pgt
pgt on Hacker News
Jev will continue to do well because people don't actually want to host their own models. The average customer just want an always-on pay-per-use API that has social proof.
Posted by
pgt on Tuesday September 15, 2026 @10:09AM
from the show-hn dept.
https://github.com/theronic/eacl
EACL is a situated ReBAC authorization library inspired by SpiceDB & Google Zanzibar, built in Clojure and backed by Datomic Pro, Datahike, Datalevin or DataScript. EACL is suitable for multi-Peer deployments. I have been working on EACL for ~2 years in my personal capacity and I use it in all my Clojure projects now.- You can try the EACL performance demo: https://demo.eacl.dev/- And play...
This is great, but why did it have to be YAML?
Does anybody have a link where I don't need to download a PDF and zoom and it's still messed up for reading? I just want HTML, mannn.
This is suicide for Gmail.
Pre-October 2025, maybe yes. But now? Couldn't disagree more. There is no insight in this post.
Most Ketamine users are unaware that even occasional usage causes
untreatable ulcerative cystitis, a.k.a. Ketamine Bladder Syndrome. Ketamine thins the urothelium while increasing the collagen to smooth muscle ratio and exacerbating interstitial fibrosis. The syndrome is
untreatable and causes serious problems down the line.
So, for anyone thinking of using Ketamine on the regular or even monthly, there is good evidence that EGCG (epigallocatechin gallate) administered 30 minutes before prevents KBS (Ketamine Bladder Syndrome), in rats: https://www.sciencedirect.com/science/article/pii/S187952261...
This promising rat study from 2015 shows that EGCG extracted from green tea has a protective effect on the bladder, reversing ketamine-induced damage to control levels. So if you are taking ketamine for major depression or recreationally, taking EGCG before may prevent long-term bladder damage.
Summary from the study:
# Objective
To investigate the protective effect of green tea epigallocatechin gallate (EGCG) on long-term ketamine-induced ulcerative cystitis (KIC) using a ketamine addiction rat model.
# Materials and methods
Thirty Sprague-Dawley rats were divided into three groups which received saline, ketamine (25 mg/kg/d), or ketamine combined with EGCG (10 μM/kg) for a period of 28 days. In each group, cystometry and a metabolic cage micturition pattern study were performed weekly. Masson's trichrome study was done to evaluate the morphologic changes. Western blot analyses were carried out to examine the expressions of inflammatory protein [transforming growth factor-β (TGF-β)] and fibrosis proteins (fibronectin and type I collagen) in bladder tissues.
# Results
Chronic ketamine treatment resulted in bladder hyperactivity with a significant increase in micturition frequency and a decrease in bladder compliance. These alterations in micturition pattern were accompanied by increases in the expressions of inflammatory and fibrosis markers, TGF-β, fibronectin, and type I collagen after long-term ketamine treatment. Masson's trichrome stain showed that ketamine treatment decreased urothelium thickness while increasing the collagen to smooth muscle ratio and exacerbating interstitial fibrosis. By contrast, simultaneous EGCG and ketamine treatment reversed ketamine-induced damage to almost control levels, showing the protective effect of EGCG.
hmm, I suspect distributed systems can be modelled with a form of probabilistic borrow-checking as long as you have control all the connected systems.
It matters because it means that lab could not train that model without distilling another frontier model, and their progress would slow once they get properly cut-off. If I funded that lab, I would want to know that.
This is helpful, because practically many functions in the wild have assert-like checks at the top of the function, e.g.
`(if-not key1 (throw (Exception ...))`
...and pre-conditions, e.g. `:pre [condition1 condition2]` do not run when `assert` is off.
Previously:
https://news.ycombinator.com/item?id=40248231tldr; Starlink doesn't work in South Africa, Elon's home country, because the ANC and its lawfare arm ICASA demands they hand over 30% to the State because of BEE laws.
Very cool! Idea: show the definition of the word when you find it, so you can learn new words.
(edit: ah, I see they are shown later)
This is such an obvious feature for Glassdoor, but needs verification to avoid abuse.
In case you want to view the menu, press Tab. Click outside menu items to resume game.
Only a fool would enter their name in this.
I’m surprised to see the model names hardcoded as an enum (e.g. `.sonnet4_6`), instead of a string with model discovery so that the user can select their preferred model without having to get a new app version through the App Store to support newer models.
How to shoot? I can't figure it out. Perhaps made harder by typing on Colemak.
Edit: OK, spacebar shoots, but it's hard to see the cannons flying due to low contrast. And it shoots in the direction of the enemy.
AI threatens the programmer identity.
Ooh, thanks for this. I'm going to try quercetin for my histadine sensitivity in red wine, which causes a skin reaction. I thought it was the sulphites, but after trying older wines (which have fewer sulphites), it made it worse. So I've narrowed it down to histadines. The dilemma here is that wines tend to improve with age, but they contain more histadines.
> "I do wish there were an easier way to move in the ]}]})))}-ness of block ends though."
If he means navigating the AST, there is Parinfer: https://shaunlebron.github.io/parinfer/
Paredit / Parinfer ruined other languages for me. It lets you navigate up/down/in/out of the Clojure AST with keyboard commands and mutate those expressions, e.g. "Split" will split open the current data structure you're in: `(a| b)` =Split=> `(a)| (b)`, where | is caret. Join is the inverse, and it works for all data structures.
Iocaine powder is a fictional, odorless, and tasteless poison from The Princess Bride.
Not a Rogue-like specialist, it's hard – I haven't made it out of level 1 yet.
If query strings are banned, tracking parameters will simply move to the URL, or use `?t=` & `?h=` if those are whitelisted.
This reminds me of the META [^1] II paper [^2]:
> META II is a domain-specific programming language for writing compilers. It was created in 1963–1964 by Dewey Val Schorre at University of California, Los Angeles (UCLA). META II uses what Schorre called syntax equations.
The interesting part of META II is that it can be defined in itself on one page (refer page 8 of the paper).
[^1] Wikipedia https://en.wikipedia.org/wiki/META_II
[^2] Paper: https://dl.acm.org/doi/epdf/10.1145/800257.808896
Datomic was the reason I switched [^1] to Clojure as my primary language in 2014. It was a gamble, but it paid off in the end.
I maintain that Clojure is the best AI-first language due to the lightning-fast iteration via the nREPL and Clojure's token efficiency.
[^1]: https://petrustheron.com/posts/why-clojure.html
thx that looks much better, but I'll forget that syntax in the next 3 months. nowadays I am just casual commenter on HN spending points on contrarian views that I know will be downvoted and I know no one else will say it.
Your device is the ultimate edge. The next frontier would be running models on your wetware.
You need to curate your algorithm. Took me 10 years before I started blocking aggressively and now my feed is amazing with 90% bangers. Twitter is by far the best product in this space. Every other platform is 2+ weeks behind. Twitter is where the news breaks.