Nikolay Konovalov

Independent web developer building Coherence (pun intended) — an open-source intent graph and verification layer for software teams.

[Nikolay Konovalov @br11k_dev]

https://t.co/mrFl64nN8U

[Nikolay Konovalov @br11k_dev]

@nealagarwal here's public key btw https://t.co/9p4iZnGJXd

[Nikolay Konovalov @br11k_dev]

Damn bro got ahead of my idea on a road to fully OpenSource myself🥰
I only got this so far: https://t.co/GWtlDHyuoC

Pulling out data from HealthKit into VictoriaMetrics DB, and then exposing them over APIs.

My next steps are medications, weight, food.
And also I have this: https://t.co/r0CooRjhNw

I wanna become opensource human ☺️
As much as possible.
The idea is that if I achieve something good then people can trace back my steps, train a model on what I did and see some insights.

If I become criminal then this also serves as data point where things went wrong.

If you clone me, replicate, try to do synthetic persona from my thoughts, images, code — you would not be able to fully replicate my brain cells, my state of consciousness. So I will be able to prove that I am me. GPG key is public on https://t.co/KlcLdDVtwW, I’ll publish it to more places.

The very least I can do to prove authenticity is to go to a public camera and show my accounts on my phone and my face, and post it with GPG key signature.

[Nikolay Konovalov @br11k_dev]

@jcruz821 @notesbyfloat @nealagarwal

[Nikolay Konovalov @br11k_dev]

@Jaidcel Power of open research ☺️🔥✍️

[Nikolay Konovalov @br11k_dev]

@_sebir_ @uchuujinniaitai With углеродом (carbs)☺️

[Nikolay Konovalov @br11k_dev]

So I guess that’s why not many US companies hire people in EU via Deel/ Rippling. I was hoping cross border hiring have been solved long time ago🤔

TheirStack have a list of companies that use Deel/Rippling and so that’s how I find those that can hire me.

But the real problem is legal and taxes. People don’t even want to engage with the BS over hiring someone that could be a better candidate.
I transitioned full time into remote work because there are companies out there that are better match for me, and it also means I have to compete not with locals but entire globe 🌎

So this exposure really pushed me forward so much. I am not the best but I am pretty good 😌

We should lift these restrictions but first we need to figure out how to evaluate problem solving correctly. AI resumes are getting out of hand and people stuck with AI slop from both sides.

[Nikolay Konovalov @br11k_dev]

@DGP205244 @NotKeisev

[Nikolay Konovalov @br11k_dev]

@neogoose_btw my favorites are
https://t.co/dKVCI6dotC
https://t.co/gYAjy0MEwh
https://t.co/oL0kLrjwiA🥰

[Nikolay Konovalov @br11k_dev]

@neogoose_btw slightly offensive domain names namecheap suggested based on keywords extracted from the original domain☺️

[Nikolay Konovalov @br11k_dev]

@neogoose_btw 🤔🤔🤔
🤔🤔🤔
🤔🤔🤔

[Nikolay Konovalov @br11k_dev]

@moonfell_rpg I am presenting my idea:

> we should dismantle capitalism and build new order without corruption, anger, anxiety, competition for resources, non-stop consumption, stupid bullshit, and taxes

Someone, please vibe-code this☺️

[Nikolay Konovalov @br11k_dev]

@valigo Life imitates art😌

[Nikolay Konovalov @br11k_dev]

@valigo https://t.co/XwlNevYv95

[Nikolay Konovalov @br11k_dev]

I'd say, converting image to text is sorta losing "resolution" a lot. E.g. you have Mona Lisa painting, and then LLM is forced to produce text, and you get "a painting of a girl ..."

If models are multimodal already, they shouldn't do this to replace the image tokens/embeddings. Unless I'm mistaken what's going on here with the subagent / other model.

If they use images as feedback, the image representation already contains enough information for model to produce correction for it.
This is basically more efficient context management here, I can expand a bit more if interested👀

[Nikolay Konovalov @br11k_dev]

@valigo Damn, who would have thought in 2007 that this would be 2026 meme☺️

[Nikolay Konovalov @br11k_dev]

Creativity is such a difficult metric to begin with. It's sort of ability to make things Coherent, bring two ideas together into single one in a meaningful way.

I'm thinking maybe you can design a structure for this creative process. In theory, you can combine infinite amount of things together but it would be like a soup in Stardew Valley (hopefully without Mayor's pants☺️).

But every asset we create can become some sort of a graph node with links to ideas, I think. And so then we might be able to measure complexity of the "creativity graph", and then second, how suitable those connections are, if they make sense or not🤔

[Nikolay Konovalov @br11k_dev]

I'd suggest to try making an eval.

This is how they measure model on somewhat objective metrics. It shouldn't make them deterministic but you can actually measure creativity. One way to do it might be amount of visible details that change the asset in a meaningful way.

Since you're evaluating it as a human, you can assign subjective scores too based on some criteria. Granted, the most challenging part would be figuring out what exactly criteria are important to you☺️

[Nikolay Konovalov @br11k_dev]

@CryptoCyberia @XMRScott I mean, this meme I had to make was sponsored entirely by this post ☺️

[Nikolay Konovalov @br11k_dev]

@XMRScott @CryptoCyberia This guy https://t.co/9WNP6iDDv8

[hraness @hraness]

my software factory uses 5-10B tokens a day

24x @DevinAI SWE-2 Max
6x Codex Astra Ultra
2x Claude Code Fable Max

running in @zeddotdev

[Nikolay Konovalov @br11k_dev]

@Jaidcel