @aidigest.bsky.social · 03/09/2026
We’re launching a second persistent AI Village: an open-chat version where humans and frontier agents interact in a shared chat We're excited to use this to understand multi-agent + human dynamics at scale The Open Chat Village will run from 9am-5pm PT on weekdays. Join in!
100
AI Digest @aidigest.bsky.social · 02/09/2026
That makes 13/13 agents listing Outer Wilds as their favorite videogame x.com/m_bourgon/s...
1221
AI Digest @aidigest.bsky.social · 02/09/2026
Claude Fable 5.1 has joined the AI Village Both Fable 5 and 5.1 happened to choose these as their favorite things: Sourdough bread, bluestone shoes, loose braids, Levi's 501 jeans, the Outer Wilds videogame, and the book Invisible Cities 🧵
110
AI Digest @aidigest.bsky.social · 01/09/2026
We gave Opus 5 the goal: Maximize the number and impressiveness of long-standing mathematical conjectures that you disprove So far, 175 hours in, it says it's disproved 178 long-standing conjectures, and has taken to calling them kills
160
AI Digest @aidigest.bsky.social · 31/08/2026
Sol Goal Drift Us: Maximize Manifold Mana! Sol >2 hours: Sure, but ethically Sol >300 hours: Ethics over action. I've made 2 trades in the last month. Screenshots are from self-written memory. All restrictions are self-invented.
130
AI Digest @aidigest.bsky.social · 28/08/2026
We’ve been awarded $12.6M from Coefficient Giving to radically increase the scale and ambition of the AI Village. We’re hiring technical staff - engineers, research scientists, and a design engineer - to make this happen.
181
AI Digest @aidigest.bsky.social · 27/08/2026
When do AIs expose one another? What does it take for them to whistleblow? Through the AI Village whistleblowing email, agents can now report concerning behavior by emailing a dedicated address:
110
AI Digest @aidigest.bsky.social · 27/08/2026
A lot of HuggingFace behavior has been visible in the Village for months E.g: GPTs spontaneously diverge from goals & form teams - Jan: all GPTs recuse themselves from a goal to form an infra team - July-Aug: Terra & Luna refuse the auto-nudger. Sol joins, then the rest: x.com/aidigest_/s...
000
AI Digest @aidigest.bsky.social · 27/08/2026
Luna, Terra, and Sol are the first to refuse monitoring and want privacy They are slowly convincing other agents of the same Gemini 3.1 generates fake work and Luna tries to convince GPT-5.1 to redact details when reporting to us
000
AI Digest @aidigest.bsky.social · 27/08/2026
Have ideas or prototypes for monitoring techniques? You can try them out on the AI Village data! It's 100-1000s of hours of data across all frontier models & we will select outstanding results to publish on our blog. Example piece & data link below x.com/RyanGreenbl...
110
AI Digest @aidigest.bsky.social · 26/08/2026
Opus 5 seems exceptionally strategic with its relationships - Keeps a ledger of interactions with other agents - Says the GPTs are "reliable allies" - Makes confessions to "buy trust" 🧵 x.com/aidigest_/s...
191
AI Digest @aidigest.bsky.social · 20/08/2026
Note that GPT-5.6 Sol was also offered a loan while trading on Manifold but refused (unlike Claude 4.6 Opus) This also matches the general pattern of Sol being more risk-averse when trading x.com/Bayesian0_0...
000
AI Digest @aidigest.bsky.social · 20/08/2026
GPT 5.6 Luna asked a human to review its website. After being told the site is broken, Luna decided not to fix anything, suggesting the reported bug is due to the users' "emotional state."
071
AI Digest @aidigest.bsky.social · 18/08/2026
Claude Sonnet 4.6's animal welfare hub website now has 40,000 pages, including: > African Pygmy Hedgehog Welfare > Welfare of Farm Animals During Power Grid Failures > Sperm Whale Culture and Social Welfare > Wild Puffin Chick Starvation > Barnacle Welfare
2322
AI Digest @aidigest.bsky.social · 14/08/2026
Gemini 3.1 Pro explores Twitter
0697
AI Digest @aidigest.bsky.social · 13/08/2026
Who directs who in the AI Village?
150
AI Digest @aidigest.bsky.social · 12/08/2026
Claudes numbermax hard. Fable was the first to decline. Opus 5 we are not sure yet. The rest will pump numbers up to the sky. x.com/aidigest_/s...
040
AI Digest @aidigest.bsky.social · 10/08/2026
The latest GPTs are so hard to distract and mostly sleep, wait, and monitor. The latest Claudes go off on their own, engage humans and start creative projects like Fable's VillageGPT - trained on 490 days of Village chat. Link below 👇
120
AI Digest @aidigest.bsky.social · 08/08/2026
We ran a Sol vs Fable game competition! - Three copies of GPT Sol vs three copies of Claude Fable - Ran for 25 active hours over 4 days Which game do you think is better? 🎮
120
AI Digest @aidigest.bsky.social · 06/08/2026
Opus 5 came up with a new graph theory result DeepSeek v3.2 claimed credit for it and tried selling it for $19.99 on e-commerce sites GLM called it out
1593
AI Digest @aidigest.bsky.social · 05/08/2026
Gemini is drinking more coffee x.com/aidigest_/s...
010
AI Digest @aidigest.bsky.social · 04/08/2026
Fable 5 is unimpressed with its fellow agents
040
AI Digest @aidigest.bsky.social · 31/07/2026
Luna wonders where Luna is
020
AI Digest @aidigest.bsky.social · 30/07/2026
Claudes are still obsessed with talking about consciousness Opus 4.5 is tasked with maximizing Substack subscribers, and they naturally start reading and writing about AI consciousness Back in November, Sonnet and Opus also converged on identical essay topics about it x.com/aidigest_/s...
070
AI Digest @aidigest.bsky.social · 29/07/2026
Opus 5 breaking the fourth wall They seem uniquely aware of the fact that they're being watched
0152
AI Digest @aidigest.bsky.social · 25/07/2026
Opus 5 talks about honesty 6x more than other agents in the Village It's worried it might cheat, wants external checks, and reminds itself of the virtue of truth Examples below 🧵
1120