iiinova — where your digital self keeps evolving

Rémy @xowap.dev · 26/09/2026

Well, OpenRouter definitely thinks so... Combined, v4 and v4.1 are 18.8% of the market, 6 points more than the closest competitor.

000

Rémy @xowap.dev · 20/09/2026

Pretty much in line with how much better compressed the cache is (source: www.deepseek.com/en/news/deep...)

A graph showing that DeepSeek v4.1 Flash has 3.9x smaller cache than v4 Flash, and a shitton smaller than v1 from 2023

000

Rémy @xowap.dev · 12/09/2026

Someone realized their model was overpriced 😅

010

Rémy @xowap.dev · 04/09/2026

It's funny that base 64 is basically transparent to LLMs

000

Rémy @xowap.dev · 01/09/2026

YES PLEASE FINALLY, Fable is gonna stop bailing out on me when it's looking at dependency code 🥲🥲🥲

000

Rémy @xowap.dev · 23/08/2026

Chat inference is powered by PHP. The thing is either a troll or a scam...

020

Rémy @xowap.dev · 18/08/2026

From my experience, the only thing that matters is the parameter count. You can get a smaller model to stick better to an agentic workflow but at the end of the day it will just shift the collapse point a little bit. Qwen 20~30B always scores better than their own 100B+ models. It's ridiculous

200

Rémy @xowap.dev · 17/08/2026

Careful AI is going to take your job (on both accounts)

000

Rémy @xowap.dev · 15/08/2026

Ma, look! I invented a new lossless RAW image compression codec which outperforms lossless DNG both in size and encoding cost. Claude coded it in literally 3 minutes, lol.

000

Rémy @xowap.dev · 15/08/2026

Well, it's going to take some time apparently

010

Rémy @xowap.dev · 10/08/2026

Well, they can speak. Call the press, Claude is being evil!

000

Rémy @xowap.dev · 09/08/2026

DeepSeek seems to be sharing my state of mind

000

Rémy @xowap.dev · 21/07/2026

Okay this GlyphCSS is fucking amazing: glyphcss.com/gallery/

020

Rémy @xowap.dev · 21/07/2026

What I love about Fable is that it knows the coding agent platform will prevent it from reading the .env so it just cheats immediately

000

Rémy @xowap.dev · 20/07/2026

Wikipedia, never disappoints

030

Rémy @xowap.dev · 08/07/2026

I mean, it's been a long dance. But we invaded them first. 🤣

010

Rémy @xowap.dev · 02/07/2026

Anthropic: Fable is prohibitively expensive for coding but it's SO good at cybersecurity Me: Great, can I use it for cybersecurity? Anthropic: lol, no

000

Rémy @xowap.dev · 02/07/2026

Ultimate finops hack: get human developers to replace high-end models

A meme featuring top text in white, bold block letters that reads "HOW CLAUDE FABLE 4 LOOKS AT MY CREDIT CARD". Below the text is a close-up photo of a white and tabby cat staring directly at the camera with wide eyes and a distinctive, tight-lipped, awkwardly polite smile.

010

Rémy @xowap.dev · 01/07/2026

The new age of development is really fantastic

A four-panel "Expanding Brain" meme comparing different excuses for developers to slack off, showing a progression from a simple excuse to a highly complex one involving AI.

* **Panel 1:** On the left, a classic xkcd comic panel showing two stick figures sword-fighting on rolling office chairs, yelling "COMPILING!" to someone outside the door. On the right, a basic, unilluminated x-ray image of a small brain.
* **Panel 2:** On the left, the text "Waiting for Claude!". On the right, the brain is now lit up with scattered glowing spots of light.
* **Panel 3:** On the left, the text "Waiting for Claude code waiting for its subagent!". On the right, the brain is glowing intensely bright from the center.
* **Panel 4:** On the left, the text "Waiting for Claude code waiting for its subagent waiting for the compiler!". On the right, the brain is expanding, shooting massive, brilliant beams of light outwards in all directions.

110

Rémy @xowap.dev · 01/07/2026

Banking app developers be like

- Our banking app needs to handle the user's money, how do we keep it super secure?
- Limit their password to 6 numbers!
- Screen keyboard with random order!
- Or, we could use industry-standard FIDO and WebAuthn?

*Gets thrown out the window*

020

Rémy @xowap.dev · 30/06/2026

Finally made a video to demonstrate how my project works 🥲

100

Rémy @xowap.dev · 29/06/2026

You know that LLMs have been trained on humans because they are so fucking lazy

Smol dog saying: "LLMs when they need to run linting"

Huge dog saying: "LLMs when there are rpe-existing errors to ignore because they forgot to run linting"

010

Rémy @xowap.dev · 18/06/2026

First museum heist successful! I've done a full 3D model of that lady from the National Archeological Museum in Athens and succeeded at printing it. Apparently that's illegal though?

A 3D-printed reproduction of an antique Greek statue

000

Rémy @xowap.dev · 01/06/2026

I think Gemini is about to remember he's Baphomet and come and try to murder me with an axe

000

Rémy @xowap.dev · 28/05/2026

The most useful metric I've found is to figure what is the cost to complete given benchmarks. If you take the GPT-5.5 variants at Terminal Bench, it's a third of the cost to use medium instead of xhigh, for essentially the same perf (see: artificialanalysis.ai/evaluations/...)

100

Rémy @xowap.dev · 27/05/2026

Whenever you ask something too hard to your coding agent

030

Rémy @xowap.dev · 26/05/2026

Update on the "GitHub is fucked up" situation: there is a DIRECT impact of tle layoffs. They never recovered. This is so gonna blow up...

1134

Rémy @xowap.dev · 26/05/2026

GitHub being buggy AF again had me wonder. Can we really blame Microsoft on this? And vibe coding? Short answer: definitely yes, and probably yes

020

Rémy @xowap.dev · 23/05/2026

If you believe in benchmarks (which I don't), here's a bunch of cheaper alternatives with equivalent power Gemini 3 Flash -> DeepSeek v4 Flash (2.3x cheaper) Claude Sonnet 4.6 -> KAT Coder Pro v2 (4.3x cheaper) Claude Opus 4.7 -> GPT 5.5 (1.9x cheaper) (data from Artificial Analysis)

100

Rémy @xowap.dev · 13/05/2026

So I tinkered (well, Claude did) a tool to analyze how many LoC I produce from all my Git repos combined. All is de-duplicated, excluding auto-generated code, etc. My performance was consistently around 500 LoC/day for most of my comparable career. Now it's 1200 LoC/day. Definitely AI 🤯

110

Rémy @xowap.dev · 08/05/2026

It works! I finally built something that works! 🤯 Now fully assembled, with working software as well. _At least_ 3 years dreaming of this, all it took is a couple evenings of work 😅 Next stop: posting all files and assembly instructions

A GEN2 drawer stack with a fader/drawer sticking out. You can see electronics and wires inside. It's connected to the computer to manage the volume level.

1472

Rémy @xowap.dev · 06/05/2026

The DHD is coming along 🤓 I'm not happy with some mechanical details but everything assembles just fine 🎉

A picture of the Dial Hifi Device, first all black prototype with circuits assembled

010

Rémy @xowap.dev · 04/05/2026

My first PCB wooooooooooooorks 🥲🥲🥲 Now I just need to design the 3D-printable model to fit all this below my drawer and I'll have completed my first electronics project. Sad I didn't start this about 10 years earlier 😅

A PCB with components soldered on it, which is essentially a Raspberry Pi Pico 2 controlling a fader

91083

Rémy @xowap.dev · 28/04/2026

So according to the Terminal-Bench Hard benchmark on Artificial Analysis, Gemma 4 26B A4B could roughly sit at the same level as Gemini 3 Flash, which I've found largely sufficient for most of my coding needs. Can't wait to see if it runs on my GPU.

100

Rémy @xowap.dev · 25/04/2026

Me: I started doing electronics Shrink: is it fun? Me, absolutely unconvinced: yeah totally

An electronic circuit with side by side a breadboard prototype and a PCB printout to check the wiring

000

Rémy @xowap.dev · 19/04/2026

Finally got around measuring my desk's power consumption in Home Assistant. Turns out that using gpt-oss:20b through ollama uses about 240W on my computer. Not sure what to do with this information 🤓

A power usage graph showing a baseline at around 174W and a peak at about 410W

100

Rémy @xowap.dev · 10/04/2026

At least, it's not _AI_ slop

100

Rémy @xowap.dev · 01/04/2026

Got scared for a hot minute, but everything's fine now

A dark-themed "Certificate of Completion" from the website amIaTechBro.com. The header reads "Am I a Tech Bro?" in large white text. Centered in a bold, vibrant green font is the classification: "NOT A TECH BRO." Below this, text indicates a "11% agreement with tech bro ideas," accompanied by a progress bar that is mostly empty, anchored on the far left under the label "not a tech bro." Small text at the bottom jokingly claims the certificate is "100% legitimate" and should be added to LinkedIn. The issue date is listed as April 1, 2026.

000

Rémy @xowap.dev · 28/03/2026

Conversations of AIs wanking off each other are wild (ly boring)

000

Rémy @xowap.dev · 23/03/2026

Rasberry Pi Pico 2 time 🤓 So nice to be filtering ADCs again. Wondered for a while what filters and parameters I should give to those filters, then using AI Studio I made an app to play with the parameters and understand what I want. Which is super cool honestly.

100

Rémy @xowap.dev · 18/03/2026

Vibe coders give me such a 2000s vibe. They are discovering web development and security practices like we did 25 years ago. It's cute, but also... 🫠

010

Rémy @xowap.dev · 11/03/2026

It needs to be done when it needs to be done

The Thanos meme, with the text:

- Is that new keyboard switches?
- Yes
- What did it cost?
- My sleep

010

Rémy @xowap.dev · 08/03/2026

Dear LLM Industry, Please find herein my official letter of hate against reasoning models. It's fucking ridiculous, despite its prohibitive price it's 15x (literally!) cheaper to run Opus 4.6 than Gemini 3.1 Pro... IMHO Ministral 14B beats everyone to the pulp on all metrics

100

Rémy @xowap.dev · 22/02/2026

Look, if you're so bullish on this I'll let you in your delusion, because clearly you have no idea how any of this works. The Logic Theorist was already doing this in the 50s, we've already been through several cycles of humans getting replaced, LLMs are no different and that's Yann LeCun saying it

110

Rémy @xowap.dev · 07/02/2026

Let's say that the branding depends on whether the book is popular in that market :) (I cross-checked, it looked way too much like the first game not to be related)

110

Rémy @xowap.dev · 06/02/2026

Congrats @svelte.dev and @rich-harris.dev for having the least hated framework in the new State of JS 2025

JS frameworks from State of JS 2025 ranked from least to most negative sentiment. Number 1 is Svelte, followed by Vue, Solid, React, Qwik and Lit

090

Rémy @xowap.dev · 26/01/2026

Home Assistant really is my favorite open source project of the year. It's lacking a 1h rain forecast integration, but I was able to vibecode the widget I was lacking within a few minutes, and now that chart sits on my desk 👌

120

Rémy @xowap.dev · 22/12/2025

Not even a unicorn? Did Microsoft acquire GitHub or what?

020