avery

swe @boundaryml

[Antonio Sarosi @antoniosarosi]

BAML is now an official language on @github with syntax highlighting.

We made the sheep instantly recognizable 🐑🚀

Congrats @boundaryML

[Antonio Sarosi @antoniosarosi]

This is how building compilers in 2026 looks like btw

[Vaibhav Gupta @vaibcode]

Imagine if v8 was tied to the visualization, and you could watch your code execute. or click on teh graph and jump to that line of code.

Or don't imagine and try this early beta in BAML.

[Anish Palakurthi @anishpalakurT]

Announcing the BAML Bounty...
For all power-users of BAML, we're giving away free BAML merch! (t-shirts, stickers, hoodies 🔥🧯).

Share what you built with BAML with #baml → Fill out https://t.co/oglxEbz5eG → Free merch!

Hurry! Supplies are limited to the first 50 posts.

[avery @codeshaunted]

@tsoding This has existed for a long time now and is called a question answering model: https://t.co/6hdwdzshrE

AFAIK models like this are what google used before AI overview to grab relevant snippets from a webpage.

[avery @codeshaunted]

@paularambles

[avery @codeshaunted]

@nexxeln @greptile claude code keeps doing this to me too

[avery @codeshaunted]

@hdevalence people still did this into the 2010s. when we were reverse engineering the network protocol for @darkflameuniv we saw random bits of XML data in the unused bytes of fixed length strings from packet captures. we’re pretty sure it came from uninitialized memory on the server

[avery @codeshaunted]

@tldraw that increases the space needed in-memory by ~33% compared to a typed array, and you get overhead with constant encode/decode… how often is the FULL drawing being synced to warrant this? i’m assuming for updating the server you just send deltas

[avery @codeshaunted]

@tldraw so your in-memory representation of a collection of points is a base64 string?

[avery @codeshaunted]

@massive_endian reporting this tweet for terrorism

[avery @codeshaunted]

@Sceneometry @DisneyPlus At least on Disney+, each episode has 3 different descriptions, "full", "medium", and "brief". Some of them are pretty bad (like this one lol) and each one is seemingly written by a different person.

[avery @codeshaunted]

@Sceneometry @futurama_hulu let's go

[avery @codeshaunted]

@Sceneometry did they show the first episode?

[avery @codeshaunted]

@Sceneometry can’t wait for it to come out

[avery @codeshaunted]

@EnemyCrabGames they automated tub gremlins

[avery @codeshaunted]

@EnemyCrabGames shots 1-5 clearly missed

[avery @codeshaunted]

@WBWarforge did you do this by somehow editing UEFN project with UE5?

[avery @codeshaunted]

It's also notable that SD v1.4 doesn't have this issue. Here are some outputs for the same prompt from v1.4.

[avery @codeshaunted]

@StabilityAI's #StableDiffusion seems to be overfit on a single Apollo 11 photo whenever "moon landing" is in the prompt. First is the original Apollo photo, following are outputs from SD v1.5 for "a photo of a moon landing". @EMostaque any chance you guys fix stuff like this?