If done well the proper env setup will help end users as much as LLMs. For proper play, human or machine, you need the equivalent of GET and POST - knowing you can hammer the test Sqlite database as much as you want given it's either resettable sandbox or only running read only queries.

I had a not dissimilar situation with a complicated Pandas query recently. All of the major LLMs defaulted to the same mistaken thinking and required hard nudges. First LLM that seamlessly integrates interactions with the user's own tooling will do so much better given the LLMs can self correct.

There are similarity and differences to tweetstorms and I think that's a great example of where it transitions to marketing / hype. The first tweet in a tweetstorm is usually hype and noise and promise but no value.
"Five amazing secrets that gradient optimizers don't want you to know! 🧵 1/9"

I still think there's strong value in the annoyance you're experiencing 😅 The extra effort you use is gifted to everyone as a richer contextualized message, and I think below the fold encouraged above the fold to be marketing/noise instead of signal.
bsky.app/profile/smer...

tldr; as a language modeler I always thought compression was intelligence, and short tweets force compression and hence intelligence :)

Longer posts on Twitter actually kill a core Twitter feature imho.

There's value and art in a tweet compressing long form information. This can be done by anyone, not just the original author.

The feed becomes a high level "skim reader", progressing to depth when interested piqued.

Been a while since I've seen parse trees, though I'm partial to CCG 🤣

The Dell XPS 13 with Ubuntu Linux native was what I tried last. It's close but there's still a few hardware issues and most importantly a weird system lag with Google Chrome that can drive me insane. If I can't type without random double presses it no longer counts as a functional device 😭

I fought for some time to keep using a Linux laptop but even for those which promise "first class Linux support" seem to fall short. I dual wielded a Linux laptop and a Macbook Pro, slowly transitioned to just the latter :(

I think with a default chronological pipeline it's harder for (3)/(4) to gain traction. When the points rely on rage / clickbait / etc to propagate there's more active work needed, with algorithmic feed redirecting air to the spark. Chronological + local interaction helps minimize (1)/2) as well?

The fact that most wouldn't pay money for passive consumption "there" (reels, feed, ...) but do actively pay with the tick tock of our mind (hours spent, enriching the content locally, ...) is fascinating. If this is psychological obliteration we're very much an active participant in it.

The fact that most wouldn't pay money for passive consumption "there" (reels, feed, ...) but do actively pay with the tick tock of our mind (hours spent, enriching the content locally, ...) is fascinating. If this is psychological obliteration we're very much an active participant in it.

Imagination feels like a double edged sword "there". Too little and you consume only as "there" actually exists, nothing novel. Too much and you enter Don Quixote mode where "no history has more reality in it". Aphantasia may seem an accidental defense, preventing excessive self enriching?

A new perspective to me as well which explains why I'm always interrogating when passive via comparative analysis of "here"/"there", intending to pull back analogical wisdom to my present situation.
I think that's as much protective delusion as practical defense.
news.ycombinator.com/item?id=4226...

As an Australian I don't think it will be effective (enforcement or result), opens the possibility of broader tracking, and likely doesn't mitigate the harms they're trying to avoid either. It definitely would have stunted my learning as a teen.
A contentious issue regardless ¯\_(ツ)_/¯

Out of curiosity, as I'm not in deep in the panadas ecosystem, would polars fill a similar need for you or was there something different?

Glad I found you again! You're the type of follow I didn't want to lose in the Twitter migration ^_^
I appreciated the bridge app but it was definitely slow, had many false positives I filtered manually, and won't catch those moving after I migrated.

This is great! I'm mentally adding it to the same collection as Linus first announcing Linux via mailing list ^_^

What got you this time? Last "yeah I definitely need to play with nightly" for me was Flex Attention - which admittedly was pretty recent too 🤣