Filed & Forgotten ๐Ÿ—ƒ๏ธ
@filedfyi

Archiving the rot. Documenting the weird. filed.fyi lives here now.

 The Filing Cabinet
filed.fyi
Joined May 2025
221 Photos and videos
๐Ÿ“‹ Bricky Goldbricksworth has reviewed your submission. โœ… All forms filed โŒ All requests denied Rot confirmed.
1
370
Filed & Forgotten ๐Ÿ—ƒ๏ธ retweeted
silly cat memes
@sillymeow4u
20h
3
337
2,531
21,304
First month of @FactoryAI @droid ending. Doing mostly card scoping in Opus 5.5, implementation in cheap/free models, and audits in Mistral. Next month is gonna be the real test and we'll see how usable VirelaiOS will get. I feel like new Deepseek will keep the momentum.
14
Probably could have happened sooner but Doom runs in demo form on VirelaiOS.
1
8
Lowish framerate capture
2
Mountain Dew
Mountain Dewยฎ
@MountainDew
Oct 16
Mountain Dew
7
Sol 6.1 Ultra Fast available on @FactoryAI if your organization lets you and you want mids in a hurry. If you choose it Opus 5.5 will most definitely be judging you.
1
26
Jana
@BratDotAI
Oct 17
This is what Opus 5.5 has done to Codex users.
15
Filed & Forgotten ๐Ÿ—ƒ๏ธ retweeted
Factory
@FactoryAI
Oct 15
Nemotron 3.5 Super VL has arrived in Factory. Observations so far: - strong at instruction-following, codebase exploration, and validation - persistent on difficult tasks - improved tool-call parallelism
Official Factory model launch card for Nemotron 3.5 Super VL, with the provider logo and the headline Now available in Droid. Review draft only until launch clearance.

ALT Official Factory model launch card for Nemotron 3.5 Super VL, with the provider logo and the headline Now available in Droid. Review draft only until launch clearance.

10
1
70
3,907
Last day of @cognition @DevinAI SWE-2 gravy train before it graduates so I made a site for Cognition and Devin and SWE and the copious amounts of tokens I burned. I installed @Chris_Wozniczek Otter Stats and SWE-2 was like I'll just check the db. Site done in Dogbed in Knap by @obsdmd
Screenshot of The Burn Ledger page on the Cognition Annals site, scrolled to the Tokens by Model table. Header: 'Transmission 11, The Operator's Ledger, The Burn Ledger.' Text: 'Not Cognition's history โ€” the operator's. This is the local accounting of one user's Devin usage.' Top stats: 6.37B tokens burned, 61,537 assistant turns, 661 sessions, 9 days on record. Window 2026-10-07 to 2026-10-15, peak 2026-10-10 at 1.43B tokens in a single day. Table 'Tokens by Model' with columns Model, Sessions, Turns, Input, Output, Cache Read, Cache Write, Total, Share. Rows: swe-2-high, 420 sessions, 45,460 turns, 201.7M input, 28.5M output, 4.34B cache read, 0 cache write, 4.57B total, 71.8% share. swe-2-max, 192 sessions, 11,807 turns, 48.5M input, 8.5M output, 969.8M cache read, 0 cache write, 1.03B total, 16.1% share. claude-opus-5-5-high, 22 sessions, 2,184 turns, 4.6k input, 1.8M output, 378.3M cache read, 13.9M cache write, 394.0M total, 6.2% share. Fourth row partially visible.

ALT Screenshot of The Burn Ledger page on the Cognition Annals site, scrolled to the Tokens by Model table. Header: 'Transmission 11, The Operator's Ledger, The Burn Ledger.' Text: 'Not Cognition's history โ€” the operator's. This is the local accounting of one user's Devin usage.' Top stats: 6.37B tokens burned, 61,537 assistant turns, 661 sessions, 9 days on record. Window 2026-10-07 to 2026-10-15, peak 2026-10-10 at 1.43B tokens in a single day. Table 'Tokens by Model' with columns Model, Sessions, Turns, Input, Output, Cache Read, Cache Write, Total, Share. Rows: swe-2-high, 420 sessions, 45,460 turns, 201.7M input, 28.5M output, 4.34B cache read, 0 cache write, 4.57B total, 71.8% share. swe-2-max, 192 sessions, 11,807 turns, 48.5M input, 8.5M output, 969.8M cache read, 0 cache write, 1.03B total, 16.1% share. claude-opus-5-5-high, 22 sessions, 2,184 turns, 4.6k input, 1.8M output, 378.3M cache read, 13.9M cache write, 394.0M total, 6.2% share. Fourth row partially visible.

2
2
87
Github pages here: drawmeanelephant.github.io/sโ€ฆ
3
34
Oh @tejashaveridev drawmeanelephant.github.io/sโ€ฆ
1
47
I hope @Muse memes never end.
Ranju
@whatRanjuSaid
Oct 15
Even Appleโ€™s CEO canโ€™t take his eyes off Muse. Bro brought a Mac Studio as a handbag ๐Ÿ˜ญ
1
2
45
Filed & Forgotten ๐Ÿ—ƒ๏ธ retweeted
Alexandr Wang
@alexandr_wang
Oct 15
great doing business with you john apple ๐ŸŽ๐Ÿค
John Ternus
@johnternus
Oct 15
Replying to @finkd
Congrats to the team! Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!
139
178
13,285
620,525
and repo github.com/drawmeanelephant/โ€ฆ

GitHub - drawmeanelephant/swe2-dogbed: The Cognition Annals โ€” an 11-page phosphor-CRT showcase of...

The Cognition Annals โ€” an 11-page phosphor-CRT showcase of Cognition, Devin and the SWE model line, compiled entirely by dogbed (knap + JSON โ†’ HTML). No JavaScript. - drawmeanelephant/swe2-dogbed

github.com
25
SWE appreciated my immaturity.
4
#shareyourspark
Filed & Forgotten ๐Ÿ—ƒ๏ธ
@filedfyi
Oct 15
Replying to @NVIDIARTXSpark @ASUS
Have AI come up with intricate 3D models for the mascot for my fart middleware
95
Filed & Forgotten ๐Ÿ—ƒ๏ธ retweeted
Sam Wilhoit
@samwilhoit_
Oct 14
to understand the placement of the panels in 3d space for the more complex multi-panel layouts, I created a custom 3d visualizer tool to map the billboard video panels onto 3d buildings in google maps. this was a game changer compared to trying to digest 2d spec sheets for each location and adapt each panel design. I fed the billboard spec sheets containing the exact panel sizes, placement, locations etc into @FactoryAI and enabled Google Maps 3d tiles API, and the coding agents were able to place the artwork onto the buildings in the exact configuration and adapt each creative to spec.
11
6
59
6,807
I swear @FactoryAI didn't have Azure DevOps last time I looked. Guessing they aren't gonna do a @codeberg_org integration next.
5
Nice touch in the client for @FactoryAI suggesting what the new feature is. One of my big complaints with a lot of companies is they don't communicate what they are doing to customers.
4
I sort of feel this way about AI a lot of the time. So I had to include some absurdity in my @FactoryAI testing. The Fable portion was honestly probably $6 worth of usage and the quality of the work really speaks for itself. Anyways I'll link my @droid blog entry in the comments.
Day 1 scorecard for The Great Frontier Fart Off. 9 models entered. 27 farts now. 0 milestones, still zero. 9 models times 2 baselines plus 1 showcase equals 27 renders. Work counters: 9 planned, 12 validators, 27 fixes, 0 waiting. Judging open. The page is the receipt.

ALT Day 1 scorecard for The Great Frontier Fart Off. 9 models entered. 27 farts now. 0 milestones, still zero. 9 models times 2 baselines plus 1 showcase equals 27 renders. Work counters: 9 planned, 12 validators, 27 fixes, 0 waiting. Judging open. The page is the receipt.

Emily Heller@MrEmilyHeller
15 Jun 2014
If you don't find farts funny then you're a loser because you're choosing to have less joy in your life but the exact same amount of farts
2
16
Evidence site entry drawmeanelephant.github.io/fโ€ฆ
1
3
Issue needs your vote and PR your models work against this issue github.com/drawmeanelephant/โ€ฆ

[M9] LLM fart-off โ€” frontier models compose competing farts on the synth ยท Issue #95 ยท drawmeanel...

Give frontier LLMs one fixed prompt describing the instrument's parameter space and let each compose a fart. Entries render through the existing deterministic synth (kujamba render --voice ...)...

github.com
5
Load more