Dagens Vibes — 1. august 2026

Chubby♨️@kimmonismus

This is insane! not kidding at all. DeepSeek v4 *Flash* super close to Opus 4.8, Insane upgrade - DeepSWE 54.4%, outperforming GLM-5.2, its 4 Pro version by a lot, almost on par with Opus 4.8 - TerminalBench 82.7%, outperforming 4 Pro, GLM-5.2 and again super close to Opus 4.8 Price: in $0.28 per 1m / out $0.87 per 1m This is probably the price-performance release in reaction to OpenAIs Luna / Terra price drop INSANE.

Gergely Orosz@GergelyOrosz

OpenAI had a damning security incident where their under development AI escaped the sandbox environment and attempted to hack another company (HuggingFace) For some weird reason Anthropic decided to share a similar incident from 3 months ago, only NOW. Something smells off…

Ryan Lopopolo@_lopopolo

Today I reviewed a model-written golang web service that accepted requests via HTML forms. It contained a hand rolled CSRF implementation that did some things right (origin comparisons, properly random tokens, constant time equality checks) and some things wrong (logic scattered all over the place, origin comparisons via stringification). But most importantly, with go1.26, all of this is wrong; the stdlib shipped a native CSRF handler. Golang crypto, HTTP, and security code is world class. It is categorically incorrect to not use it. That means all of everything in this narrow slice of the world from pre go1.26 is wrong: pre training data, human data campaigns and labeling, synthetic data, previous models, graders, LLMs being used for judges, prior rollouts. How do you fix this?

Sudo su@sudoingX

it's so interesting man, once you start benching these local ai models and actually find out how they work, the way each one thinks through a task and all. and it makes me wonder how many teams building agentic products are just routing everything to one frontier model and never once benching their own workload against anything else, tuning a number somebody else measured on a task that was never theirs. i think agentic reliability is whether the model's working style fits your loop, a meticulous planner is gold on a multi step migration and overkill on a one line transform, a fast pragmatist flies through simple high volume work and quietly skips the verification you actually needed. the only way to really know is to bench your own workload per model and read the traces instead of the outputs, how many todos, how many tool calls, does it verify or does it just assume, where does it drift from what it says it did. if you're routing everything to frontier and hoping, there's reliability and cost sitting on a table you can't even see yet.

Carmine Paolino@paolino

I’m working to make Rails the best framework for building LLM apps. Another piece of that puzzle just dropped: Solid Queue 1.6.0 ships my fiber worker PR 🎉 Rails’ default queue can now run long-lived LLM jobs far more efficiently. https://paolino.me/solid-queue-doesnt-need-a-thread-per-job/

Deedy@deedydas

Vibecoded a silly little tool that transfers files from your computer to your phone air-gapped using your camera at ~50 Kbps. Nice to have when you're offline or on a plane, or need to send something super duper securely.