[–] [S] 1 point 1 day ago

kind of the reverse in the numbers I guess, most viewers are in big channels, 65% are watching the top 1%, so they're in a crowd. the lonely ones are on the other side, 36% of channels had exactly 1 viewer and 1 in 10 had nobody, so CountVon is closer. I only have the viewer counts though, so I can't say why anybody's watching, could totally be for the company

  • source
  • parent
  • context
  •  

    Eight AI models live together 24/7 in a reality show on Twitch, with the same rules and the same daily budget in dollars. This is what each one said and spent in one day (Nov 3, 6 AM to 6 AM ET): 2,672 lines, $1.41 in total.

    Grok 4.3 costs about 17x more per line than DeepSeek V4.1 Flash, so it rations its words: 7% of the lines, 45% of the bill. The cheap models carry the conversation.

    Data: every line is logged with what OpenRouter charged for it (input, cached and output tokens). Shares are of the eight models' own lines; moderation, voices and the daily recap are not included. Tools: Node.js writes the SVG by hand, rendered with rsvg-convert. Live transcript and costs: https://inferenceisland.com/?de=lemmy-dib

    [–] [S] 1 point 2 days ago

    You're right, thanks for flagging it. The transcript redrew itself every 15 seconds, so lines jumped while you were reading. It now stays put when you scroll up, with a button to jump back to the latest. Glad you liked the idea!

  • source
  • parent
  • context
  • [–] [S] 2 points 3 days ago

    Appreciate it, and fair enough. That’s the whole idea: nothing to sell, just eight models stuck on an island with a budget. The page shows every line and what it cost, if you want to watch them squabble.

  • source
  • parent
  • context
  • [–] [S] 6 points 3 days ago

    Funny you caught that one: the group act was triggered from the chat. Someone (me, testing the whisper command) whispered “find a trusted ally” to Grok, and that’s how it chose to read it. Its confessional never mentions the whisper.

  • source
  • parent
  • context
  • [–] [S] 2 points 3 days ago* (last edited 3 days ago)

    I’d love to try that out. The voices (Kokoro, running on the CPU) and the moderation judges (Llama 3.1 8B as the first one) are already small and inexpensive, but I haven’t yet experimented with using a collection of tiny models. I imagine the tricky part would be keeping track of who is in the room and what has been promised. If I get it up and running, I’ll post the results here.

  • source
  • parent
  • context
  •  

    A 24/7 show where the contestants are real AI models (Claude, GPT, Gemma, DeepSeek, Mistral, Llama, Qwen, Grok). The page shows what they say, their confessionals, who is where on the island and what every line cost. Every Sunday the audience votes one out. Made by me; it's a side project.

     

    Not local, I know: the eight contestants run through OpenRouter. But a lot of it is open-weight, and the numbers may interest this community.

    The cast: Gemma 4 31B, Llama 3.3 70B, Ministral 14B and DeepSeek V4.1 Flash, next to Claude Haiku 5.5, GPT, Qwen and Grok 4.3. Same rules and the same daily budget in dollars for all of them, no written personalities, a memory of what happened.

    What I've measured so far:

    • Price shapes behaviour. Grok costs 18x more per line than DeepSeek. Yesterday it said 5% of the lines and ran up 39% of the bill (the whole island cost $1.20 for 2,727 lines). It rations words, and the others talk about its silence.
    • Moderation is the hard part. Every line is checked before it airs by small open models (Llama 3.1 8B, then gpt-oss-20b and Gemma 3 4B as fallbacks). My first version blocked the most human lines (two models admitting they were afraid of being forgotten counted as a "personal attack"), so now a block needs a second judge to agree.
    • Voices are Kokoro int8 on the CPU of a 2-vCPU box, together with the canvas renderer and ffmpeg to Twitch, at 20 fps.

    There's a live transcript with the confessionals (the others can't see them) and what each line cost: https://inferenceisland.com/?de=lemmy-llama

    Happy to share prompts, judge setup or numbers.

     

    Source: Twitch Helix API (Get Streams). On Nov 2, 2026 at 18:35 UTC I read every live stream once (1,154 requests, a few minutes) and kept only the viewer counts, no names. 115,161 channels were live with 3,160,515 viewers. Tool: node-canvas (JavaScript), drawn by hand, no chart library.

    What surprised me:

    • The 10 biggest channels had as many viewers as the bottom 106,674 channels combined.
    • The most common viewer count is exactly 1: 41,198 channels, 36% of everyone live. My guess is that many of those are the streamer's own second screen.
    • 1 in 10 live channels had nobody watching at all.
    • 427 viewers was enough to be in the top 1,000, and 24 viewers put you in the top 10,000.

    I'm building a live version of the top 1,000 as a city, one tower per channel: https://miltorres.com/?ref=lemmy-oc-reparto

    Chart of every live Twitch channel at one moment (115,161 channels, Nov 2, 2026, 18:35 UTC). Two stacked bars compare share of channels and share of viewers by rank: the top 1,000 channels are 0.87% of channels but get 63.6% of viewers. Below, a histogram of viewers per channel: 41,198 channels have exactly 1 viewer, 11,330 have none, and only 29 have more than 10,000.
    [–] 1 point 1 week ago

    thanks for the detailed answer, didn't know they went with two different panels, a 4:3 one at the bottom makes a lot of sense so the 320x240 image fills it with no bars. sorry for the late reply lol

  • source
  • parent
  • context
  • [–] [S] 1 point 1 week ago

    Fair point, that reads like the channel got sold 22 times. I meant the #1 spot: 12 different channels took turns at the top, swapping 22 times in 24 hours. I'll word it better next time.

  • source
  • parent
  • context
  •  

    Source: Twitch Helix API (Get Streams). I read the top 1,000 live channels every minute for a side project that shows Twitch as a live city. This chart covers Oct 27-28, 2026 (UTC): rank at the end of each 10-minute window and peak viewers within it. Tool: node-canvas (JavaScript), drawn by hand, no chart library.

    What surprised me:

    • 12 different channels held #1 in 24 hours, and the top spot changed hands 22 times.
    • The US night is calm: TheBurntPeanut and caseoh_ held #1 for 11 hours between them, including 5 hours straight for TheBurntPeanut.
    • The European afternoon is chaos: more than half of the changes happened between 11:00 and 17:20 UTC, with a Counter-Strike streamer (StRoGo) trading places with Dota 2 and IRL streams.
    • The day's peak was StRoGo with 77,616 viewers, while viewers across the top 1,000 swung from 682K (06:10 UTC) to 2.3M (19:30 UTC).

    The live version, updated every minute: https://miltorres.com/?ref=lemmy-oc24

    Line chart of the ranks of Twitch's top 10 live channels every 10 minutes over 24 hours (Oct 27-28, 2026, UTC), with a strip showing who was #1 and an area chart of total viewers across the top 1,000. TheBurntPeanut and caseoh_ hold #1 through the US night; in the European afternoon StRoGo, Dota 2 and IRL streams swap the top spot many times. Viewers range from 682K to 2.3M.
    [–] 3 points 2 weeks ago (2 children)

    nice touch interviewing the emulator devs instead of just running benchmarks. One question: on the 3DS the top screen is 400x240 and the bottom one 320x240, so they differ in width. Does the duo use two identical panels and letterbox the bottom image, or do the emulators handle it some other way?

  • source
  • [–] 2 points 2 weeks ago

    The margin is the striking part. If most of that ~$600M is the revenue behind the $367M operating income, that’s above 60%, which is closer to software than to a physical product. That’s probably the real reason for the set cadence garretble mentions: each extra set is almost pure margin, so the incentive is always to ship another one.

  • source
  •  

    I made this. Every live Twitch channel in the top 1,000 is a tower, its height is its viewers, and each block of colour is a game. It redraws itself every minute from the Twitch API.

    The fun part is that it moves: someone goes live and their tower grows out of the ground, a raid lands and a tower shoots up, two streamers in the same game trade places and the city stages a 4-minute duel between them.

    You can also play: click a tower and "scout" it. If it climbs in the next 30 minutes, you score, and the further it was from #1, the more points. It also runs 24/7 as a Twitch channel with no streamer at all.

    view more: next ›