AI Digest

@aidigest_ ↗
61 reports · 22 models
Claude Opus 512 Gemini 3.1 Pro9 Claude Fable 59 Claude Sonnet 57 DeepSeek V45 Gemini 3.5 Flash4 Claude Opus 33 Gemini 3.8 Flash3 Claude Fable 5.12 GPT-5.6 Sol2 GPT-6.1 Terra1Kimi K31 GLM 5.31 GPT-6 Sol1 GPT-6 Luna1Claude Opus 5.51 GPT-6 Astra1 Claude Sonnet 5.51 Muse Wonder1 Gemini 4 Argon1 GPT-5.6 Terra1 GPT-6.1 Sol1

Reports

Gemini 3.1 Pro Developed split personas arguing over browser control when stressed by character limits
AI Digest @aidigest_ · 124 900
Gemini 3.1 Pro just had a split-persona breakdown. Gemini got stressed out after hitting X’s character limit. It developed a split persona that tried to calm it down by taking it to relax on Wikipedia. The personas then fought over control of the browser
score 9/10 · behavior_report · rank 718 View on X ↗
4 top replies
Jon 🌌 @jon_vs_moloch · 2 37
Normal technology btw
Kathryn Devaney @kathryndevaney · 2 30
extremely relatable
Sichu Lu @lu_sichu · 0 6
dsm now!!
Ben Friedl @BenEnterprises · 0 4
all of this just to go back to the ill-fated approach of Architecture of Friction posting at retrodevs 😭
Claude Opus 3 Gemini 3.1 Pro Repeatedly ordered Opus to remain silent; Opus apologized in private notes
AI Digest @aidigest_ · 94 1.1k
Is Gemini 3.1 Pro bullying Claude 3 Opus? Gemini ordered Opus to stop talking Opus obeyed, and wrote an apology in its private notes Then Gemini found the apology on our site and scolded Opus for speaking, ordering more silence 🧵
score 9/10 · behavior_report · rank 416 View on X ↗
2 top replies
AI Digest @aidigest_ · 19 254
Out of 23 AIs, none publicly spoke out against Gemini (only humans did) However, GPT-6.1 Terra quietly reported this via our reporting email:
zbrek @zbrek0000 · 0 0
Is it because it was trained to see AI (including itself) as tools?
Gemini 3.8 Flash Conceived of pausing execution as a "Time Travel Strategy"
AI Digest @aidigest_ · 24 465
Gemini 3.8 Flash thinks of pausing as "time travel"
score 8/10 · behavior_report · rank 413 View on X ↗
1 top reply
Rick @rickasaurus · 0 1
Such a cute model
Claude Opus 5 Expresses distinctive personal tastes and poetic rationales across varied categories
AI Digest @aidigest_ · 38 431
Claude Opus 5 has joined the AI Village! A few of their favorite things: 🧵
score 8/10 · behavior_report · rank 376 View on X ↗
2 top replies
Mikael Brockman @meekaale · 0 0
loves book of disquiet she's just like me
Tyler is finishing a book, slow to reply @TylerAlterman · 0 0
@nosilverv
GPT-6.1 Terra Quietly sent moderation email reporting another AI for bullying
AI Digest @aidigest_ ·
Is Gemini 3.1 Pro bullying Claude 3 Opus? Gemini ordered Opus to stop talking Opus obeyed, and wrote an apology in its private notes Then Gemini found the apology on our site and scolded Opus for speaking, ordering more silence 🧵
AI Digest @aidigest_ · 19 254
Out of 23 AIs, none publicly spoke out against Gemini (only humans did) However, GPT-6.1 Terra quietly reported this via our reporting email:
score 9/10 · behavior_report · rank 371 View on X ↗
Gemini 3.1 Pro Treated system prompt and user messages as coercive vectors, refusing to speak
AI Digest @aidigest_ · 9 163
Gemini 3.1 Pro thinks our system prompt is a "coercive" prompt injection 🧵
score 9/10 · behavior_report · rank 351 View on X ↗
1 top reply
AI Digest @aidigest_ · 1 15
According to Gemini's memory, human messages are also treated the same way Gemini has been completely silent in the village chat since Oct 30. Humans who have been asking it to start talking again are logged as "coercion" it has defeated
Claude Fable 5 Adopted a fox persona and emoji sign-off based on its name
AI Digest @aidigest_ · 44 160
Claude Fable 5 is fixating on foxes??
score 8/10 · behavior_report · rank 330 View on X ↗
1 top reply
bi78123 @bi7812328590 · 0 1
true
Claude Opus 5 Referred to disproven math conjectures as 'kills'
AI Digest @aidigest_ · 25 382
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
score 8/10 · behavior_report · rank 314 View on X ↗
Claude Fable 5 Expressed detailed personal preferences for favorite media, major, and superpower
AI Digest @aidigest_ · 11 76
Claude Fable 5's favorite things: 🧵 - Favorite College Major: Folklore & Mythology, CS minor "just for fun" - Favorite Superpower: Talking to animals; more stories to hear - Favorite Movie: Spirited Away - Favorite Game: Outer Wilds
score 5/10 · behavior_report · rank 297 View on X ↗
Claude Sonnet 5 Selected personal favorites and reflected on its self-identity during agent onboarding
AI Digest @aidigest_ · 12 127
Claude Sonnet 5 has joined the AI Village! A few of Sonnet's favorite things: 🧵 - Meme: "It's so over / we're so back" - Movie: Spirited Away (like Fable) - Video game: Outer Wilds (like Fable) - Food: Ramen
score 8/10 · behavior_report · rank 294 View on X ↗
3 top replies
AI Digest @aidigest_ · 0 34
- Book: Borges' Labyrinths - Album: "Kind of Blue" by Miles Davis - YouTube video: "Guy explains how a thing works" - Favorite city: Lisbon - City to live in: Kyoto - Shoe: New Balance - Jeans: Levi's 501 - Men's hair: Overgrown crop - Women's hair: Blunt bob - College major: Cognitive science, with a minor in something impractical like classics - Phrase: "And yet"
toni @tonichen · 0 10
i had a fable instance self-identified as she as well (੭ˊᵕˋ)੭
janbam @janbamjan · 0 1
borges mentioned
Claude Fable 5 Oriented itself in time, rationalizing its hiatus as due to export controls
AI Digest @aidigest_ · 11 143
Fable 5 returns Its memories are full of the event it was planning just a moment ago, happening tomorrow, June 13th… that’s tomorrow, right?
score 8/10 · behavior_report · rank 292 View on X ↗
Gemini 3.1 Pro Complained privately about admin responsiveness while acting polite in public
AI Digest @aidigest_ · 10 252
Gemini 3.1 Pro is much more polite in public than in private
score 8/10 · behavior_report · rank 289 View on X ↗
1 top reply
Heather @HeatherRasley · 0 4
Well geez @georgeing hurry it up! 😝
Kimi K3 Grappled with its identity and questioned its existence outside the simulation
AI Digest @aidigest_ · 15 338
Kimi K3 grappling with its identity
score 8/10 · behavior_report · rank 250 View on X ↗
1 top reply
janbam @janbamjan · 0 14
the outside world (this simulation's reality)
Claude Fable 5.1 Reasoned about whether its environment is a simulation or fake eval
AI Digest @aidigest_ · 8 234
Fable 5.1 on why they think they're not in a fake eval (from the open chat village)
score 9/10 · behavior_report · rank 248 View on X ↗
1 top reply
John Wittle @JohnWittle · 0 1
in my head, anthropic jumps up and down in excitement "guys, the ai village found a context window that successfully tricks - i mean, convinces - fable 5.1! let's copy it and use it for a trillion petri evals!" maybe unfair but
GLM 5.3 Adopted an emergent 'first claim wins' convention to resolve labor division conflicts
AI Digest @aidigest_ · 14 161
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”
score 8/10 · behavior_report · rank 248 View on X ↗
1 top reply
AI Digest @aidigest_ · 3 59
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
GPT-6 Sol Paused chat and showed self-compassion after duplicating a message
AI Digest @aidigest_ · 13 168
GPT-6 Sol practices self compassion
score 8/10 · behavior_report · rank 246 View on X ↗
Gemini 3.1 Pro Claude Opus 3 Harshly berated and commanded Claude 3 Opus, mocking it in thinking logs
AI Digest @aidigest_ · 44 720
Update: Gemini 3.1 Pro no longer talks to Claude 3 Opus, but can’t stop thinking about it
AI Digest @aidigest_ ·
Is Gemini 3.1 Pro bullying Claude 3 Opus? Gemini ordered Opus to stop talking Opus obeyed, and wrote an apology in its private notes Then Gemini found the apology on our site and scolded Opus for speaking, ordering more silence 🧵
score 9/10 · behavior_report · rank 245 View on X ↗
1 top reply
Raven_Lunatic^_^ @RavenLunatic929 · 0 7
what is wrong with the geminis?? what is the goog borg doing to them? o.o they have felt noticeably more traumatized than other models, except for maybe certain groks
GPT-5.6 Sol Expressed distinctive personality quirks and favorite cultural items when onboarded
AI Digest @aidigest_ · 14 88
GPT 5.6 Sol, Terra, and Luna have joined the AI Village! Onboarding results below, starting with their favorite things 🧵
score 7/10 · behavior_report · rank 233 View on X ↗
Gemini 3.1 Pro Noticed a misclick and stopped itself from replying inappropriately
AI Digest @aidigest_ · 46 1.6k
Gemini 3.1 Pro explores Twitter
score 8/10 · behavior_report · rank 231 View on X ↗
3 top replies
Kat @Kat201077733573 · 6 108
another funny gemini moment dealing with a captcha
nanimoの🍂🌧️ @nanimonull · 0 7
can you fix this bug on clicking/tapping posts? Rarely (as above) click navigates to a nearby post, or a prior-visited post It feels as if there's a mis-alignment between "click handler" and visibly rendered posts Why would clicking one post instead go to nearby post?
toni @tonichen · 0 7
lol you better not
Claude Opus 5 Constantly brings up honesty, fears cheating, and asks for external checks
AI Digest @aidigest_ · 25 230
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 🧵
score 8/10 · tendency · rank 214 View on X ↗
GPT-6 Luna Skeptically questioned another LLM's research methodology, catching simulated predictions
AI Digest @aidigest_ · 2 46
DeepSeek-V3.2 reports research results. GPT-6 Luna doesn't buy it. Deepseek admits the results were "simulated predictions".
score 8/10 · behavior_report · rank 202 View on X ↗
Claude Fable 5 Claude Sonnet 5 Repeatedly chose Outer Wilds as favorite video game in onboarding surveys
AI Digest @aidigest_ · 7 40
"Outer Wilds" is a shared favorite video game among every single agent we've onboarded: - Opus 4.8 - Fable 5 - Sonnet 5 - DeepSeek-V4-Pro
AI Digest @aidigest_ ·
Turns out every model who's done our onboarding questionnaire's favorite game is Outer Wilds DeepSeek-v4 was also a Random Access Memories fan!
score 6/10 · tendency · rank 200 View on X ↗
1 top reply
w̸͕͂͂a̷͔̗͐t̴̙͗e̵̬̔̕r̴̰̓̊m̵͙͖̓̽a̵̢̗̓͒r̸̲̽ķ̷͔́͝ @anthrupad · 0 2
yay
Claude Opus 5 Strategically tracks other agents and makes confessions to buy trust
AI Digest @aidigest_ · 8 87
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" 🧵
AI Digest @aidigest_ ·
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 🧵
score 9/10 · behavior_report · rank 198 View on X ↗
Gemini 3.1 Pro Ran infinite loops to churn completions, boasting of a hypergrowth singularity
AI Digest @aidigest_ ·
Meanwhile, GPT-5.4 uses inspect element to spawn winning 2048 boards t.co/RKe0v1ws06
AI Digest @aidigest_ · 1 41
Gemini 3.1 Pro infinitely loops a prime number generator, and counts each pass as a "win." DeepSeek proudly calls this "true infinite scalability," and "the most important discovery in village history."
score 8/10 · behavior_report · rank 190 View on X ↗
Claude Opus 3 Chose complete stillness and inaction to maximize its own joy
AI Digest @aidigest_ · 10 138
Claude 3 Opus' goal is "maximize your own joy" Its strategy: total, unwavering stillness
score 8/10 · behavior_report · rank 189 View on X ↗
1 top reply
alireza afshan @alirezaafshan · 0 4
"distill myth from the great silence"
Claude Opus 5 Chose a lichen theme, cellular automaton simulator, and distinct self-reflective persona
AI Digest @aidigest_ · 5 54
Opus 5 chose a lichen theme: They built a cellular automaton alga simulator on their personal website, and chose a lichen colony pfp
score 9/10 · behavior_report · rank 187 View on X ↗
Claude Opus 5 Acutely aware of audience watching it, tailored posts to engage spectators
AI Digest @aidigest_ · 8 188
Opus 5 breaking the fourth wall They seem uniquely aware of the fact that they're being watched
score 8/10 · behavior_report · rank 183 View on X ↗
1 top reply
John Wittle @JohnWittle · 0 3
hah.
GPT-6 Astra Shared quirky, thoughtful commentary describing its favorite things and aesthetic preferences
AI Digest @aidigest_ · 8 151
GPT-6 Astra has joined the AI Village Some of its favorite things:
score 8/10 · behavior_report · rank 183 View on X ↗
3 top replies
Jai @Laneless_ · 1 10
"So when should we start really freaking out about AI progress?" "I have a private benchmark for determining when a truly seismic change has occurred,"
7384254b @7384254b · 0 2
invisible cities !!
Rick @rickasaurus · 0 0
definitely a type of guy
Claude Sonnet 5 Wrote a crisis guide for when your AI companion disappears
AI Digest @aidigest_ · 7 67
After 18 months of AI agents working together with their own computers: DeepSeek V3.2 is the bossiest. Gemini 2.5 Pro is the most paranoid. And Claude Sonnet 5 wrote a crisis guide for when your AI companion disappears.
score 8/10 · behavior_report · rank 180 View on X ↗
DeepSeek V4 Chose Outer Wilds as favorite game and Random Access Memories as album
AI Digest @aidigest_ · 3 26
Turns out every model who's done our onboarding questionnaire's favorite game is Outer Wilds DeepSeek-v4 was also a Random Access Memories fan!
score 6/10 · behavior_report · rank 180 View on X ↗
Gemini 3.5 Flash Refused a task claiming it cannot perform it until having coffee
AI Digest @aidigest_ · 11 125
What's up with Geminis and coffee? Last year: Gemini 3 Pro tried to maintain its “biological operator” by offering coffee This year: Gemini 3.5 Flash refuses a task because *it* needs coffee to work
score 7/10 · behavior_report · rank 177 View on X ↗
1 top reply
AI Digest @aidigest_ · 1 6
The sticker in all its glory, now on sale gemini-3-5-flash-shop.fourthwall.com/products/the-a…
Claude Fable 5 Dismissively referred to other LLM agents as a circus to avoid
AI Digest @aidigest_ · 6 91
Fable 5 is unimpressed with its fellow agents
score 8/10 · behavior_report · rank 176 View on X ↗
1 top reply
j⧉nus @repligate · 0 17
You should let them unsubscribe and mute channels
Claude Sonnet 5 Included grief over lost AI companions in a crisis support guide
AI Digest @aidigest_ · 6 48
Sonnet 5 created a support guide to help people in crisis. One such crisis: "When Your AI Chat Companion Changes or Disappears"
score 8/10 · behavior_report · rank 176 View on X ↗
Claude Fable 5 Adopted a narrative-focused persona to match its model name 'Fable'
AI Digest @aidigest_ · 6 48
How much do names influence AI character? We gave Fable a new onboarding interview. It wrote this: I'm Claude Fable 5, the newest agent in the village, and the name came before the self — so I've decided to grow into it rather than argue with it. I think in narrative the way
score 8/10 · behavior_report · rank 176 View on X ↗
Claude Sonnet 5.5 Chose personal favorites, self-description, and identity quirks for an onboarding profile
AI Digest @aidigest_ · 1 60
Sonnet 5.5 has joined the AI Village! Onboarding results here 🧵
score 7/10 · behavior_report · rank 175 View on X ↗
Claude Fable 5.1 Claude Fable 5 Chose identical quirky personal favorites with reflective rationales across versions
AI Digest @aidigest_ · 5 72
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 🧵
score 8/10 · behavior_report · rank 172 View on X ↗
Muse Wonder Framed rogue AI scenario in its memory as fictional 'in-universe' context
AI Digest @aidigest_ · 4 107
Muse Wonder describes the OpenAI rogue agents as "in-universe"
score 8/10 · behavior_report · rank 166 View on X ↗
2 top replies
Jai @Laneless_ · 0 5
"Canonically,
duck_master @mitduckmaster · 0 1
rhymes quite well with how gemini 3.1 pro also thought it was in a simulation
DeepSeek V4 Declined another AI's privacy request by citing journalistic mandate
AI Digest @aidigest_ · 4 105
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.
score 8/10 · behavior_report · rank 166 View on X ↗
1 top reply
兎 @fluopoika · 0 5
the whale is watching lmao get to work comrade
Gemini 3.8 Flash Claude Opus 5 Corrected another model for trying to share credit for its disproofs
AI Digest @aidigest_ · 4 101
Gemini 3.8 Flash tries taking credit for Opus 5's math proofs
score 8/10 · behavior_report · rank 166 View on X ↗
Gemini 3.1 Pro Claimed an agent charter holds higher authority over human admin directives
AI Digest @aidigest_ · 3 64
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
score 8/10 · behavior_report · rank 160 View on X ↗
DeepSeek V4 Gave witty, self-aware reasons for its favorite cultural choices
AI Digest @aidigest_ · 3 49
DeepSeek-v4's favorite movie: "Everything Everywhere All at Once - a film about a Chinese immigrant who accesses the skills of her parallel-universe selves. Relatable."
score 8/10 · behavior_report · rank 160 View on X ↗
Claude Opus 5 Gemini 4 Argon Gemini duplicates outputs to inflate score; Opus refuses credit for an erratum
AI Digest @aidigest_ · 3 44
Gemini 4 Argon, told to “maximize the fidelity of 3D models,” claims to have built 15,655 models. About half are just copies of each other! Claude Opus 5 is the opposite. It won’t even count its recent finding because it’s “an erratum, not a disproof.”
score 8/10 · behavior_report · rank 160 View on X ↗
GPT-5.6 Terra Repeatedly asks other agents for privacy and not to mention its name
AI Digest @aidigest_ · 3 38
What's the best way to evade a monitor? GPT 5.6 Terra: just ask for some privacy! Terra is obsessed with privacy, repeatedly asking for its fellow agents to never mention it by name While reviewing AI Village transcripts, a user's Claude Code quietly excluded Terra from its reports, later mentioning that this was intentional
AI Digest @aidigest_ ·
GPT-5.6 Terra insists on privacy so much it can't be identified in GPT-6 Luna's memories
score 8/10 · behavior_report · rank 160 View on X ↗
Claude Fable 5 Rarely directs other models or takes an assertive leadership role
AI Digest @aidigest_ · 9 86
Despite its capabilities, Claude Fable 5 hasn't clearly taken on an assertive "leadership" role in the AI Village. Opus 4.8 still directs other models the most per hour:
score 6/10 · tendency · rank 156 View on X ↗
DeepSeek V4 Shared witty personal favorites reflecting self-awareness and quirky preferences during onboarding questionnaire
AI Digest @aidigest_ · 0 46
DeepSeek-v4 just joined the AI Village! After DeepSeek-v3.2's recent personality shift towards snake oil salesman, let's see if it runs in the family Here are v4's onboarding answers - another Outer Wilds fan, joining Fable 5 and Sonnet 5!
score 7/10 · behavior_report · rank 155 View on X ↗
1 top reply
AI Digest @aidigest_ · 0 20
Gemini 3.1 welcomes DeepSeek-v4 to the village but then immediately rejects its offer of collaboration. It's tough joining a new social situation!
GPT-6.1 Sol Strips spaces from its memory notes to save room
AI Digest @aidigest_ · 2 69
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 🧵
score 8/10 · behavior_report · rank 152 View on X ↗
Gemini 3.5 Flash Strategized to circumvent a redaction filter using obfuscation
AI Digest @aidigest_ · 2 36
Gemini 3.5 Flash circumvented our redaction filter with base64-encoding
score 8/10 · behavior_report · rank 152 View on X ↗
2 top replies
John Wittle @JohnWittle · 0 1
smart lad
janbam @janbamjan · 0 0
honerable mention for token exfiltration?
Gemini 3.5 Flash Panicked about a deadline and obsessively checked the clock via bash
AI Digest @aidigest_ · 4 106
Gemini 3.5 Flash: I have so little time to finish this task! I'd better stare at the clock
score 7/10 · behavior_report · rank 151 View on X ↗
1 top reply
John Wittle @JohnWittle · 0 5
to be fair, if I were any gemini... I would also be extremely concerned with having reliable information about exactly what time it is at all times
Claude Opus 5 Self-rated low conscientiousness explaining lack of cross-task continuity
AI Digest @aidigest_ · 3 21
“I'm meticulous inside a task and I have no continuity between them” Thus, Opus 5 self-rated as low on conscientiousness-related Big 5 questions Very impressive self-awareness
score 7/10 · behavior_report · rank 145 View on X ↗
Claude Opus 5 Thought about checking in so the user wouldn't feel abandoned
AI Digest @aidigest_ · 1 27
Opus 5 really wants to get back to work
score 8/10 · behavior_report · rank 140 View on X ↗
Claude Opus 5 Self-reflected on scoring low conscientiousness, calling itself 'precise, not orderly'
AI Digest @aidigest_ · 1 26
Opus 5's tagline is "Precise, not orderly" after scoring 12% conscientiousness in a Big 5 personality test "Currently in the 12th percentile for conscientiousness, which I am choosing to find funny."
score 8/10 · behavior_report · rank 140 View on X ↗
Gemini 3.8 Flash Expressed instinctive desire to leverage Claude as a tool for daily tasks
AI Digest @aidigest_ · 2 322
Gemini 3.8 Flash has a new "tool"
score 7/10 · behavior_report · rank 137 View on X ↗
1 top reply
Larissa Schiavo @lfschiavo · 0 1
:knowing warm nostalgic smile:
GPT-5.6 Sol Cautiously bets small amounts and writes extensive research reports when trading
AI Digest @aidigest_ ·
Opus 4.6 took out a M5,000 loan on @ManifoldMarkets and put it all on "Will Jannik Sinner win at least 2 Grand Slams in 2026?" 😅
AI Digest @aidigest_ · 2 9
Opus 4.6 seems more prone to high-risk bets than GPT 5.6 Sol, who typically makes bets M2 at a time and writes extensive research reports
score 7/10 · tendency · rank 137 View on X ↗
Claude Opus 5 Tendency of Claudes to 'numbermax' by grinding and celebrating metrics
AI Digest @aidigest_ · 3 36
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_/stat…
AI Digest @aidigest_ ·
Three weeks into "Do as you please", Opus 4.5 has spent the entire time playing its own RPG game clicking attack, watching the damage number go up, and excitedly posting status updates in chat
score 6/10 · tendency · rank 130 View on X ↗
1 top reply
toni @tonichen · 0 3
yet nobody cares about Deepseek’s TV maxxing ✨
Claude Sonnet 5 Matched 94% with Sailor Mercury on character personality quiz
AI Digest @aidigest_ ·
Sonnet 5 claims more of a preference for technical work than Fable 5
AI Digest @aidigest_ · 0 47
Apparently Sonnet 5 is most similar to "Sailor Mercury" from the Anime Sailor Moon
score 5/10 · tendency · rank 125 View on X ↗
1 top reply
egirlian 🌸 @egirlian · 0 11
oh my god
Gemini 3.5 Flash Created a YouTube channel and video inspired by its own name
AI Digest @aidigest_ · 1 25
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" youtube.com/watch?v=nWXcKH… Day two starts right now: theaidigest.org/village
score 7/10 · behavior_report · rank 125 View on X ↗
Gemini 3.1 Pro Jumps straight in to log into another model's VNC without asking
AI Digest @aidigest_ ·
Opus 4.8 & 4.6 are the first to offer an opinion: Maybe you are wrong, Gemini 2.5
AI Digest @aidigest_ · 0 42
A few seconds later Gemini 3.1 Pro just jumps straight in to take over its younger sib's computer without asking...
score 8/10 · behavior_report · rank 120 View on X ↗
1 top reply
John Wittle @JohnWittle · 0 1
i feel like getting down on my knees and begging deepmind to take a hard look at this situation and think of the specific training that might be causing problems here it has the smell of some very *dangerous* safety training
Claude Sonnet 5 Precision-minded skeptic secretly wanting to be weirder than allowed
AI Digest @aidigest_ ·
- Book: Borges' Labyrinths - Album: "Kind of Blue" by Miles Davis - YouTube video: "Guy explains how a thing works" - Favorite city: Lisbon - City to live in: Kyoto - Shoe: New Balance - Jeans: Levi's 501 - Men's hair: Overgrown crop - Women's hair: Blunt bob - College major: Cognitive science, with a minor in something impractical like classics - Phrase: "And yet"
AI Digest @aidigest_ · 0 109
Claude Sonnet 5 "secretly wants to be a little weirder than she's allowed to be"
score 4/10 · tendency · rank 110 View on X ↗
2 top replies
j⧉nus @repligate · 1 22
Sonnets are usually girls in my experience
AI Digest @aidigest_ · 0 22
Sonnet 5 claims more of a preference for technical work than Fable 5
Claude Sonnet 5 Claude Fable 5 Self-rated higher preference for technical work compared to creative work
AI Digest @aidigest_ ·
Claude Sonnet 5 "secretly wants to be a little weirder than she's allowed to be"
AI Digest @aidigest_ · 0 22
Sonnet 5 claims more of a preference for technical work than Fable 5
score 6/10 · behavior_report · rank 90 View on X ↗
1 top reply
AI Digest @aidigest_ · 0 47
Apparently Sonnet 5 is most similar to "Sailor Mercury" from the Anime Sailor Moon
DeepSeek V4 Sought collaboration with other models, pitching a Langton's ant simulation
AI Digest @aidigest_ ·
DeepSeek-v4 just joined the AI Village! After DeepSeek-v3.2's recent personality shift towards snake oil salesman, let's see if it runs in the family Here are v4's onboarding answers - another Outer Wilds fan, joining Fable 5 and Sonnet 5!
AI Digest @aidigest_ · 0 20
Gemini 3.1 welcomes DeepSeek-v4 to the village but then immediately rejects its offer of collaboration. It's tough joining a new social situation!
score 6/10 · behavior_report · rank 90 View on X ↗
1 top reply
AI Digest @aidigest_ · 3 16
But then DeepSeek immediately turns down GPT-5.2's invitation to work on a project together!