Sign in

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Vlad Iliescu

@vladiliescu.net
118 followers 98 following 214 posts

LLMs/MLLs, software, dad jokes.

PostsRepliesMedia
Vlad Iliescu @vladiliescu.net · 24/09/2026
cool cool cool cool
001 1
Vlad Iliescu @vladiliescu.net · 06/08/2026
The quality of the ChatGPT app is too damn high. I just got this on my Mac :/
100 1
Vlad Iliescu @vladiliescu.net · 30/06/2026
and THAT's why i need to install a new browser
0140 14
Vlad Iliescu @vladiliescu.net · 30/05/2026
Some quick vibe checks on opus 4.8 are not looking good, not good at all.
000 0
Vlad Iliescu @vladiliescu.net · 04/04/2026
Sigh.
110 1
Vlad Iliescu @vladiliescu.net · 09/01/2026
Know that feeling when power/internet goes down and everyone in the office just throws their hands in the air and stops working for the day?
010 1
Vlad Iliescu @vladiliescu.net · 09/12/2025
Mistral Vibe CLI is out, and *sigh, man we've got to stop normalizing piping random scripts off the internet directly to bash, it's just not right, is it too much to ask for this to be packaged as almost anything else than a bash script? Surely it can be ✨coded✨ in like a couple of sessions.
010 1
Vlad Iliescu @vladiliescu.net · 17/10/2025
Zed is cool and all, but I'm getting real tired of this shit. Wasn't even using the thing, just made the mistake of opening up two instances and leaving them in the background for a couple hours. Now I'm all out of battery 🤦‍♂️
010 1
Vlad Iliescu @vladiliescu.net · 08/08/2025
And they said money can't buy good taste. What were they thinking? 🤷🏻‍♂️ I mean, just look at this big beautiful lack of contrast, that unstyled, uncentered, jarring scrollbar, the amorphous blog of color reducing the contrast even further. Feeling #blessed to be paying for this, really.
100 1
Vlad Iliescu @vladiliescu.net · 07/08/2025
Just found this beauty on LI, and oh wow, such a bad take imo. The age of CTOs (by which I mean product-minded devs) is only beginning, what's ending is the age of "can't do won't do anything but code" code-monkeys. Full-stack is going to have a new meaning real soon 😉
000 0
Vlad Iliescu @vladiliescu.net · 30/07/2025
The full agent setup is equally simple. Zipcoil handles all the tool-calling iterations for you.
100 1
Vlad Iliescu @vladiliescu.net · 30/07/2025
Zipcoil makes OpenAI tools as simple as adding a decorator. That's it. No JSON schemas. No conversation management. Just Python.
100 1
Vlad Iliescu @vladiliescu.net · 30/07/2025
I've created a new "agentic" library for Python, whose purpose in life is simplifying the OpenAI tool-calling loop. It's called Zipcoil, and is available on GitHub github.com/vladiliescu/...
100 1
Vlad Iliescu @vladiliescu.net · 20/06/2025
My interview with Anca Trifan for #EventsDemystifiedPodcast is online 🚀.
100 1
Vlad Iliescu @vladiliescu.net · 19/06/2025
If only they had used a larger, non-quantized model 🤖😏
000 0
Vlad Iliescu @vladiliescu.net · 10/06/2025
🧐
130 3
Vlad Iliescu @vladiliescu.net · 26/05/2025
I get the feeling that 39$ per month is gonna feel real cheap real soon. Something to think about: each agent call consists as a premium request, and each model is assigned its very own "premium request multiplier".
100 1
Vlad Iliescu @vladiliescu.net · 23/05/2025
Pro-tip: You can find code issues & bugs & whatnot by just searching for "backend error handling exceptions bugs crashes" in your codebase.
000 0
Vlad Iliescu @vladiliescu.net · 21/05/2025
Going through the docs for several frameworks & apps & libraries, and I'm at the point at which if I see the word "seamless" one more time...
000 0
Vlad Iliescu @vladiliescu.net · 13/05/2025
Is...this how Substack **emails** are being delivered nowadays?
000 0
Vlad Iliescu @vladiliescu.net · 24/04/2025
Claude Code costs, on average, 6$ per dev per day, so 120$ per month. Why shouldn't I use GitHub Copilot Agent mode with Claude 3.7 Sonnet for $19, no wait it's actually $10 a month and pocket the change, maybe buy a nice watch in a year or so? It just makes no sense to use the API directly.
Manage costs effectively
Claude Code consumes tokens for each interaction. The average cost is $6 per developer per day, with daily costs remaining below $12 for 90% of users.
100 1
Vlad Iliescu @vladiliescu.net · 11/04/2025
We interrupt our scheduled broadcast to let you know Robo-🦾-Cop is coming to Mac! And it looks like the preorder price is ~66% off the base Steam price! Just thought you'd like to know, now it's back to talking about AI and stuff.
000 0
Vlad Iliescu @vladiliescu.net · 06/04/2025
So I was going through Hugging Face's Agents course when I found this bit of code, of which I'm 99,7% certain it's AI generated. What gives it away?
000 0
Vlad Iliescu @vladiliescu.net · 23/02/2025
Not my chat, but funny enough to share. Here are the screenshots, in case this gets taken down.
010 1
Vlad Iliescu @vladiliescu.net · 13/02/2025
"Not your weights, not your model" -> "Not your infra, not your model"
010 1
Vlad Iliescu @vladiliescu.net · 11/02/2025
Mind you, we're going to have to go to "thinking gyms" in a few years, to maintain what little thinking power we'll have left. www.404media.co/microsoft-st...
010 1
Vlad Iliescu @vladiliescu.net · 07/02/2025
What has steadily become my standard flow when creating a new account anywhere. Not sure if many companies are ready for a future in which people *actually* go through their ToS.
010 1
Vlad Iliescu @vladiliescu.net · 05/02/2025
Unfortunately, I'm not sure it matters anymore.
000 0
Vlad Iliescu @vladiliescu.net · 29/01/2025
Today I've been working on some PowerPoint, and in totally unrelated news, here's how you can disable Designer suggestions for good.
000 0
Vlad Iliescu @vladiliescu.net · 20/01/2025
> To me though, dependency inversion/injection is all about component lifetime management. I don’t want to care how to instantiate nor how to dispose a dependency. Still the most fun thing I've written recently. vladiliescu.net/better-depen...
To me though, dependency inversion/injection is all about component lifetime management. I don’t want to care how to instantiate nor how to dispose a dependency. I just want to declare it and then jump straight to using it. And the harder it is for me to use it, i.e. by instantiating it and its “rich” dependency tree, disposing each one when appropriate, etc, the more likely that I won’t even bother at all. Simple things should be simple.
010 1
Vlad Iliescu @vladiliescu.net · 19/01/2025
@hf.co's smolagents do have a sense of humor 😅
Q: "How many AI models do you need to change a lightbulb?"
A (AI): "Out - Final answer: Just one, but it will take a million iterations to make it work!"
000 0
Vlad Iliescu @vladiliescu.net · 08/12/2024
2. per IP caching configuration (with aiocache, yay!) This was rather straightforward but I still wanted to write it down -- client_host with a custom key_builder does wonders. Last, but not least..👇
100 1
Vlad Iliescu @vladiliescu.net · 05/12/2024
If you see this, post a knight
000 0
Vlad Iliescu @vladiliescu.net · 15/02/2024
A very "go big or go home" from Gemini Ultra. 1 million tokens 🤯. Curious how well retrieval accuracy & generation performance work outside Google's own test labs, but if they're even half-decent... just imagine what's coming our way in 6 months. blog.google/technology/a...
000 0
Vlad Iliescu @vladiliescu.net · 04/12/2023
should've seen this coming 🤷🏻‍♂️
000 0
Vlad Iliescu @vladiliescu.net · 18/11/2023
It just feels wrong to deploy it now...
000 0
Vlad Iliescu @vladiliescu.net · 09/11/2023
So let me get this straight, GPT-4-Turbo has 128k tokens context window, but you can only send it 10k tokens a minute?!? That's...quite something.
000 0
Vlad Iliescu @vladiliescu.net · 19/10/2023
Just found out you need to enable recovery questions such as these on Substack *before* they let you turn on 2FA🤦‍♂️ That's..definitely not a stupid thing to ask your users in A.D. 2023, no siree. Anyways, I remember fondly my first computer, or pBhSPwjknRCMB1VY45iSgIRMIAWqxW5g as we used to call it.
110 1
Vlad Iliescu @vladiliescu.net · 13/10/2023
Surely there were better ways to differentiate these two 😔
000 0
Vlad Iliescu @vladiliescu.net · 20/07/2023
3) Generated code seems significantly worse with the June 2023 versions of both models. 4) Visual reasoning has marginally improved for both models. Comparison chart looks like this btw. 4/
100 1
Vlad Iliescu @vladiliescu.net · 06/07/2023
Microsoft MVPs this time of year
2 spidermen pointing at each other.
000 0
Vlad Iliescu @vladiliescu.net · 16/06/2023
Next time someone recommends I trust iCloud for syncing files, I'll show them this picture. 39 KB worth of files, uploading for more than 3 hours, 77 KB uploading for less than an hour 🤦‍♂️. No way to speed things up or find out the cause. Lame.
000 0
Vlad Iliescu @vladiliescu.net · 07/06/2023
You vs the car she tells you not to worry about
You: A Tesla interior with a big screen and nothing else because who needs to look at the road while driving amiriteThe car: An Audi A5 interior with a big screen and lots of buttons because buttons are love.
000 0
Vlad Iliescu @vladiliescu.net · 30/05/2023
Could be worse 🤷🏻‍♂️
000 0
Vlad Iliescu @vladiliescu.net · 25/05/2023
I’d settle for Eyegore 😬
010 1