Mahy

what is wrong with gpt 5.5 it's been super dumb for days now (Opus aswell but thats another story) For instance, I literally asked it to create a website based on the reference images I provided. It spent 17 minutes thinking, only to give me an HTML file where the “website” was basically just the same images slapped into tags.

Holy sh*t 😳 Le gros chaton by Mistral beats both opus 4.8 and gpt 5.5 in DeepSWE for less than 1% of the cost! Interestingly enough it gets worse when it reasons more

After fable is gone, I finally understand how the 4o gpt AI psychosis feels like...

@AdamHoltererer @robinebers lmao Codex.

@_mahybe @robinebers But are you gonna build the bot with Codex or Claude?

blocking 10 accounts like this every single day

Is it just me or is claude fable 5 so bad now?

@huntharo @jxnlco Yep use Hyperframes instead.

@jxnlco Check the license. Do not use remotion.

Fable is so so good at remotion wow.

@LLMJunky oops forget to mention made this with github-unwrapped but its limited to 2025 only.

@LLMJunky You're killin it man I need to lock in

I may not have the most impressive Github portfolio, but I'm quite proud of the work I've done over the last year. I can honestly say I've shipped some really cool software. Post your activity chart, I wanna see how I stack up against all of you goats.

@alex_whedon @vxnuaj This paper is never gonna get released lol.

@vxnuaj This week!

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

@ondrejgmux @shaun_on_x Looks good! gonna give it a try

@shaun_on_x @_mahybe you can use gmux.sh and create your down shortcuts x.com/ondrejgmux/sta…

Herdr is all you need

Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers. Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?

@steipete yeah if you want to end with pure slop maybe the better way is to sequentially plan and be questioned by the ai for every aspect of the plan if you want to end up with an actually production ready software lol

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.

@aitexzaz @hqmank Medium

@_mahybe @hqmank high or medium?

Google AI Pro subscribers now get Claude Opus 4.6 included in Antigravity. No extra cost. Just tested it. Works✅

@YashHustle_22 Sqlite and PostgreSQL most of the time SQLite does 99% of the job

Which database are you using most? - PostgreSQL - MySQL - MongoDB - SQLite - Supabase - Firebase

I can't stand how FAST macbooks get dirty I literally cleaned it up 3 hours ago and it's already super dirty

Is it just me or is gpt 5.5 super dumb today?

@LLMJunky in my account current state even bot replies make me happy lmao

got smacked by the algo so hard I dont even get bot replies anymore 😅