Andrew Kuncevich
297 posts
Andrew Kuncevich
@AndrewK404
Building complex production AI systems · ML/AI research, context engineering, agents, RL, evals · Poland
Katılım Mart 2026
113 Takip Edilen85 Takipçiler
Something I told 14 yo: There's a kind of politician who tells people "Your life is bad because <outgroup> stole what's rightfully yours. Vote for me and I'll get it back for you." They do it on both the left (Lenin) and right (Hitler), and they're invariably bad news.
English
Good point that you shouldn’t lock yourself into specific providers.
I think we’re only just entering the phase where OSS models start handling complex tasks.
They’ll keep getting smarter, even if they lag behind by 6–12 months (I doubt this gap will shrink much in the next few years). And eventually, they’ll be able to handle tasks at the level of today’s Claude Code.
I think I’ll finally build a server by the end of summer and start running agents locally XD
English
I'm 100% Codex pilled now
Been using Codex and Claude Code side by side hours a day for 2 months straight
No longer using them side by side. Codex has become incredible
What did it for me is the self testing. Every change it makes it self tests in it's own browser
I went from about 40% of my changes being buggy on first go to at most 3% maybe? So much more reliable and allows me to get in an awesome flow state
Listen, Claude can literally drop an update tomorrow that changes all of this, but for now I'm really blown away by Codex
Do yourself a favor and don't have loyalty to any company. Use every tool. Use whatever is the best at the moment. Switch whenever they're no longer the best. No point in tribalism
But at the moment I'm REALLY enjoying my time with Codex
English
@SkylerMiao7 Cool visual.
I think Sparse attention will be one of the top-10 architectural innovations this year.
English
Andrew Kuncevich retweetledi
the guy’s average morning, apparently:
- launched internet on planes
- tagged his mom’s photoshoot
- drank coffee
- talked about why colonizing other planets matters
(and he’ll probably ship another 2–3 releases during the day)
Funny how it feels like he’s personally doing all of this. If I were only reading the SpaceX account, I’d probably perceive it more as some abstract company.
Makes you think about how much a personal brand shapes perception…
English
Mostly true. What matters is securing the long-term future of consciousness, both on Earth and other heavenly bodies.
We cannot just focus on Earth, because there are irreducible external (eg massive meteor) and internal (eg global nuclear war) cataclysmic risks.
The Moon is faster to make self-growing, but is more susceptible to problems on Earth. Mars will take longer to make self-growing, because it is so hard to reach, but is more secure from Earth disasters for that same reason.
Both the Moon and Mars should have self-growing civilizations. Making this happen is the prime directive of SpaceX.
Jaynit@jaynitx
Former SpaceX astronaut Garrett Reisman reveals the single prism Elon Musk runs every major decision through "He measures pretty much every major decision by whether or not it brings the day when we have a self-sustainable colony on Mars sooner or later" "That's the prism by which he makes every single decision he makes" "He's got an idea and he'll keep pushing, and he gives us aggressive timelines that we have to work to" "We work really hard to try to meet them. It's hard when you're doing stuff that's this complicated to predict exactly how long it's going to take" "We end up falling a little bit behind, but we do our best"
English
Thanks for idea :)
Prompt for a createing same cool images:
Prompt:
```
Create a clean whiteboard / glass-board diagram in the style of a hand-drawn technical architecture sketch.
STYLE:
- Looks like a photo of a real glass whiteboard mounted on a wall.
- Slightly off-white / pale mint background from the glass surface.
- Hand-drawn black marker typography, casual but readable.
- Imperfect lines, wobbly arrows, rough boxes, sketchy icons.
- Mostly black ink with 1–2 accent colors: teal/blue arrows and small orange loop annotations.
- Minimal shading, no gradients, no glossy digital UI.
- Layout should feel like a startup/AI research architecture diagram drawn by an engineer.
- Use simple icons: folders, clouds, documents, browser windows, terminal, magnifying glass, checkmark, database cylinder, graph nodes.
- Keep generous whitespace and strong visual hierarchy.
- The image should look like an explanatory diagram, not a poster or infographic.
COMPOSITION:
- Large handwritten title centered at the top.
- Smaller subtitle underneath.
- Main flow goes left-to-right through 4–6 stages.
- Each stage has:
- a bold stage label
- one simple hand-drawn icon or box
- 1–3 tiny handwritten notes
- Use curved arrows between stages.
- Add one feedback loop arrow in orange labeled something like “self-improving loop”.
- Add a small “future direction” box in the bottom-right.
- Add a small “tools” row near the bottom with 2–4 icons.
- The whole diagram should feel dense but still readable.
CONTENT TRANSFORMATION:
Turn the supplied post into a visual system diagram.
Do not copy the post verbatim.
Extract the core mechanism, workflow, and surprising insight.
Represent the idea as stages, components, arrows, and feedback loops.
Prefer concise labels over long sentences.
TEXT RULES:
- Use short labels only.
- No paragraphs.
- Maximum 6–8 words per label.
- Make text look handwritten.
- Avoid tiny unreadable text.
- Use English.
MOOD:
Technical, clever, slightly viral, “AI engineer explaining a cool workflow on a whiteboard”.
OUTPUT:
A single 16:9 landscape image.
High resolution.
No photorealistic people.
No 3D render.
No corporate vector illustration.
No dark background.
```
English
🚨 @karpathy literally ditched traditional RAG for an autonomous Obsidian file system.
Instead of writing code, he dumps raw AI research into a local folder and lets an LLM convert it into an interconnected markdown wiki.
He rarely edits the text manually.
By relying purely on dynamically updated index files, the system navigates the exact context it needs natively without relying on flawed vector embeddings.
Because the LLM fully understands the file structure, it executes advanced autonomous workflows:
→ Operates a custom vibe-coded local search engine
→ Renders complex charts and formatted markdown slides
→ Continuously compounds a 400,000-word knowledge base
The most fascinating mechanic is the self-healing loop.
He triggers background health checks where the LLM natively spots structural gaps, scrapes the internet for missing data, and cleans the articles perfectly.
This feels the absolute blueprint for managing complex technical data 🔥
btw, he also plans to fine-tune a local model directly on the wiki so the research is baked into the neural weights rather than relying on limited context windows 👀
English
Damn, this is actually a very good argument.
It’s easy to forget that there are still a lot of AI labs in the world that haven’t really broken out yet.
Anthropic already showed how a single lab can become a market leader with massive revenue — which matters even more for investors.
The market is so large that a company capturing just 3% of it in 10 years could be worth $1T.
English
What you say is true, but nonetheless our AI will be great. Whether it is the best remains to be seen, but I will never give up. Never.
Space(XAI) is only 3 years old. That’s half the age of Anthropic and quarter the age of OpenAI.
Let’s see where things stand 3 years from now.
English
Elon doesn't need to run the best AI model. He needs to control the best hardware. Nvidia is the most valuable company on Earth and their models aren't the most popular. The hyperscaler play works.
English
They added cost to usage.
Probably worth mentally preparing for the fact that in a year or two, everything will go through API — obviously cheaper than now, but still several times more expensive than a subscription.
And we’re the ones giving them the data, heh
English
Out with the old, in with the bold ✨
We gave the Google Workspace icons a sleeker look to meet this new era of helpfulness. Check them out! goo.gle/4diuCsz
English