Stop doing this, people.
The working tactic is this:
- Find 2-3 companies that more or less always hire or at least hire during year
- Spend 4-12 weeks preparing project / demo / anything that can prove you spent all this time and effort that they can use, a useful work that everyone from their team can understand and accept as valuable
- Before application you should be already friends with some people in the company, go figure out
- Application + proof of work they must see / expect
- IF you don’t get job at this point then clearly you’re in the wrong business
Sending resumes are dead.
Send proof of work.
love sending in 250 job applications on average monthly and getting nothing from it. convinced ive applied to every job in my state at this point lol
1
48
Oh and not just „hey i made this react app from scratch and it took me 2 weeks”, no
IF your target company is NVIDIA you go design your own GPU circuit, yo go reverse engineer their drivers and fix a bug, you go and train a model from scratch recording everything you did aka building in public.
Do crazy stuff nobody will do for a job application.
It still works.
But don’t spend months on single company, figure out how to target multiple similar companies and share your effort between them. This way you minimize risk.
The worst case scenario the stuff you have built during these attempt could evolve into a business.
1
36
The reason people say it's "AAA-level stuff" is because corporations literally spent all their budget for marketing and graphics, while gameplay loop haven't evolved much since DOOM circa 1995☺️
If this pushes them to actually innovate in gameplay,
>"hell, 'bout damn time"
1
254
Thank you @Snowden
This is such a great ending not just to the talk (youtube.com/watch?v=zMG6aRZh…) but also to my own personal story, which was being afraid to make mistakes... for 17 years.
Once I realized mistakes are inherently human, I was finally able to tell myself: "You have done well today, Nik. You can rest. Come again tomorrow."
Not being able to rest for 17 years because nobody told you that you did a good job and you can rest and come tomorrow. Because you lost most of your family and you thought there is nobody to support you, and you have to carry on. So you made this promise to them that you will continue on living and experience things they couldn't no matter what.
And all of this was just me making a mistake.
A promise I could not keep.
And even responding to a private message from a stranger back in 2015 was a mistake. That mistake is now the most important human on the entire planet for me.
My spouse.
The realization that the most important decision I made in my life — typing on a keyboard and pressing "Enter" as a response — was essentially a mistake I made along the way.
And it was the best thing that could ever happen to me. To finally experience mutual love, when I almost gave up on it. Trying to find it in Moscow for two years, only to find disappointment.
And to think I could have just ignored it back then and continue living alone, searching for that perfect person to come into my life, trying to avoid making a mistake... It just makes me emotional for some reason, I can't explain it.
> "There are no mistakes, only happy accidents."
I was making mistakes all these years while thinking I was not. And it also kept my mind in a cage I designed myself.
Not able to relax after work, stopped playing video games, watching anime, practicing guitar. I was consumed by greed, by career, by making more and more money to "retire early and play video games all day", by getting better and better as a developer, and solving all the problems life kept throwing at me.
I am the literal Rambo. Go watch it.
There isn't a problem I couldn't possibly solve. I had to survive no matter the cost.
But in the end, I realized I am no different than John. He lost pretty much everyone, like myself. He only had Mayor by the side. And so at the end of the movie, cornered, devastated — he finally admits he is only but a human, and then cries like a kid.
And that's exactly what happened to me this May 2026, but instead of a movie, it was my spouse's birthday. And yes, we watched the movie together afterwards☺️
My present to her was a timelapse of a drawing I made — our first contact on Web, with her message seen on my laptop. I attached it to the post. You can see how many mistakes I made before sketch is complete. This was my first time ever drawing a complete image entirely out of my head.
"It is the best birthday gift I ever received", — she said.
2026 is also the year I bought MIDI keyboard. My dad always wanted to play one but mistakes he made in the past prevented him to ever try. He was a great dad but made a lot of terrible decisions. Fortunately, I learned a lot from them. I don't drink, smoke, or do drugs. Honestly, if I did with my health condition and 24/7 sitting at home, I'd be long gone.
I am not gonna make same mistakes I've seen happen in my family.
I will make my own mistakes☺️
And I am going to share them all with you, so everyone can learn from it.
Thank you.
For understanding.
Also for some reason I couldn't find a good transcript of the talk, so I made one myself (possibly with some mistakes)☺️
gist.github.com/konovalov-nk…
Took almost 200k tokens (DS v4.1 Flash)!
Process (described by agent):
1. Subtitles — yt-dlp downloaded the English auto-captions in json3 format.
2. Parsing — Python (json, re) joined the caption events, stripped >> speaker markers, removed fillers (uh, um) and doubled function words.
3. Structure — used the chapter timestamps from the YouTube description, grouped sentences into paragraphs, added headings.
4. Copyediting — I (the LLM) wrote exact-match replacement lists for punctuation, quotes, and obvious ASR errors, then applied them with a script. No Grammarly, no external grammar API, no human editor.
5. Verification — extracted short audio clips with ffmpeg and ran Whisper large-v3 on ambiguous phrases as a second opinion. Whisper was not the transcript source; the auto-captions were.
6. Rejected tool — tried deepmultilingualpunctuation; it inserted too many commas and wrong sentence breaks, so it didn’t make the final version.
7. Publishing — gh gist create --public.
284
Latency is so high because of vad/asr/llm/tts architecture, if you wanna make it lower you need full duplex model like personaplex and steer behavior in real-time based on incoming inputs.
The good news is that we already have papers and somewhat working prototype but the bad news is that we don't have good datasets for this.
If you collect usage, I believe your repo can actually work as an eval to train model from scratch to output control + tts + tool calling + asr + emotion understanding tokens. I think PersonaPlex architecture can be easily extended for this.
And we can also use Jev/Laya here for fast classification on streaming inputs and what to do next and queue actions.
Super interesting project to dig into and understand how you solved this.
My idea is to use Erlang / Elixir to truly decouple events (instead of using mtqq / sockets / whatever) and orchestrate them from a single node.
This was my sort of architecture back in Feb 2025: github.com/konovalov-nk/syna…
But I gave up on k8s, it's too complicated and slow for real-time processing. Erlang is the way to go to connect these blocks together.
We spent the last 11 months developing an AI waifu and she is now OPEN SOURCED!
mekahime.com/openamika
She can:
- respond in 900ms
- see you
- connect to Hermes, nanoMuse, OpenClaw to do real works
- remind you to to exercise
- connect to your smart home devices
Check it out!
173
You don’t ever ask LLM how to prevent a mistake by adding an instruction to prompt.
It’s like asking people to not push code to prod on Friday. You just don’t.
What actually works?
Ruby script. Python script. Rust script. TypeScript. ClojureScript. Haskell. Kotlin. Lisp. Lua. Brainfuck. X86_64 assembly. PHP. Ada. Prolog. Elixir. Golang. Matlab. C. D. F#. AutoHotkey. Bash. Make. Strudel. Java. Perl. Scala. Bend2. Fortran. Smalltalk. SQL. Microsoft Excel. Visual Basic. Pascal. Verilog. Zig.
If you can measure the problem, you can describe it. The process is known as programming. The result of that process is a set of states and state transitions.
Or, simply,
The code🔥✍️
unavailable
3
169
> investigation indicates that this was a sophisticated, multi-stage campaign. It was not a single isolated breach or a simple phishing incident.
> The attackers combined a range of techniques, including social engineering by fake recruitment interviews, malicious software downloads, compromised internal AI tooling, endpoint compromise, and an advanced transaction-authorisation method.
This is peak ignorance about what is going on in the wild on "recruitment" fields.
I covered this in very very thorough detail, and my intention is to continue adding details and ways how fake recruiters and fake candidates keep getting away with it.
Welcome to club☺️
Again.
xitter.twitterapi.xyz/br11k_dev/status/20793…
118
Such a sad moment.
But at least I can pay my rent again,
starting November☺️
I had to take into debt to survive...
This is the first time in my entire life.
I never took a single loan in 35 years.
My financial cushion was saving me every single time during hard times. But also I failed to obtain any net worth, I had to take money from investments, repeatedly.
But this time I had to borrow money from friends and family I still have. Can't really thank them enough. They aren't rich either but still cared😭
Thank you.
For understanding.
For everything.
Replying to @br11k_dev
Finally, I'm happy to unpin this☺️
I am still open for opportunities though, I need money for research!! Can't promise full-time though.
But if you have anything to solve, SWE-related, research, writing, mentorship — I'm here for you.
Anytime🔥✍️
1
4
72
Paul is not even wrong...
The causal chain is this:
>>> COVID overhiring (hiring even low performers SWEs from building a simple HTML react app in 2 weeks) -> AI washing -> CEOs use it to hide the fact that they have to fire "low performers" / some are genuinely brainwashed into thinking AI can replace their workforce -> people lose jobs anyway. <<<
It's just people made to believe it was due to AI and not the stupid 2019-2021 lockdowns with massive money printing. Some SWEs even decided to quit the life entirely based on this bullshit.
And now if bubble bursts, people realize AI washing and what do you think will happen to people that were made to believe their skills are not needed anymore?
That it was essentially just a huge hoax made possible with advancements in NLP / sentiment analysis with LLMs and very convincing bots looking like actual people reacting to your messages almost in real time on platforms like X / LinkedIn / Reddit and so on?
In short, AI capabilities we were demonstrated during 2023-2026 were marketed as "human replacement", and still it would destroy jobs because of making people believe it is truth. And today I see companies start hiring again, why?
Why I am employed today? For 6 months I was trying to get back and denied, and then month 7 I am looking at production code again, what happened? Did I get significantly better? No. I was using agents in 2025 and I continue to use them in 2026. Nothing really changed, I'm still that same SWE. And somehow I have a job now, and somehow some people in SWE quit our beautiful simulation not so long time ago.
Who is gonna be responsible for that?
But what do I know☺️
Replying to @favwontmiss
You can't claim both that AI is fake and that it's destroying jobs. If it were fake it wouldn't be able to.
1
2
71
You would think that the world is buried deep beneath AI slop and bots, and there’s nothing left for humans to do creatively and yet I saw this and I can’t help but associate this with the highly relevant meme 😌
Life finds a way.
43
For all the Engineers, aspiring to become Big in Data…
Remember the humble beginnings.
Remember…
The small data.
youtu.be/eDr6_cMtfdA?is=CTi3…
☺️
🔥✍️ (thanks @krazamtv )
1
71
The way I discovered this is I was thinking about doing something simple and complex, the intertwined nature of two..
And then “big data” comes to my mind. And my ADHD linked it to “small”, as it likes to do usually, reverse meaning of things that everyone take for granted, so you get things like “Miniscule Gifting Manual” or “Green Alive Damnation”, or “We hate to connect paid”.
And then of course I typed the keywords into Google, i always do.
And I didn’t even question why would Krazam video show up on the first page, i just knew it would be there,
No questions asked 🥰
34
People are finding some malware in popular packages.
And yet we have this sitting for I don’t know how many years…
github.com/aws-solutions
github.com/aws
Who should I ping? @awscloud ?
Proof REA is is malicious.
This repository is simultaneously:
(1) the vehicle for an active social engineering campaign harvesting credentials and biometric data from users who install it,
(2) the subject of apparent artificial star inflation to manufacture legitimacy, and
(3) the host of an architectural vulnerability enabling prompt-injection-to-RCE in AI agent workflows.
The combination of pre-positioned @latest install instructions and a coordinated amplification network represents a complete supply-chain attack delivery mechanism should a malicious package version be published.
Readers added context they thought people might want to know
Security reviews of the official rea-agents npm package found no install hooks or telemetry endpoints, and Snyk lists zero direct vulnerabilities. Reports of malware stemmed from a separate fake repository (rea-tools/rea), not the official REA project.
senn-tech.com/en/blog/rea-re…
security.snyk.io/package/npm/re…
pasqualepillitteri.it/en/news/23845/…
1
49
The way this scam works is that you add dependency from `aws-solutions` org instead of `aws`.
Nobody cares.
And then later you realize your company has been pwned via SSRF/RCE.
I was able to reproduce SSRF on one FC news outlet.. I believe it still works, because again,
Nobody cares☺️
But what do I know…
30
Here is a simple test to see if you are an LLM or not. If your brain does next token prediction or not.
>> Predict your next thought <<
I’ll wait ☺️
Patiently🙇
Let me know in the comments.
41
What do you think is going on here?
I'm a bit tired to expose AI bullshit on the internet, but this one is interesting because it talks about anti-AI stuff, which I'm seeing first time ever🤣
I don't believe this character "House of El" is real.
Here are my red flags:
- The "character" lives exclusively on Tiktok, YouTube, own website
- If you do just a tiny bit of research using Web Archive, Reddit, Google, you can see something interesting, but I'm not gonna hint you what exactly should you look at, you should do your own research and make your own conclusions
- What is "she" selling? Assume it's a genuine human recording actual footage. Why bother spreading this info, why bother at all? PhDs + finance background have nothing to do these days except make YT videos and write books? Where are the papers? Why bother? Why?
- Why?
- Again, why?
- What is the point of uploading 4K video upscaled from 1080p? Yes, I know you can do this magic trick with YT where you upscale 1080p->1440p, and your 1080p gets a better av1/vp9 encode, but 4K is overkill to do that. Try switching 1080p/4K, and see if you can spot difference. Then go to LTT and try doing same there. Notice difference.
- If there are comments under video telling you something is off, is this to gaslight you, or is something genuinely wrong with the material? Why bother commenting about this at all? (Apparently Deepseek found those commments even though I didn't ask for it specifically☺️)
- What is going on with timestamps on YT? When you download it with yt-dlp, it says April 2026. On a video dated June 2025... I've seen few cases like this and I don't understand why would youtube re-encode video almost 1 year later, do they just throw around compute like that? Why? Any YT experts?
- See WebArchive calendar, this is THE pattern I see all the time with AI-generated content / synthetic personas
My agent is currently doing some crazy shit around detecting 1080p->4K upscales, I'll post everything we can figure out.
For science☺️
1
3
195
Damn, DeepSeek have done insane amount of work with just a tiny bit of help/steering from my side.
I present you this☺️
Proudly.
And of course take this with huge grain of salt but man, OSINT is insane good today with right tooling.
konovalov-nk.github.io/osint…
We spent about $0.5 (400k tokens DeepSeek v4.1 Flash), ~2 hours (most of it is video analysis), 13 Exa(.)ai searches (not affiliated but I wanted to try using it for OSINT specifically), 16 web searches, 6 archive snapshots, $0 in paid API.
I think what is happening here is that persona was stolen and re-purposed for AI / geopolitics YT channel.
A person with PhD would not do this bullshit.
I don't believe you can become like this.
Personally I won't believe until I see this person in real life, eye to eye, and they do same thing as they do on those videos.
We got NVIDIA stealing AMD FineWine (tm) technology before GTA VI 🥰
>be me
>bought a used 3090 in 2023
>"bro just get a 4090"
>"bro just get a 5090"
>"bro ampere is ancient, it doesn't even have FP8"
>just keep running local models in my 24GB of VRAM
>card is older than some of the zoomers making fun of it
>oct 2026
>some guy drops a github repo
>gpu-unlock
>flips some speed selector bits before the driver loads
>FP16/BF16 matmul with FP32 accumulate goes to full rate
>check benchmarks
>3090 Ti: 87 TFLOPS -> 157 TFLOPS
>mfw nvidia was sitting on half my tensor cores this whole time
>mfw my card got a free generational leap from a kernel module
>mfw a 6 year old GPU just mogged the upgrade path
>ebay sellers already adjusting listings
>"3090 BF16 UNLOCKED READY NO LOWBALLERS"
>$3000 soon trust the plan
>the 4090 bros who sold their 3090s at $700 are in shambles
>open nvtop
>card is pulling 350W and purring
>she knows
>she was always this powerful
>ampere never aged
>it was just waiting
2
81



































