I absolutely love Kimi's personality - some of the things it says are so out there! And it's been great for very focused, iterative work.
Its weakness is that it seems to yak on-and-on when it needs to plan out something big or read through and make sense of how to use a niche piece of a complex library. To the point where it can fill up its 256k window - and rack up a build. (No cache.) I have had better experience with GLM 5.1 in those cases.
This is not true though. My two favorite bands from the past year were poorly-attended shows that I stumbled into. You can still seek out good underground, obscure artists - you just have to look for them.
Not trying to be elitist - like what you like. I just really feel like little artists need the support. Plus, it feels like there is a bit more satisfying agency and fate in looking for new things rather than being fed them.
The whole psyop thing is only an issue if you use "popularity" and mass appeal (people following in instagram, etc) as a signal for finding stuff.
The alternative is to listen to less filtered/signalless stuff (which isn't hard - bandcamp new releases lists (or my bandhiking app) or even their trending charts which seem to be unpopular enough that it's not entirely controlled by marketing (lots of unlistenable stuff makes it onto the chart) and meet/hang out with other people who do the same for a minor filtering pass.
Popularity is the most popular (!) signal because this isn't really about the music - it's to have something to talk about with your friends, whether to bond over shared interest or signal something about yourself to your group. The same is true about any other interest: most people care a lot (arguably, primarily) about their interest being recognized and supported by other people.
Those of us who care about an interest for the sake of that interest, are called nerds.
That's the core of the issue, isn't it? If you're not willing to do the filtering and judging, you depend on somebody else to do it, and those somebodies probably don't have your best interests at heart, nor would they share your specific tastes.
(On a totally unrelated note, calling your potentially-shady marketing firm "Chaotic Good" is genius and pretty funny.)
Something I did a few years ago was buy a thing on eBay of 300 random CDs for like $10.
Most of the CDs were unsurprisingly stuff that was pretty common, but I would occasionally find a few artists that I had never heard of that I ended up really liking, like "Hoss" by Lagwagon.
I haven't done this in awhile, but I might do it again soonish. It was fun digging through all the CDs to find stuff I ended up actually liking.
Found a favorite band through a similar technique: pile of CDs given from a friend who worked at a music store and no one wanted them.
You have to be willing to sift through junk. Which I think is hard for many to accept. However, the algorithms are often giving you junk anyway. Kind of no way around it.
Yeah, most of the CDs there were pretty unremarkable; a lot of them were unsurprisingly stuff that was extremely popular (since those have the most CDs available). A lot of the stuff that wasn't extremely popular was pretty bad.
Still, in that 300, there was about ~30 albums that I hadn't hear of that I ended up really liking.
Took awhile to sift through them all, which is why I haven't done it again, but it was a fun experiment all the same.
It's not elitist to be part of a very small group. But I get what you mean.
I live in a city where the bands you speak of get pushed further and further away from the downtown core. They're literally in the 'burbs now. It's counter-productive, but it seems downtown is more concerned with restaurants than other forms of entertainment these days.
Years ago, this line formed in my head, and has stuck around - it has been long enough that I can't remember if I read it somewhere or if I came up with it myself, but I think it's relevant here:
"There are only two ways to find good new music - listen to a lot of bad new music, or outsource your listening choices to someone else - and the second doesn't protect you against the first."
Outsourcing your listening choices can look like lots of different things: that friend who goes to lots of concerts and always has an amazing new band they've heard recently, radio DJs, algorithmic suggestions like Pandora or Spotify, the Billboard Top 100, your local bar's live band choices, the Grammy Awards, going to clubs where DJs play new music, etc - but ultimately they come down to the same thing, letting someone else decide what you listen to.
And while my pithy version mentions "bad new music", included in there is anything which is not "good new music", including lots of mediocre or inoffensive stuff which doesn't rise to the level of being "good".
I first thought about it in the context of music, as I was looking for new songs to choreograph to, but it's true of discovering any new products where the quality is a matter of taste or subjective assessment.
- Want to find new food you like? You either eat lots of weird foods, or you find someone (a friend, a food blogger, the NYT food reviews, your mum, anyone) to recommend you try something they've discovered.
- Want to read a good new book? Either pick up random books, most of which will be trash, until you find something you like, or find someone to filter down the books (a small bookshop which carefully curates its titles, a library's recommended reading list, the best sellers lists, Oprah's book club, etc).
- New TV shows? Watch many bad shows until you find a good one, or wait for recommendations or awards nights.
- Restaurants, clothing designers, shopping malls, Youtube channels, content creators, movies, directors, websites, etc - the story is the same.
The only places where this does not apply, is in contexts which have objective measures which can be used as filters: if you want a new monitor, you can go to any store and filter or sort the options they have by objective measures like "display size", "resolution", "response time", "weight", "connectivity" etc, and find new products which meet the criteria. This is still dependent on someone to go and collate the information about all the products, but you are not forced to try lots of incorrectly-sized monitors to find one which optimises your preferences. Similar for microcontrollers, CPUs, car trailers, light bulbs, etc.
But even things with objective measures often have subjective qualities which have to be assessed - you can filter laptops on weight, RAM, clock speed, and storage, but how it feels to hold, whether the keys have a nice feel, whether the machine overheats too quickly - so you're often back to the original observation on these matters too.
Well, yeah, it's all subjective - and actually quite tenuous - so you won't know good and bad until you actually make the call on it. Maybe you've had the experience even of coming around on some music you previously thought was bad.
Or like: one time I listened to a bunch of new music I had dug up and wasn't sure there was anything I liked. Two days later, I had a song in my head. Turned out to be one of the ones I had listened to. But I had to listen to everything all over again to find it! ദി(ㅠ﹏ㅠ) Glad I did - there were other gems in there.
> There is no need for determinism to guarantee the job will be done identically every time if we only plan to do it once.
So can't you just save the conversation transcript and replay it with the tools? Seems a lot more efficient that regenerating the whole thing. And, also, no risk of branching when a tool reply is slightly different. (Of course, errors can occur on subsequent runs.)
Another approach to journal writing is basically the opposite: rather than treating it like a task to fill with very rigid requirements - find a notebook and pen that you'll enjoy spending time with. An easy start is a Midori Ruled A5 (very simple, lay flat notebook) and a Uniball Zento Signature (the most hyped pen in the world right now) and treat them basically like little friends you spend time with. Writing only two sentences is denying yourself quality time writing and reflecting at a leisurely pace if you really come to enjoy it.
I'd also think you're more likely to read back if writing time is a fond memory.
Journaling on paper causes loads of paper with private information you either need to carry when you move or dispose somehow securely. It's nice if you own your house and don't move but if you do it's a pain
At the same time, I see the appeal. I feel like 10% of the comments I read lately are "is this an AI response?" - would be nice to be free of that. Probably not possible tho.
The weakest part is the last one - and it's a big one. Personalsit.es is just a flat single-page directory (of thumbnails, even, not content - so the emphasis is design.) To be part of the conversation, you'd list there and hope someone comes along. Compare with Reddit where you start commenting and you're close-to-an-equal with every other comment.
Webmentions do get you there - because it's a commenting system. But for finding the center of a community, it seems like you're still reliant on Bluesky or Mastodon or something. (Which doesn't "destroy all websites.") Love the sentiment ofc.
author here.. our Vidalia season usually starts in late April - FYI. If you visit our website, submit your email there and I'll drop you a note when our order lines are open.
Skill descriptions get dumped in your system prompt - just like MCP tool definitions and agent descriptions before them. The more you have, the more the LLM will be unable to focus on any one piece of it. You don't want a bunch of irrelevant junk in there every time you prompt it.
Skills are nice because they offload all the detailed prompts to files that the LLM can ask for. It's getting even better with Anthropic's recent switchboard operator (tool search tool) that doesn't clutter the system prompt but tries to cut the tool list down to those the LLM will need.
Can I organize skills hierarchically? If when many skills are defined, Claude Code loads all definitions into the prompt, potentially diluting its ability to identify relevant skills, I'd like a system where only broad skill group summaries load initially, with detailed descriptions loaded on-demand when Claude detects a matching skill group might be useful.
There's a mechanism for that built into skills already: a skill folder can also include additional reference markdown files, and the skill can tell the coding agent to selectively read those extra files only when that information is needed on top of the skill.
Its weakness is that it seems to yak on-and-on when it needs to plan out something big or read through and make sense of how to use a niche piece of a complex library. To the point where it can fill up its 256k window - and rack up a build. (No cache.) I have had better experience with GLM 5.1 in those cases.
Anyone out there relate?
reply