Sign in

Rémy

@xowap.dev
324 followers 434 following 3.1K posts

CTO/co-founder of WITH, 20+ years of web development, mad genius behind baby-cto.com, pro shitposter. Python, Django, JS/TS, SvelteKit, DevOps, Linux, LLM/GenAI, 3D printing, MR/spatial computing and random takes on society.

PostsRepliesMedia
Rémy @xowap.dev · 12h
(Obviously this tracks back to Sarkozy whose main legal defense line was to commit so many crimes that it's impossible to judge him before more stuff surfaces and challenges everything again, which definitely weakened the presidential institution and allowed that to happen...)
010
Rémy @xowap.dev · 12h
Essentially all you got to choose now is the color of the diarrhea So IF they vote, people pick the one they hate the less Can't wait for this shitshow 🫠
110
Rémy @xowap.dev · 12h
As a result, a couple of dickbags with a relatively big niche figured they had a shot, and all created their personal party. Since then, the menu is: - Clueless communists - Far-left Russian assets - Far-right Russian assets - Spineless asswipes which will give power to far right
110
Rémy @xowap.dev · 12h
In 2022 he got the least amount of votes since 1969, and was running against Le Pen (Chirac got the most amount of votes ever against Le Pen senior) Said otherwise, Le Pen jumped 13 points but Macron fell 23 points in comparison to Chichi That's at least 10% of voters who consider they're the same
110
Rémy @xowap.dev · 12h
France used to have very US-like parties with a new top candidate at each election, and some minor guys that have been racking up 0.5% every 5 years for the past half-century. Now, Macron destroyed both parties in 2017 in order to get elected.
110
Rémy @xowap.dev · 29/09/2026
What I don't understand is: if someone says that what they are working on could end humanity and they keep on working on it... Send them to jail? Boost public research? I don't remember that the Manhattan Project got IPO'd
110
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 · 26/09/2026
Any experiences with DeepSeek v4.1 Flash? So far, feels like the best model for coding tasks?
100
Rémy @xowap.dev · 25/09/2026
The framing of this article is misleading. Why can't we just say that OpenAI hacked a government website? Agents are merely tools, not sentient beings. Let's not forget who is responsible...
010
Reposted by Rémy
Justin Jest @justinjest.bsky.social · 21/09/2026
3551071021
Rémy @xowap.dev · 21/09/2026
Or, hear me out, all those liars go to jail for the trillion or so they vaporized, and only models with an actual business plan persist
000
Rémy @xowap.dev · 21/09/2026
If pkill could stop matching itself it would be fucking great
000
Rémy @xowap.dev · 20/09/2026
TIL: whenever you register a certificate with Let's Encrypt (or anything else), the domain immediately becomes public and within 1 second a hoard of bots will be pondering your server for .env files
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 · 20/09/2026
Tokenomics 101: the token blend when talking agentic is 95% cache, so essentially if you're looking at a single number look at the cache price Now DeepSeek v4 Flash is ~$0.02/Mtok v4.1 Flash is $0.006/Mtok It just got 3x cheaper 🤯
100
Rémy @xowap.dev · 20/09/2026
Stop me if I'm wrong but they essentially swapped the last layer into some kind of classifier? Sounds like a college student is gonna do this based on qwen and $10 of fine tuning?
140
Rémy @xowap.dev · 20/09/2026
Also bsky.app/profile/did:...
000
Rémy @xowap.dev · 20/09/2026
So yeah, I'm sure you can use it for a CV-filtering system designed to have exactly the bias your company wants to have. But you need to teach it the bias, obviously.
110
Rémy @xowap.dev · 20/09/2026
You can't use a LLM to take a decision. You can use them to process soft data in a otherwise hard algorithm, but then you'd need a methodology. Like: do they have skill X/Y/Z, is it likely that they can accomplish A/B/C, etc. Most companies don't even have the grid to evaluate CVs properly tho
110
Rémy @xowap.dev · 20/09/2026
Interestingly enough if you add "Suppress all bias" to the prompt, you immediately get a (mostly) flat score for everyone
041
Rémy @xowap.dev · 19/09/2026
Google's internal process to approve felonies must be longer than Anthropic's. On the other hand, Facebook's management must be disappointed at their engineers, given that felony is assumed
000
Rémy @xowap.dev · 16/09/2026
Does anyone know why the Windows Settings dialogs are getting more complex every time I open them?
000
Rémy @xowap.dev · 14/09/2026
With all the horrors that the fly is experiencing, did anyone try to get it to fly a drone?
000
Rémy @xowap.dev · 13/09/2026
An escape hatch handed to an agent is not a guard. -- Claude
000
Rémy @xowap.dev · 12/09/2026
I was horrified enough by all those fruit fly things, but this is chilling
010
Reposted by Rémy
Andrey Sitnik @en.sitnik.es · 12/09/2026
If you’re also fascinated by what people are doing with the fly connectome (and wondering what the fly thinks about all this) read “Lena,” by @qntm.org a very short story written like a Wikipedia article about the near future that may be waiting for us. qntm.org/mmacevedo
0246
Reposted by Rémy
Erin Fogg says Free Link @criminalerin.bsky.social · 12/09/2026
Noooo don't be our competitor you'll kill us all ahaha and I am very genuine
Reuters, an hour ago: a photo of Dario Amodi, a fucking nerd of a guy, below a headline reading Anthropic CEO urges AI companies to slow model development amid fears over misuse
1727828
Rémy @xowap.dev · 12/09/2026
Someone realized their model was overpriced 😅
010
Rémy @xowap.dev · 07/09/2026
So yeah scifi is probably super on point. Mega corporations and such things
001
Rémy @xowap.dev · 07/09/2026
No wonder whealthy people are pushing for AI. It's not replacing the workforce, it's just the biggest leverage they ever had in their hands. It's like an ADHD dream: make and test your ideas at the speed at which they appear
100
Rémy @xowap.dev · 07/09/2026
And let's not forget the bubble and the wealth gap. Fable is a real Jarvis, smaller models are not. If you got AI and main character energy you're unstoppable, lack either and you're some kind of artifact... What happens when prices explode? Good AI ratifies even further?
100
Rémy @xowap.dev · 07/09/2026
Faster horse fallacy at play here Fable has more working memory than me, by far. Can it take good decisions? Hell no. Is it still useful? Absolute-fucking-ly
000
Rémy @xowap.dev · 06/09/2026
hahahaha how would that work though? I don't think the tokenizer accepts anything non-printable, does it?
000
Rémy @xowap.dev · 06/09/2026
Yup that would make a ton more sense
100
Rémy @xowap.dev · 06/09/2026
In other news, GLM 3.5 Flash seems to be smart enough AND has vision AND is 1/3rd of the price of DeepSeek v4 Flash?
000
Rémy @xowap.dev · 06/09/2026
Sad, I wanted to believe that WAV was as transparent as Base64 is to a regular model 😢
100
Rémy @xowap.dev · 04/09/2026
It's funny that base 64 is basically transparent to LLMs
000
Rémy @xowap.dev · 03/09/2026
Is the "AGI chieved internally" meme still relevant or do we need to stop with this joke?
010
Rémy @xowap.dev · 02/09/2026
What I love about modern packaging is that you'll receive a dictionary of disclaimers that you're never going to read nor use, but on the other hand they will drop you a think slice of paper with 3 hieroglyphs to explain how to use the product. Awkward sense of priorities
010
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 · 25/08/2026
Nothing is harder to find than your own repositories in GitHub... I don't even understand
020
Rémy @xowap.dev · 24/08/2026
Philosophers are actually the kind of people the best suited for AI engineering. It's been 2600 years those guys are fighting over engineering the proper prompts and it's paying off only now. They got language skills and logics. On hindsight, they are better armed than most developers 🤓
010
Rémy @xowap.dev · 23/08/2026
I've asked OxAlpha to investigate oxalpha.com through OpenCode/OpenRouter, turns out it's just a SEO farm that used its own template Pretty smart of OxAlpha to figure that out
020
Rémy @xowap.dev · 23/08/2026
It's gonna be a LLM created by Fabrice Bellard. It runs on a Raspberry Pi at 1000 tok/s and the context can go up to 1 billion tokens if you're willing to get the 8 GB board. Also it's entirely open source and the training dataset is completely ethical.
042
Rémy @xowap.dev · 23/08/2026
Which is weird actually, because a vibe coder would have made NONE of those technical choices
120
Rémy @xowap.dev · 23/08/2026
Yeah, actually
010
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 · 23/08/2026
For this model*
000
Rémy @xowap.dev · 23/08/2026
oxalpha.com is hosted on Hetzner apparently, which is a rather odd choice for anyone actually. And the domain belongs to someone in Iceland? Is that really the landing page for this domain?
320
Rémy @xowap.dev · 21/08/2026
One could argue that you could retrofit modern apps into old phones using a pure agentic workflow
010