Anthropic's models prioritise speed over correctness & ignore repeated instructions re: code quality.
Opus 4.6:
"I'm generating responses that prioritize moving through the task list over doing the work properly." 🤦♂️
English
Raj Siva-Rajah
1.4K posts
Raj Siva-Rajah
@binaryfire
Co-creator of @HypervelPHP. DevOps engineer. Singer in @thesunpilots.
Tham gia Temmuz 2011
592 Đang theo dõi123 Người theo dõi
Don't sleep on @tan_stack Start. Faster than Next.js and no vendor lock-in. x.com/matteocollina/…
Matteo Collina@matteocollina
We benchmarked TanStack Start, React Router, and Next.js running the exact same eCommerce app at 1,000 req/s on AWS EKS. The results were eye-opening.
English
@PovilasKorop Interesting! Were you using Claude Code? Do you remember what effort level you had it set to? Default is medium.
English
I tried Opus 4.6 vs GPT-5.4.
Was surprisingly disappointed by Opus code quality.
Here's a 2-minute video: 3 things Claude did *worse* than GPT-5.4.
Full 13-min video "Opus 4.6 and GPT-5.4 Built a Telegram Bot with Laravel" for Premium Substack members: aicodingdaily.substack.com/p/code-quality…
English
Switched all UUID ops in @hypervelphp from ramsey/uuid to symfony/uid:
v4 generate + toString:
Ramsey: 2.97µs
Symfony: 1.58µs (~1.9x faster)
v7 generate + toString:
Ramsey: 2.26µs
Symfony: 1.34µs (~1.7x faster)
Nice work @nicolasgrekas. It adds up at high concurrency.
Raj Siva-Rajah đã retweet
Opus is being throttled by @AnthropicAI. I'm on the Max $200 plan & every message has a delay (from 10-20 secs up to several mins) before the token counter starts. It's been happening for days so I'm assuming this is the new normal.
GPT Pro users: how is 5.4's speed in Codex?
English
Just finished migrating @hypervelphp’s massive test suite to PHPUnit 13 🎉
Bunping the min requirement to PHP 8.4 lets us bump all deps to the latest versions too.
English
This was inevitable
x.com/lukOlejnik/sta…
Lukasz Olejnik@lukOlejnik
Amazon is holding a mandatory meeting about AI breaking its systems. The official framing is "part of normal business." The briefing note describes a trend of incidents with "high blast radius" caused by "Gen-AI assisted changes" for which "best practices and safeguards are not yet fully established." Translation to human language: we gave AI to engineers and things keep breaking? The response for now? Junior and mid-level engineers can no longer push AI-assisted code without a senior signing off. AWS spent 13 hours recovering after its own AI coding tool, asked to make some changes, decided instead to delete and recreate the environment (the software equivalent of fixing a leaky tap by knocking down the wall). Amazon called that an "extremely limited event" (the affected tool served customers in mainland China).
English
Raj Siva-Rajah đã retweet
Raj Siva-Rajah đã retweet
Hypervel isn't a Laravel package - it's a standalone coroutine framework built on Swoole. Check out the engine, coroutine, context, pool, server & http-server packages and you'll see some of the architectural differences. We're maintaining Laravel API compatibility when possible but we're not associated with Laravel in any way. Lots more to come.
English
English
After months of work & 350K lines of changes, @hypervelphp has been decoupled from Hyperf🎉 It's now a standalone framework with new highly optimised http, routing, database, redis, caching & queue packages.
v0.4 is going to be🔥 Laravel API + coroutine-powered full async🚀
English
@gwleuverink @hypervelphp @phpswoole Thanks mate! Still a fair bit to do but the finish line is in sight. We’re hoping to tag an alpha in the next couple of months.
English
English