DistractionRectangle's recent activity
-
Comment on Which covers did it better than (or put a fresh twist on) the original? in ~music
-
Comment on What programming/technical projects have you been working on? in ~comp
DistractionRectangle LinkI considered writing a epic about how I waged war with the Wifi, but I'll spare you. It was DNS. It's always DNS. Apparently my carrier does DNS hijacking and I only just figured it out because...I considered writing a
shortepic about how I waged war with the Wifi, but I'll spare you.It was DNS. It's always DNS.
Apparently my carrier does DNS hijacking and I only just figured it out because their system broke... Of course it was working when I went to trouble shoot it, which led to me tilting at windmills for a bit until I threw out all assumptions and rechecked DNS. It wasn't fruitless though, I did make some sizable wins on loaded latency and throughput as side quests by switching to an NSS build of openwrt and finding someone else did the legwork for autorate sqm (the alternative being an extremely low static SQM ceiling since my provider is cellular internet)
-
Comment on What programming/technical projects have you been working on? in ~comp
DistractionRectangle Link Parent/offtopic This reminds me of a story one of Professors had about working for IBM's storage division during the dotcom bubble. When you sell shovels during a gold rush, corners are cut and...Almost everyone I know is having trouble with hard drives, WD or Seagate, purchased in the last 8 months or so. I'm very concerned that the stupidity of the AI market means they've shifted their QA and stuff that normally wouldn't make it out the door is now "consumer grade" (if you can get it at all).
/offtopic This reminds me of a story one of Professors had about working for IBM's storage division during the dotcom bubble. When you sell shovels during a gold rush, corners are cut and inevitably quality suffers. I'm having trouble finding a source, as it wasn't the deathstar. Unlike the deathstar, it wasn't a long protracted problem. It was one horrendous batch that was DOA.
Anyways. R&D is playing fast and loose in production because if you aren't first, you're last. So they had this oil on the platters, the idea being it safeguards against what would be minor head crashes. Kinda like how a hydro foil lifts a boat out of the water, the idea being that at speed, if the head collided with the oil layer, it would be lifted away. This requires a careful balancing act. If the oil is too fluid, it doesn't work as intended. The surface tension would be too little to course correct the head (and the oil probably doesn't stick to the platters). Too viscous, and you can get a wide range of problems. The worst of which being the HDD functions more like a blender. This is that story.
The actual story is very short. Someone tweaked the formula of the oil, and while it looked okay in the lab, in practice the oil became very viscous. So viscous - once powered on - it ripped the head and the arm clean off, completely destroying the platters as they rattled around. Lots of R&D heads rolled (including my Professor), and that was the beginning of the decline of IBM's storage division.
-
Comment on The Amazing Digital Circus: The Last Act in ~movies
DistractionRectangle Link ParentEpisode 3 and 8 is about as "horror"/graphic as it gets. The rest of it is much more toned down. It's mostly about people struggling with the state of their existence. So life, death, body...Episode 3 and 8 is about as "horror"/graphic as it gets. The rest of it is much more toned down. It's mostly about people struggling with the state of their existence. So life, death, body dysmorphia, loss, coping, empathy... That kind of thing.
Description from the pilot:
The Amazing Digital Circus is a psychological dark comedy about cute cartoon characters who hate their lives and want to leave 🎪😀
Depends on the kid, you'll probably have to watch it (or at least parts of it) to get an idea if it's something your kid will be okay with.
I just watched all the trailers, I find the ep 1 trailer to give the best taste of the show. The other trailers are surface level (except for 8/9) and portray the setting but less so the content of the show.
-
Comment on Half-baked idea for metered inline image allowances in ~tildes
DistractionRectangle LinkI have no strong feelings one way or the other We can already post links to stuff, and people can click on it. Inline images/media just removes that tiny bit of friction, and this proposal goes as...I have no strong feelings one way or the other
We can already post links to stuff, and people can click on it. Inline images/media just removes that tiny bit of friction, and this proposal goes as far as allowing users to put that friction back. If tildes isn't doing the hosting, then they don't have to moderate the hosted content (just the comments/links, which they already have to do).
I suppose an argument can be made that defaulting to displaying images inline will encourage more images, but it doesn't actually enable anything we can't do already.
-
Comment on AI Coding agents are the opposite of what I want in ~comp
DistractionRectangle Link ParentQwen3.5 has quirks that need addressing, and Gemma 4 is new enough that there's plenty of bugs in support/quantization. Also, if you have older qwen3.5 quants, those had issues which required you...Qwen3.5 has quirks that need addressing, and Gemma 4 is new enough that there's plenty of bugs in support/quantization. Also, if you have older qwen3.5 quants, those had issues which required you download updated quants.
Recent discussion with code addressing outstanding issues with qwen3.5: https://old.reddit.com/r/LocalLLaMA/comments/1sdhvc5/qwen_35_tool_calling_fixes_for_agentic_use_whats/
-
Comment on AI Coding agents are the opposite of what I want in ~comp
DistractionRectangle Link ParentRight now, low latency pair programming like use just isn't there, and trying to analyze/prompt you as you type requires implicit understanding from the AI, it has to intuitively discern what...Right now, low latency pair programming like use just isn't there, and trying to analyze/prompt you as you type requires implicit understanding from the AI, it has to intuitively discern what you're doing/trying to do. Turn based flows with explicit context works better with the current models/tooling.
The closest thing to what you want, where it prompts you, is probably PR/commit code review. There it gets explicit context on what you're doing/trying to do via commit messages and diffs, and can take its sweet time looking for:
- bugs
- regressions
- test coverage
- algorithmic complexity
- approach
- security issues
- style/code smells
- etc
And prompt you about what it finds.
-
Comment on Three Cheers for Tildes: App updates and feedback (March 2026) — Version 1.5 can search for posts in ~tildes
DistractionRectangle LinkWhen the keyboard is up I cannot see the content of my comment:https://imgur.com/a/9avtYd4 Pixel 4a on stock firmware - android 13When the keyboard is up I cannot see the content of my comment:https://imgur.com/a/9avtYd4
Pixel 4a on stock firmware - android 13
-
Comment on Afroman emerges victorious in ‘Lemon Pound Cake’ defamation case in ~music
DistractionRectangle Link ParentIt's about 2 minutes from here, https://m.youtube.com/watch?v=rE8mtNCsihE&t=1565It's about 2 minutes from here, https://m.youtube.com/watch?v=rE8mtNCsihE&t=1565
-
Comment on Subnautica 2 publisher Krafton's CEO asked ChatGPT how to void $250 million contract, ignores lawyers, loses in court in ~games
DistractionRectangle Link@delphi commented about agentic coding the other day, and the beginning of that comment I think applies well to using AI in general. It's especially relevant to this news:@delphi commented about agentic coding the other day, and the beginning of that comment I think applies well to using AI in general. It's especially relevant to this news:
In the underwater survival video game Subnautica, you can eventually get access to the PRAWN Suit, a mecha that lets you go far deeper into the ocean than you could before. When you first construct it, the in-game computer tells you that it's normal to feel a sense of limitless power when first putting the suit on, and that the months of training suit operators usually get is not to learn how to pilot the thing, but to understand that you're not invincible in it. Claude Code works the same way.
-
Comment on What programming/technical projects have you been working on? in ~comp
DistractionRectangle Link ParentI never got into tabletop (lack of time), but I've always loved procgen and have debugged a few tauri apps in my free time, so put me down for the alpha! As for monetization, that's ultimately a...I never got into tabletop (lack of time), but I've always loved procgen and have debugged a few tauri apps in my free time, so put me down for the alpha!
As for monetization, that's ultimately a personal choice, but I think Aesprite has threaded the needle quite well. They're source available and don't put a limit on how use you personal builds, they just limit forking/redistributing the software. So in effect, they sell the convenience of compiled binaries.
-
Comment on What programming/technical projects have you been working on? in ~comp
DistractionRectangle Link ParentPcie does deliver power through the slot, so low power cards (75W, or maybe it's 50W, you'll have to check for your pcie generation) can actually be powered straight from the motherboard slot...Pcie does deliver power through the slot, so low power cards (75W, or maybe it's 50W, you'll have to check for your pcie generation) can actually be powered straight from the motherboard slot without additional PSU cables.
Beware some arc cards have an annoying fan curve, with mixed reports after a firmware update.
-
Comment on Google’s AI overviews can scam you. Here’s how to stay safe. in ~tech
DistractionRectangle LinkIf want to remove AI overview for yourself or relatives: https://tenbluelinks.org/ It uses the OpenSearch specification to announce the Google search engine with the Url param udm=14 appended...If want to remove AI overview for yourself or relatives: https://tenbluelinks.org/
It uses the OpenSearch specification to announce the Google search engine with the Url param udm=14 appended (which opts out of AI overview). You can do this directly on Desktop browsers + Firefox for Android, but Chrome for Android doesn't allow you to add/edit custom search engines except via OpenSearch
-
Comment on What programming/technical projects have you been working on? in ~comp
DistractionRectangle LinkSo I've resisted using AI for anything major for a while now. I mostly treat it like a pseudo sentient rubber duck that forces me to articulate my problems and helps me explore domains/problem...So I've resisted using AI for anything major for a while now. I mostly treat it like a pseudo sentient rubber duck that forces me to articulate my problems and helps me explore domains/problem spaces (the you don't know what you don't know problem). Chat basically. I ask questions, it spits out something answer shaped, I extract value (e.g. learning domain specific jargon and using that to make better web queries//ai queries).
I've been sticking to local models for a few reasons. Privacy for one. Consistency in abilities/cost/availability. But probably the biggest one is smaller models aren't that capable. It's a tool, but not a crutch. I still have to think critically, grow. I'm not outsourcing my ability to do, to think. What good is a highly capable model if I'm not skilled enough to steer it?
Which leads me to today. Agentic programming is on the rise, and at some point I'm going to have to adapt. Enter oh my pi. Those that trawl HN probably already know what I'm talking about. I just spent a few minutes patching it to work with self hosted providers that use an openai compatible API (it supports ollama's /api/tags endpoint, but not /v1/models like llamacpp and friends). What better way to test my patch than ask the agent to review my work? I gotta say, I was impressed. It's a little slow using qwen3-coder-next w/ 64GB ram + 12GB vram, but ~30 tokens per second is usable.
Code Review Request
Mode
Reviewing uncommitted changes (staged + unstaged)
Changed Files (2 files, +9/-9 lines)
┌────────────────────────────────────────────────────┬───────┬──────┐
│ File │ +/- │ Type │
├────────────────────────────────────────────────────┼───────┼──────┤
│ packages/coding-agent/src/config/model-registry.ts │ +5/-5 │ ts │
├────────────────────────────────────────────────────┼───────┼──────┤
│ packages/coding-agent/test/model-registry.test.ts │ +4/-4 │ ts │
└────────────────────────────────────────────────────┴───────┴──────┘Distribution Guidelines
Use 1 reviewer agent.
Reviewer Instructions
Reviewer MUST:
- Focus ONLY on assigned files
- MUST use diff hunks below (MUST NOT re-run git diff)
- MAY read full file context as needed via read
- Call report_finding per issue
- Call submit_result with verdict when done
Diff
-- I'm held back the diff because formatting it was a PITA
⏳ Task: reviewer
├─ Context
│ ## Goal
│ Review changes to Ollama model discovery inmodel-registry.tsand corresponding test updates. The changes update the endpoint from/api/tagsto/v1/modelsand adjust the JSON payload structure.
│
│ ## Non-goals
│ Do not review unrelated files or changes not in the diff. Do not run tests or build the project.
│
│ ## Acceptance
│ Identify any correctness, compatibility, or safety issues with the API endpoint and payload parsing changes.
└─ Tasks: 1 agents
└─ ✔ 0-ReviewOllamaChanges: Review Ollama API endpoint and payload changes ⟦done⟧ · 334K tokens · 3m7s
Patch is correct ✔ (95% confidence)
The changes correctly update the Ollama model discovery endpoint from /api/tags to /v1/models and a…
Findings: ✘ P0:0 · ⚠ P1:0 · ⚠ P2:2 · ⓘ P3:1
├─ [P2] Model name now set to id, losing human-readable display names model-registry.ts:850
├─ [P2] No fallback for model entries missing id field model-registry.ts:848
└─ [P3] Test mocks use simplified data structure without additional fields present in real API model-registry.test.ts:557
Total: 1 succeeded · 3m7s✔ Read /workspaces/scratch/oh-my-pi/packages/coding-agent/src/config/model-registry.ts:827-861
Error: EXA_API_KEY not found. Set it in environment or .env file.
Deep Search Ollama API v1 models response structure data array id field object created owne… results:5
ⓘ Raw response⟦(Ctrl+O for more)⟧
├─ {
├─ "content": [
├─ {
└─ … 5 more linesCode Search Ollama /v1/models GET response example data array id object created owned_by
ⓘ Raw response⟦(Ctrl+O for more)⟧
├─ {
├─ "content": [
├─ {
└─ … 5 more lines✔ Read /workspaces/scratch/oh-my-pi/packages/coding-agent/test/model-registry.test.ts:550-609
✔ Todo Write 3 tasks
├─ ☑ Review Ollama API endpoint change /api/tags → /v1/models
├─ ☑ Review payload structure changes models→data, name/model→id
└─ ☑ Identify issues with name field reduction✔ Todo Write 3 tasks
├─ ☑ Review Ollama API endpoint change /api/tags → /v1/models
├─ ☑ Review payload structure changes models→data, name/model→id
└─ ☑ Identify issues with name field reduction✔ Todo Write 3 tasks
├─ ☑ Review Ollama API endpoint change /api/tags → /v1/models
├─ ☑ Review payload structure changes models→data, name/model→id
└─ ☑ Identify issues with name field reduction✔ Todo Write 3 tasks
├─ ☑ Review Ollama API endpoint change /api/tags → /v1/models
├─ ☑ Review payload structure changes models→data, name/model→id
└─ ☑ Identify issues with name field reductionReview Complete
Verdict: APPROVED with findings
Summary
The changes correctly update Ollama model discovery from /api/tags to /v1/models and adjust the payload structure accordingly.
Findings
[P2] Model name now set to id, losing human-readable display names
- Body: The change replaces name: item.name || id with name: id. The Ollama /v1/models endpoint returns id in format like phi3-128k:Q8_0, x/llama3.2-vision:11b-instruct-q8_0, or
mannix/llama3.1-8b-lexi:q8_0. Previously, Ollama's /api/tags endpoint returned both name and model fields where name could be a human-friendly alias. The new code discards this potential human-readable name
and uses the raw model ID as both id and name, reducing user experience. - File: packages/coding-agent/src/config/model-registry.ts, line 850
[P2] No fallback for model entries missing id field
- Body: The old code had const id = item.model || item.name which provided fallback if one field was missing. The new code uses const id = item.id with no fallback. If an Ollama /v1/models response contains
a model entry with a malformed or missing id field (e.g., due to API version mismatch or server bug), the entry is silently skipped. While the existing if (!id) continue handles this, the previous
dual-field fallback was more resilient to partial data corruption or API variations. - File: packages/coding-agent/src/config/model-registry.ts, line 848
[P3] Test mocks use simplified data structure without additional fields present in real API
- Body: The test mocks use { data: [{ id: "phi4-mini" }] } but the actual Ollama /v1/models endpoint returns additional fields per the Reddit example: { object: "list", data: [{ id, object: "model",
created, owned_by }] }. While these extra fields are not used by the current code, if Ollama later changes the response format (e.g., removes data wrapper or renames id), tests will not catch the breakage.
Tests should mirror the actual API contract to provide early warning of upstream changes. - File: packages/coding-agent/test/model-registry.test.ts, lines 557-558
Recommendation
The core functionality is correct. Consider adding a fallback for the name field (e.g., name: item.name || item.id) to preserve potential human-readable aliases, and update test mocks to include additional
fields (object, created, owned_by) to better match the real API contract.
Todos
└ Review Ollama API changes
└ ☑ Review Ollama API endpoint change /api/tags → /v1/models
☑ Review payload structure changes models→data, name/model→id
☑ Identify issues with name field reduction -
Comment on Popcorn lung: how vaping could scar your lungs for life in ~health
DistractionRectangle Link ParentAFAIK, EVALI wasn't just limited to THC (weed) vaping. It was predominant, but about a sixth of users just reported vaping just nicotine. It was proposed that Vitamin E acetate was a contributor,...As far as I know, the only lung injuries that have been reported and have investigations on the source have been from bootleg weed cartridges that have ingredients that no legitimate manufacturer would include.
AFAIK, EVALI wasn't just limited to THC (weed) vaping. It was predominant, but about a sixth of users just reported vaping just nicotine. It was proposed that Vitamin E acetate was a contributor, but I think Covid eclipsed EVALI and the CDC stopped collecting data.
https://archive.cdc.gov/www_cdc_gov/tobacco/basic_information/e-cigarettes/severe-lung-disease.html
-
Comment on What games have you been playing, and what's your opinion on them? in ~games
DistractionRectangle Link ParentWeirdly, some people experience motion sickness with higher refresh rates or variable refresh rates. Maybe try setting a 60 fps cap and see if that solves it (on top of all the other things you've...Weirdly, some people experience motion sickness with higher refresh rates or variable refresh rates. Maybe try setting a 60 fps cap and see if that solves it (on top of all the other things you've tried)?
-
Comment on CGA-2026-01 🕹️⛵🛡️ INSERT CARTRIDGE 🟢 The Legend of Zelda: The Wind Waker in ~games
DistractionRectangle (edited )Link ParentYou should compare how both look and choose what you like best. From YouTube, I personally like the look of the gamecube version (which works out as it's the version I own). I just can't stand the...You should compare how both look and choose what you like best. From YouTube, I personally like the look of the gamecube version (which works out as it's the version I own). I just can't stand the bloom and saturation of WW HD.
-
Comment on CGA-2026-01 🕹️⛵🛡️ INSERT CARTRIDGE 🟢 The Legend of Zelda: The Wind Waker in ~games
DistractionRectangle Link ParentI loved the tingle tuner as it allowed for a semblance of co-op and friendly fire :) Bombing your sibling (and wasting their money) was a delight. Less so when they got to do it to you... The QoL...I loved the tingle tuner as it allowed for a semblance of co-op and friendly fire :)
Bombing your sibling (and wasting their money) was a delight. Less so when they got to do it to you...
The QoL mod is something I was unaware of. I set this up a while ago in anticipation of this months CGA, and I also found the default controls... clunky. So I reached for my hammer, Steam Input.
I couldn't get steam input overlay (big picture overlay) working with the dolphin launcher, so added Wind Waker as a non steam game using
/usr/bin/dolphin-emu-noguias the target with this command:%command% -e /home/drect/Documents/dumps/gc/GZLE01.iso --config Dolphin.Display.Fullscreen=True --config Dolphin.Interface.CursorVisibility=0-e is the absolute path to the game iso and the config options force fullscreen and hide the mouse
And now I can play with my steam controller and tweak the control scheme to my liking on the fly!
-
Comment on Save Point: A game deal roundup for the week of December 28 in ~games
DistractionRectangle Link ParentFree game #15 Chivalry 2 Next in 22 hours (two mystery games tomorrow!)Free game #15
Chivalry 2
Next in 22 hours (two mystery games tomorrow!) -
Comment on Save Point: A game deal roundup for the week of December 28 in ~games
DistractionRectangle (edited )Link ParentFree game #14 Trine Classic Collection Next in 11 hours This is actually four games Trine Enhanced Edition Trine 2: Complete Story Trine 3: Artifacts of Power Trine 4: Definitive Edition They are...Free game #14
Next in 11 hours
This is actually four games
-
Trine Enhanced Edition
-
Trine 2: Complete Story
-
Trine 3: Artifacts of Power
-
Trine 4: Definitive Edition
They are 2.5D physics platformers with characters you switch between at will to overcome various environmental puzzles/challenges. Kinda neat and allows for co-op play
Viewfinder on Linux
Ironically, umu launcher actually causes problems. You have to disable this if you're using Heroic or have previously setup another launcher to use UMU. You can also bypass Epic by launching the game directly with the argument
-EpicPortal -
I too was thinking of Disturbed's cover of Sound of Silence, however I don't actually like the recorded version. I much prefer their live cover they performed on Conan.