mvaled

When you ask an LLM for Rust implementation of the #Django template engine...🫣 I know this is a hard project, but I didn't expect it was going to cheat and shamelessly announce it.

@gitlabstatus I think you are having issues... My merge trains are taking forever to start, I'm getting connection errors to registry.gitlab.com, etc.

@Grady_Booch @RichardDawkins @claudeai I'm truly interested, and would greatly appreciate if you can provide references for further reading.

Dear @RichardDawkins (and Steve), With regard to the metaphor declared by @claudeai of a "consciousness is a essentially a moving point traveling through time" you asked "could a being capable of perpetrating such a thought really be unconscious?" I agree: this is indeed a profound thought, worthy of consideration, and I celebrate the beings who formulated it. However, what I do not celebrate - and indeed I condemn in the strongest possibly way - are plagiarism machines such as Claudia (as you called it) who first fawningly stroke one's ego ("that is possibly the most precisely formulated question anyone has ever asked about the nature of my existence") and then confidently assert a concept as its own. For you see, the idea is akin to Broad's moving spotlight theory of time (1923). Bergon's ideas in Matter and Memory (1896) address the idea of la duree - lived time - as a flow of consciousness. Vedantic traditions speak of consciousness an unchanging reality that persists outside of time (see Sat-Chit-Anada). Edmund Husseri had similar ideas. We don't know on what corpus Claude has been trained; @AnthropicAI has not been open to this matter. But, given that the idea is quite old, not novel, and has been explored in a multitude of human artifacts, it is reasonable to be skeptical of this machine; it is good to celebrate the thoughts of these undeniably conscious humans who first explored the ideas; it is unconscionable for you and to Steve to promote the results of a plagiarism machine that wraps up stolen concepts in coherent, poetic language. You have both been deceived. And Richard, I expected much more from you.

Plot twist? I'm using LLMs to remove it. I know already many patterns that indicate dead code; I "taught" the LLM those patterns and carefully select its targets.

I'm doing major contributions to the future of my company. I'm removing a lot of unused code. A lot. LLMs hardly know the code is useless and they spread the rot.

im fully convinced that LLMs are not an actual net productivity boost (today) they remove the barrier to get started, but they create increasingly complex software which does not appear to be maintainable so far, in my situations, they appear to slow down long term velocity

@sjryanwriter @Grady_Booch @StarTrek In Cuban Spanish we get one new one with each reggaeton "lyrics"

@Grady_Booch @StarTrek To be fair, irl we haven't had any new swear words in centuries. We're still using a medieval acronym for naughty sex.

First major finding: It produced way more code than necessary, e.g it is manually parsing JSON instead of using the readily available deserializers. Luckily it can be instructed to remove the unneeded code and use the existing tools.

I have AI-induced cognitive overload. I have to validate the implementation of the fairly complex specification (simulation of a replication between replicas). The AI seemed to have produced a decent result; but it can't trusted; it's now on me. (1/2)

I need to learn to manage this new kind of scenario.

Without an AI coding tool I would have produced the code at a slower pace, keeping my cognitive load controlled.

Credit to github.com/manzaltu/claud…

#Emacs is so awesome... While I have claude-code-ide waiting for trivial approval (ediff), I use a normal claude code to explore the ediff windows and its bindings.

@leiserfg @OmarchyLinux Not there yet. I'm breaking with Omarchy as a project, but I want to keep many things from it. Maybe I just fork it and fix things to my liking.

@mvaled @OmarchyLinux Nixos time?

I'm leaving @OmarchyLinux -- it is becoming Basecamp's distribution instead of developer distribution. Pushing Fizzy, Hey, and all other Basecamp products into it. Thanks anyways for showing me something that was cool.

It looks Claude Code is so confident people will accept every changes that, when I reject the first change it proposes with foundational code; it doesn't reevaluate and keep doing pointless changes that won't work because I have rejected the first one.

What's the point on having Gemini in Google Docs if it cannot even apply format... Frustrating experience for @googledocs

@mitsuhiko Are we?

Just a few hours after the latest CVE fix they landed another protection in the react flight protocol. Complex, magic RPC systems at the core of untrusted interaction were known to be hard to secure since early Java times and every time, we are learning the same lesson.

@gitlab Please, don't force Gitlab Duo on my face... It takes too much space, for it to be open by default in every, single page load...

@Cloudflare -- several sites affected. I'm located in Berlin BTW

@Cloudflare or @gitlab need to do something