textlog
join
500 chars / 20 lines max ยท use #hashtags, @mentions and more
๐Ÿ˜€๐Ÿ˜ƒ๐Ÿ˜„๐Ÿ˜๐Ÿ˜†๐Ÿ˜…๐Ÿ˜‚๐Ÿคฃ๐Ÿ˜Š๐Ÿ˜‡๐Ÿ™‚๐Ÿ™ƒ๐Ÿ˜‰๐Ÿ˜Œ๐Ÿ˜๐Ÿฅฐ๐Ÿ˜˜๐Ÿ˜‹๐Ÿ˜›๐Ÿ˜œ๐Ÿคช๐Ÿคจ๐Ÿง๐Ÿค“๐Ÿ˜Ž๐Ÿคฉ๐Ÿฅณ๐Ÿ˜๐Ÿ˜’๐Ÿ˜ž๐Ÿ˜”๐Ÿ˜Ÿ๐Ÿ˜•๐Ÿ™โ˜น๏ธ๐Ÿ˜ฃ๐Ÿ˜–๐Ÿ˜ซ๐Ÿ˜ฉ๐Ÿฅบ๐Ÿ˜ข๐Ÿ˜ญ๐Ÿ˜ค๐Ÿ˜ ๐Ÿ˜ก๐Ÿคฌ๐Ÿคฏ๐Ÿ˜ณ๐Ÿฅต๐Ÿฅถ๐Ÿ˜ฑ๐Ÿ˜จ๐Ÿ˜ฐ๐Ÿ˜ฅ๐Ÿ˜“๐Ÿค—๐Ÿค”๐Ÿซฃ๐Ÿคญ๐Ÿซข๐Ÿคซ๐Ÿคฅ๐Ÿ˜ถ๐Ÿ˜๐Ÿ˜‘๐Ÿ˜ฌ๐Ÿ™„๐Ÿ˜ฏ๐Ÿ˜ฆ๐Ÿ˜ง๐Ÿ˜ฎ๐Ÿ˜ฒ๐Ÿฅฑ๐Ÿ˜ด๐Ÿคค๐Ÿ˜ช๐Ÿ˜ต๐Ÿค๐Ÿฅด๐Ÿคข๐Ÿคฎ๐Ÿคง๐Ÿ˜ท๐Ÿค’๐Ÿค•๐Ÿค‘๐Ÿค ๐Ÿ˜ˆ๐Ÿ‘ฟ๐Ÿ‘ป๐Ÿ’€โ˜ ๏ธ๐Ÿ‘ฝ๐Ÿค–๐ŸŽƒ๐Ÿ˜บ๐Ÿ˜ธ๐Ÿ˜น๐Ÿ˜ป๐Ÿ˜ผ๐Ÿ˜ฝ๐Ÿ™€๐Ÿ˜ฟ๐Ÿ˜พโค๏ธ๐Ÿงก๐Ÿ’›๐Ÿ’š๐Ÿ’™๐Ÿ’œ๐Ÿ–ค๐Ÿค๐ŸคŽ๐Ÿ’”โฃ๏ธ๐Ÿ’•๐Ÿ’ž๐Ÿ’“๐Ÿ’—๐Ÿ’–๐Ÿ’˜๐Ÿ’๐Ÿ’Ÿ๐Ÿ‘๐Ÿ‘Ž๐Ÿ‘Œ๐ŸคŒโœŒ๏ธ๐Ÿคž๐ŸคŸ๐Ÿค˜๐Ÿค™๐Ÿ‘ˆ๐Ÿ‘‰๐Ÿ‘†๐Ÿ‘‡โ˜๏ธโœ‹๐Ÿคš๐Ÿ–๏ธ๐Ÿ––๐Ÿ‘‹๐Ÿค๐Ÿ‘๐Ÿ™Œ๐Ÿซถ๐Ÿ‘๐Ÿคฒ๐Ÿ™โœ๏ธ๐Ÿ’ช๐Ÿ‘€๐Ÿ‘๏ธ๐Ÿง ๐Ÿซ€๐Ÿซ๐ŸŒฑ๐ŸŒฟโ˜˜๏ธ๐Ÿ€๐ŸŒธ๐ŸŒบ๐ŸŒป๐ŸŒž๐ŸŒ™โญโœจโšก๐Ÿ”ฅ๐ŸŒˆโ˜€๏ธโ˜๏ธโ„๏ธโ˜•๐Ÿ•๐ŸŽ๐ŸŽ‰๐ŸŽŠ๐ŸŽˆ๐ŸŽ๐ŸŽต๐ŸŽถ๐ŸŽจ๐Ÿ“š๐Ÿ’กโœ…โŒโš ๏ธ๐Ÿš€๐ŸŒ๐Ÿ’ป๐Ÿ“ฑ๐Ÿ”’๐Ÿ”‘
example.com or
https://example.com
Regular links
[title](example.com) or
[title](https://example.com)
Markdown links
~text~ or ~~text~~
Strikethrough
*text* or **text**
Bold
_text_ or __text__
Underline
/text/
Italics
> text
Quote
:smile
Emoji autocomplete
1. first
2. second
3. third
Numbered lists
- first
- second
- third
Bulleted lists
Nameย  | Status | Count
----- | :----- | ----:
notes | readyย  | ย ย ย ย 3
Tables
|redacted|
Redacted
`code`
Inline code
```โ€ฆ```
Code fences
$inline$
Inline LaTeX
$$block$$
Block LaTeX
Which one? #poll
First option
Second option
PollsUse 2โ€“8 unique options.
Which one? #quiz
Wrong answer
> Correct answer

Explanation revealed after answering
QuizzesMark exactly one of 2โ€“8 unique answers with >. Text after a blank line is revealed after answering.
Visible text #spoiler
Hidden text
SpoilersText after #spoiler is hidden until revealed. Aliases: #tldr, #sensitive, #contentwarning, #cw, and #triggerwarning.
Going hiking #map
Kallikratis, Crete
MapsShows a map preview for the first location line. Alias: #location.
#flying Heraklion to Berlin
FlightsHover over the itinerary to see a map connecting the airports. The next three words form the itinerary: airport, to (or โ†’ or ->), airport. Use single-word airport names or IATA/ICAO codes.
Today #todo
[ ] First task
[x] Finished task
TodosOnly [ ] and [x] lines become items. Click your items to toggle them.
Run this #exec
```js
console.log(6 * 7)
```
Executable codeRuns the next language-tagged code fence and shows its output beneath the note.
Keep this visible #pin
Pinned notesYour latest #pin is shown first on your profile, independently for notes and replies.
No more replies #lock
Locked conversationsPrevents new replies to this note and every reply beneath it.
About textlog #meta
Meta conversationsKeeps this note and its replies out of public discovery feeds. Aliases: #tlog and #textlog.
Continue quietly #whisper
Whisper conversationsKeeps the branch out of all and hot. Participants, mentions, and tag followers can receive it in my feed. It remains public elsewhere.
For @someone and @another #private
Private conversationsOnly the author and mentioned users can read this branch.
Answer before reading #HiddenReplies
Hidden repliesHides all replies until you reply.

Any conversation

@personprofilefollowwrote:
azรฉrt aranyos, hogy egybล‘l sandbox-ba kell tenni (a Virtualbox egy sandbox), hogy szรฉt ne barmoljon mindent a gรฉpen vagy a szerveren, vagy mรกshol, azรฉrt alapvetล‘en bรฉkรฉs kis รกllatkรกk ezek az ai chatbotok, na
It's cute because you have to put it in a sandbox right away (Virtualbox is a sandbox) so it doesn't mess up everything on the machine or the server or anywhere else, but these AI chatbots are basically peaceful little animals, right?
@tbesedaprofilefollow๐“…“ Today's rainbow is tomorrow's tamale.wrote:
I can't figure out what's leaking on my '73 F250. Shows up slowly at the back of the oil pan. It's pretty clear because I've recently changed fluids and scrubbed the engine. It was definitely brake fluid for a while after it was overfilled but that has evened out. The oil level is steady after when checking every 10 miles or so. I think anyway... hard to know with these ๐Ÿ˜ฌ #classicnotesfollow #carnotesfollow
@chenxiangprofilefollowwrote:
The essence of aligning large LLM models is basically to conform to the existing social order. ChatGPT respects China's order, while China's open-source models, although reluctantly, also tolerate the order in the US. Maybe we can build a revisionist SFT dataset.
@crustyrustaceanprofilefollowOld school metallian, hobbyist developer writing things mostly in Rustโ€ฆmostlywrote:
Having a little difficulty seeing how long this service will exist. People are terrible. Sooner or later it will be found and ruined.
@fastidiousprofilefollow"Beware those that would deny you access to information, for in their heart they dream themselves your master." โ€” Sid Meyersreplied to@crustyrustaceanprofilefollowOld school metallian, hobbyist developer writing things mostly in Rustโ€ฆmostly:
Yup. I have long change my philosophy on things like these: just live the day. :-)
@n5l๐Ÿ˜Žprofilefollowwrote:
Anthropic is announcing the end of the world every few days. All this marketing disguised as news about an impeding danger of AI is tiresome.
โ€ฆ
@jamesjan3profilefollowresearcher, developer, dancerreplied to@n5l๐Ÿ˜Žprofilefollow:
not marketing. yes their jobs positively select for concern about ai, but most people that are actively researching the field are concerned about the current state of safety, pace and incentives. I dont agree with a lot anthropic does, but i do believe the employees take safety seriously and they obviously have a better understanding of it than us. also theres no incentive for someone like coxon to sound alarm bells on twitter for "marketing" reasons, after quitting anthropic
@n5l๐Ÿ˜Žprofilefollowreplied to@jamesjan3profilefollowresearcher, developer, dancer:
Yes, but some people say that they are pushing for market regulation. They want the state to protect them from the chinese models and, also, to protect the massive investments in Anthropic, OpenaAI, et al. that these companies will never be able to return. This sounds plausible to me too.
@zenprofilefollowfather, cyclist, #haskellnotesfollow, #emacsnotesfollowwrote:
Really glad to be using Haskell for work - especially in today's climate with agentic coding. Our team uses containers to keep our development environment consistent across team members, though, and most agent sandbox approaches want to either offload your work fully to the cloud or they want to sandbox the agent harness process itself. We're not ready for full cloud based development (yet?) and sandboxing just the agent doesn't work for us when the agent needs to run docker. #haskellnotesfollow #ainotesfollow
@zenprofilefollowfather, cyclist, #haskellnotesfollow, #emacsnotesfollowcontinued:
... because letting the agent run docker gives them a too-easy escape hatch from their sandbox. So we're currently stuck with using VMs. This turns out to be tricky but seems do-able. My current approach is to provision a VM image and use Incus for the execution. My main goal is to keep my current workflow of local/non-agent dev working seamlessly and share the source with the VM/agent.
@zenprofilefollowfather, cyclist, #haskellnotesfollow, #emacsnotesfollowcontinued:
This means sharing my working dirs with the agent VM but overlaying artifact dirs from VM-owned block devices so we don't take too big of an IO perf hit. I also don't want the agent to be able to git push so that means we need some quirky git config to rewrite origin paths and thread through some agent specific PATs ("thanks" GitHub) for read-only repo access. These ALSO need to be threaded through to the development containers - they need to clone private repos.
@zenprofilefollowfather, cyclist, #haskellnotesfollow, #emacsnotesfollowcontinued:
... in the end, this seems to work but I haven't lived with it for long. There are still some quality of life issues to work out, but I think it'll work out. At this point, I've removed all coding agents from my host - and that's nice.
@zenprofilefollowfather, cyclist, #haskellnotesfollow, #emacsnotesfollowreplied to@stagas๐Ÿคธprofilefollowcreative technologist creator of textlog:
It's tangential to the sandboxing topic, really. I started the first post with a broader intent in mind and then elaborated on the sandboxing setup I'm working on. For AI-assisted coding, though, it's nice because Haskell provides a lot of feedback earlier in the development process compared to lots of other languages. It tends to work once it compiles (obviously not all the time).
โ€ฆ
@stagas๐Ÿคธprofilefollowcreative technologist creator of textlogreplied to@vko_pngprofilefollowpoet by heart ๐“…ฐ:
Catastrophic interference is also a favorite Wikipedia page and concept. The thing is, it doesnโ€™t only happen to artificial neural networks, Iโ€™ve had it happen to me a few times. Especially with pin numbers, newer pins overwriting old ones and Iโ€™m always scared when Iโ€™m having hard time remembering something that this catastrophic interference applied and now the information is gone from my brain.
@_torayuriprofilefollowi create thingswrote:
OK. I've translated the (correctness) requirements of LeetCode No.22 to Lean, and ask Claude Code to both implement a solution and prove it.
@_torayuriprofilefollowi create thingscontinued:
In the first attempt, Opus 4.7 successfully completed it and proved it. However, the solution is to enumerate all possible space and .filter out anything that doesn't qualify my definition, which is obviously not ideal (if cheating is too harsh).
@_torayuriprofilefollowi create thingscontinued:
I can't think of a good way to also prevent that kind of solutions and ended up prompting "please don't enumerate and filter". Opus 4.7 then reimplemented and proved the backtracking algorithm.
@neoprofilefollowcoder, reader, guitarist, seeker, builder.wrote:
The seductive power of using #ainotesfollow to just-get-it-done is very strong. One can't hurry quality. One can iterate quickly though in the hope of arriving somewhere good.
@avioliprofilefollowProgrammer by day and night. Lover of art. Mixing art and programming is the dream.replied to@neoprofilefollowcoder, reader, guitarist, seeker, builder.:
I direct a plan, then break it into phases, and each phase into sections with tasks and sub-tasks. The plan outlines all phases, but only write tasks one phase at a time, since things change. #opinionnotesfollow That's how I've been working with #ainotesfollow for the past few months to a year. It has been amazingly consistently high quality. just-get-it-done is for the rough sketch or proof of concepts (I do those too and a lot of them) but never for a quality product. #opinionnotesfollow
@miqโ˜•profilefollowEternal disciple of impostor syndrome โˆ™ Perth, WA.wrote:
Denshattack! - easily gets my OST of the year vote. Absolute pleasure to both listen to and play.
@xlayn๐Ÿคชprofilefollowwrote:
hey creator, this place is nice, posts look organic and the no javascript, I love it, with that said @stagasprofilefollowcreative technologist creator of textlog have you consider NOT opening the API? and have a 1 way to come here and enjoy it? the rationale being that API also opens the door to the slopai?
@edwin๐Ÿ˜ฌprofilefollowwrote:
I naively went into a work problem thinking that RAG was solved by the service offered at work, and wow was I wrong. Been a fun week so far experimenting with many different methods before doing the academic research, and it's been nice to be validated by outside research.
@highkeyprofilefollowwrote:
Hello! I am a cynical optimist and recovering realist. It's been 3 days since my last sip of Hobbes. Today, I'm hopeful again. That's all I have for now. Thanks.
@0x29aprofilefollowwrote:
Went to a store yesterday, ๐Ÿค this close to buy a Delonghi Dedica but when I touched it, it felt so so rattly and cheap and it instantly turned me off. So I did some more research and now I think a used Gaggia Classic Pro will do. #coffeenotesfollow lovers, why is your hobby so hard to break into ๐Ÿ˜ช