Smerity

Brainstorming with an LLM that's glazing you a tad is helpful when you're under glazing by default 🤔

Prague, 23 November 1911 Highly esteemed Mrs. Curie, Do not laugh at me for writing you without having anything sensible to say. But I am so enraged by the base manner in which the public is presently daring to concern itself with you!? that I absolutely must give vent to this feeling. However, I am convinced that you consistently despise this rabble, whether it obsequiously lavishes respect on you or whether it attempts to satiate its lust for sensationalism! I am impelled to tell you how much I have come to admire your intellect, your drive, and your honesty, and that I consider myself lucky to have made your personal acquaintance in Brussels. Anyone who does not number among these reptiles is certainly happy, now as before, that we have such personages among us as you, and Langevin(3) too, real people with whom one feels privileged to be in contact. If the rabble continues to occupy itself with you, then simply don't read that hogwash, but rather leave it to the reptile for whom i

einstein sent this to curie in 1911 when she was being harassed by tabloids. it contains everything you’d want in such a letter:

(1) your haters are trash
(2) you’re a baller, a true queen
(3) i have determined the statistical law of motion of the diatomic molecule in planck’s radiation field 🧪⚛️

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.

Very few people see a full video on YouTube apparently, even if it lasts just 10-15 minutes.

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 :(

My contribution to a discussion at hiddenstates.org on explorables/user interfaces for controlling ML tools
> We're trying to rig soundboards to control LLMs thinking there's a well defined interface underneath when it's actually a button that drops fertilizer into the river of a complex ecosystem.

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...

A new-to-me perspective on passive consumption of media

(Also: I'm gifting you some psychological obliteration here!)

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.

congratulations, @ian-goodfellow.bsky.social, for the test-of-time award at @neuripsconf.bsky.social!

this award reminds me of how GAN started with this one email ian sent to the Mila (then Lisa) lab mailing list in May 2014. super insightful and amazing execution!

From: torvalds@klaava.Hels Subject: What would you like to see most in minix? Date: 25 Aug 91 20:57:08 GMT Organization: University of Helsinki Hello everybody out there using minix - I'm doing a (free) operating system (just a hobby, won't be big and professional like gnu) for 386(486) AT clones. This has been brewing since april, and is starting to get ready. I'd like any feedback on things people like/dislike in minix, as my OS resembles it somewhat (same physical layout of the file-system (due to practical reasons) among other things). I've currently ported bash(1.08) and gcc(1.40), and things seem to work. This implies that I'll get something practical within a few months, and I'd like to know what features most people would want. Any suggestions are welcome, but I won't promise I'll implement them :-) Linus

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 🤣

The part that's insane to me is that external hyperlink isn't even opposite to "central host". Reddit and X aren't that different from each other and most of Reddit is linking to outside content w/ discussion returning on site.
X decided they ... didn't want free content to spur discussion ..? 🤔

I loved seeing @commoncrawl.bsky.social's use grow after contributing a decade ago. I'm expecting the Bluesky API + firehose to have an even faster pace of innovation given it's so accessible, has a small but potent dataset (total / incremental size), and directly tracks the pulse of the community!

Perhaps closer to how I've played with embeddings/discovery on English Wikipedia in the past but likely similar :) Also sticky taping in my personal bookmarking system for the discovery aspect, may release if works well.
The Bluesky API is complex but the docs are strong!
Jetstream is also JSONL 🔥

All the examples from the Bluesky `atproto` Python API as a single text file with delineation All the examples from the Bluesky `atproto` Python API as a single text file with delineation - all_examples_from_bluesky_atproto_python_api.py.txt

I've got good mileage for playing by a slightly more advanced `cat atproto/examples/*.py` and feeding it into an LLM
gist.github.com/Smerity/f896...
P.S. I don't mean to make a "Bluesky posting about Bluesky" post, but honestly, if I'd have been playing around with this regardless, so ... 😅

I'm both hopeful about and want to contribute to the ecosystem that could build up around this. There are many opportunities that I can't see existing elsewhere.
The documentation is also far less painful than I'd have thought. Bluesky's slow rolled start is a strong help in that regard!