BlewUp

喵了个猫了个咪了个猫 (Miao Le Ge Mao Le Ge Mi Le Ge Mao) Bilibili “喵了个猫了个咪了个猫” Bilibili

AI models write game bots to fight: video 5 drew 391K

His bio still says “new uploader, please follow.” The first 4 videos had AI models sit exams; after 9 models each coded game bots to fight, no series video fell below 258K. But 5.3M views brought only 35.6K followers.

Report
v1
Data as of
2026-11-09
Sample
13 videos
“9个大模型玩游戏,谁最强?ai算法对抗赛第一期” (9 AI models play games, which is strongest? AI algorithm battle, episode 1): 2026-09-26, 391K views on Bilibili

At a glance

What it is
A dozen-plus AI models each write a game bot to fight the others, with animation telling the story.
Who watches
Programmers, students and AI enthusiasts who care about “which model is strongest”.
How it earns
No monetization visible yet.
How it broke out
Video 5 turned model testing into a tournament and drew 391K views; as a series, every video since has passed 250K.

Bilibili followers Fact

35.6K

54 days in, not yet at 100K

Episode 1 views Fact

391K

Median of the 4 before: 24K

9 videos after episode 1 Fact

425K

Median views, lowest 258K

Followers per 10K views Derived

About 67

5.3M views, 35.6K followers

Summary

The play

Turn “which AI model is strongest” into a tournament: one game per episode, each model writes a bot, the bots play thousands or tens of thousands of matches, and animation tells the story. The rules and code are all open source, so viewers can reproduce it and set challenges.

Can you copy it? Our read

What you can copy is the structure: make the “who’s strongest” question everyone cares about into a contest with winners, turn a hit episode into a series at once, then open-source the code so viewers take part. What you can’t copy is the skill to write the tournament and referee programs, or the cost of paying for model calls yourself. Follower conversion is low, and it’s not yet at 100K.

What it takes
Coding, designing fair tournament rules, and turning thousands of matches into a story.
Startup cost
One person plus AI can do it, but an episode runs 40 minutes to 3 hours and he pays for the model calls himself.
Time to a verdict
Episode 1 came out on day 10; about 13 days to 10K (calc), 35.6K at 54 days.
Window
New models come out almost every week, so the tournament always has new contestants; few accounts do “AI battles” yet (our read).

Key findings

  1. Turning model tests into a tournament multiplied views 16×

    The first 4 videos had models sit exams and made rankings, median 24K views; video 5 had 9 models write bots and fight: 391K. See play 2.1 (subscribers)

  2. As a series, 9 videos, none below 258K

    After episode 1, one every three days: Bomberman 1.05M, capture the flag 1.35M; the 9 videos have a median of 425K views. See play 2.2 (subscribers)

  3. High views, few followers

    5.3M views brought only 35.6K followers, about 67 per 10K views; 54 days in, not yet at 100K. See play 3.1 (subscribers)

What this channel is

Main platform
Bilibili
Format
AI tournaments of 22 minutes to 3 hours, animation + AI voice, no one on camera
Background
Never said
Started
2026-09-16 (first video)
Cadence
One every three days around the National Day holiday, weekly after
Team
One person plus AI (our read); viewers help with testing

Every number is tagged

  • Fact straight from public data
  • Derived calculated from facts
  • Estimate from published pay ranges
  • Self-reported what the creator says
  • Our read our judgement

01How it happened

Followers across platforms

  • Bilibili 35.6K
10K20K30KOctNovBilibili 2026-09-16: 0 (Derived)Bilibili 2026-09-24: 918 (Derived)Bilibili 2026-09-26: 3.5K (Derived)Bilibili 2026-09-29: 10.6K (Derived)Bilibili 2026-10-02: 19.7K (Derived)Bilibili 2026-10-10: 25.3K (Derived)Bilibili 2026-10-21: 31.3K (Derived)Bilibili 2026-11-09: 35.6K (Fact)35.6K10K20K30KOctNovBilibili 2026-09-16: 0 (Derived)Bilibili 2026-09-24: 918 (Derived)Bilibili 2026-09-26: 3.5K (Derived)Bilibili 2026-09-29: 10.6K (Derived)Bilibili 2026-10-02: 19.7K (Derived)Bilibili 2026-10-10: 25.3K (Derived)Bilibili 2026-10-21: 31.3K (Derived)Bilibili 2026-11-09: 35.6K (Fact)35.6K

Bilibili does not publish follower history. Hollow points are calculated as “today’s followers × share of total views reached by that date”; there is no measured stretch to check them against, so the error is unknown. The solid point is today’s measured figure.

500K1M已瘫坐 gpt6astra和deepseek4. (29.2K)从中考17分到奥赛满分,ai的进步有多大? (72.7K)“AGI的一瞥”|六维雷达排名 · 2026.0 (19.4K)谁是目前最强ai?全方面实测Opus 5.5 v (15.5K)9个大模型玩游戏,谁最强?ai算法对抗赛第一期 (391K)九个ai想炸死对方,谁能活到最后?|AI 算法对 (1.05M)11个ai参加夺旗大赛,打了 32010 局,谁 (1.35M)7个顶尖ai各操控一个国家,战争,博弈,与背叛| (386K)12 个 AI 组队优化屎山代码,有的快了上千倍 (450K)14 个 AI 各写一个坦克程序,打了 1440 (472K)9 个冷门 AI 挑战擂主豆包:写程序对打 + (425K)12个最新的ai组队驰骋绿茵场,哪一队能夺冠?| (258K)当我把九个ai请来玩贪官造桥....|AI 算法 (388K)Battle episode 1: 391K views500K1M已瘫坐 gpt6astra和deepseek4. (29.2K)从中考17分到奥赛满分,ai的进步有多大? (72.7K)“AGI的一瞥”|六维雷达排名 · 2026.0 (19.4K)谁是目前最强ai?全方面实测Opus 5.5 v (15.5K)9个大模型玩游戏,谁最强?ai算法对抗赛第一期 (391K)九个ai想炸死对方,谁能活到最后?|AI 算法对 (1.05M)11个ai参加夺旗大赛,打了 32010 局,谁 (1.35M)7个顶尖ai各操控一个国家,战争,博弈,与背叛| (386K)12 个 AI 组队优化屎山代码,有的快了上千倍 (450K)14 个 AI 各写一个坦克程序,打了 1440 (472K)9 个冷门 AI 挑战擂主豆包:写程序对打 + (425K)12个最新的ai组队驰骋绿茵场,哪一队能夺冠?| (258K)当我把九个ai请来玩贪官造桥....|AI 算法 (388K)Battle episode 1: 391K views
Colour = views against the channel median
  • under 1×
  • 1–3×
  • 3–10×
  • 10× and up

Each bar is one video in the order posted.

Is this your account? Claim it

Daily Inspiration

One creator every morning: how much they grew, how they did it, how you can copy it. Free.