arshxyz

Scientists say this is the first occurance of the word "model" in the title of a post on HN that does not refer to a neural network of some sort since 2023

This is amazing. How is the Did You Know section generated?

> AI’s benefits are shared by everyone

Okay so where's my free access to Claude Code

As a rule of thumb if you see "It didn't stop there" you're supposed to stop reading

The prose is extremely grating to read. The tables presented are redundant. The entire article could've been 50 words. Give your own AI generated article a read once or twice before posting slop

even if all meta employees see the source they cannot read messages as long as keys as generated on-device, which is verifiable

What is the purpose of gcc in the new age of AI? Can't you ask your agent to build a compiler if you want to run Hello World?

Video game piracy isn't new and indie games have always been trivial to pirate. In fact a lot of indie studios opt to put their catalog up on GOG, which is a famously DRM-Free store.

By far my favourite vibecoded clone has been a WASM port of CSGO surf maps: https://surfd.net/

If you stretch the definition of "previously-best 2026 model" a bit, Haiku 5.5 outperforms Opus 4.6 which was their best at some point in 2026 while being 10x cheaper.

You'd think with all those tokens they'd be able to vibecode internal replicas of these randomly selected websites without having to send requests outbound

The README is geared towards technical people (complete with the HN screenshot) but when I see a tool like this all I can think of is how helpful this would be for my mom when I'm trying to tell her how to download and print a document over the phone

The Wake Lock API was made for stuff like this: https://developer.mozilla.org/en-US/docs/Web/API/WakeLock

> Meta’s new general-purpose AI agent Muse sent him an unsolicited notification referencing a thread between him and a co-worker over Apple Messages. Aten said he never granted Muse permissions to read his messages

Can someone explain how this is possible? If an app can do this without Full Disk Access or a popup of some kind that's a way bigger lapse on Apple's part than Meta's.

> a typical client-server exchange creates a trail of user data, like the client’s IP address or TLS fingerprint. This level of visibility can be a burden.

Does Cloudflare's WAF (which relies on TLS Fingerprinting) stop working if OHTTP is enabled? If not, does this imply the client metadata is read and processed by Cloudflare but not passed on to the application server?

<50MB .dmg and no electron? I'm sold

Who needs Sonnet 12817 when 5 is maintaining is already maintaining Omarchy? And it's only 5% JS!

Cloudflare WARP suffered a similar downgrade recently when it went from a nifty menubar app on macOS to a vibecoded window built with flutter

For some reason slop UI labels always refer to the user in second person. Solid tell when something says "Needs You" instead of "Requires Attention"

This used to happen a lot with GPT 5.4. It would start outputting entire sentences in Korean for no apparent reason.

Cost per Token is not representative of the actual cost incurred because a cheaper model can waste a lot of tokens to accomplish the same task. Also this does not consider cost for cached tokens which can often make a "cheaper" model more expensive even when total tokens spent are constant

I had Jev play a bunch of games testing:

- Various encoding schemes

- Belief tracking

- Susceptibility to prompt injection

It was surprisingly good at decoding Base64 but failed at something far simpler for humans: ROT13.

I don't understand the point of putting visible Agent Instructions on the page taking up real estate while a human has the page open in a browser. Won't an agent be able to trivially extract these instructions if they came with a display:none rule?

Amazing how many products exist purely due to Google's lack of motivation to do things right. Will be using this a lot.

Does Google Apps Script expose an API to do this though? It would be easier to download an html page from within GDocs without having to run something locally.

> Start llama-server on your model and run a prompt

If this is supposed to target closed-weight models it would be naive to assume they will work out of the box with llama

way more fun version of this went viral on twitter a while ago: https://youraislopbores.me

3 different shades of blue intersecting each other representing different models. If you want a graph to be easier to read use colours that aren't visually similar

Dreadful color-coding on the output tokens table

They seem to be using the provider's harness for each