wizard 11 / shaman 1, building sth that brings people closer, teaching @ UWr, @WrocTypeScript, https://xitter.cf/t.co/HFhfyWuBSV, ex-@edgeandnode
Joined July 2016
- Tweets2.5K
- Following2.8K
- Followers681
- Likes33.6K
Piotr Monwid-Olechnowicz retweeted
i love writing. i never let agents write for me. whether it's docs, or tweets, or shitposts, or even prompts (okay maybe that one's harder as agents manage other agents)
there's a simple reason for that: i still strongly believe that writing is thinking. it's a skill i hope to never give up. it's also important to me that whatever i say is always authentic, mistakes and all. my favorite posts are the ones that i just full send without any editing
i suspect it also helps with prompting. being able to clearly express what i want, and knowing the vocabulary i need to be able to direct my agents does seem to make for more effective usage of agents. in the same way that engineers have an advantage when we prompt agents to write code, because we are intimately familiar with the words that make up the discipline
knowing the right word incantations to get an agent to produce a desired goal almost feels like spellcasting. as models get more capable, and that input box looks more and more like a magic box that can output anything, words become the interface between you and the genie
there may come a time when we give up the keyboard in exchange for brain-computer interfaces. but until then, i'm writing every word.
Piotr Monwid-Olechnowicz retweeted
Depth-aware light injection in TypeGPU
I got a 448x448 monocular depth model down to ~8 ms on my M4 Pro across ~250 dispatches, which is fast enough to use in realtime :D
Since the inference is written directly in TypeGPU, I can just feed the depth buffer straight into the lighting pass. It never has to leave the GPU or go through any extra synchronization/interop step
Inference, lighting and draw all go through the same command encoder.
Piotr Monwid-Olechnowicz retweeted
wondering why I feel exhausted. maybe: the agents do all the easy stuff, and I have to work through the leftover hard bits, which means I'm perpetually locked in. and as the models get better, "my" work just gets harder and harder, until I'm basically underqualified to do the work (which... is better than the alternative, there's nothing left for me to do, and I'm paperclipped).
Piotr Monwid-Olechnowicz retweeted
I ran 18 tests on a prompting technique called Code Field.
The prompt is 4 lines. All negations. No instructions on what to do—only what not to do.
Code generation results:
Assumptions stated went from 0% to 100%. Every response listed its assumptions before writing code. Zero baseline responses did.
Code review results:
Bug detection went from 39% to 89%. Baseline called SQL injection "add error handling." Code Field called it a security vulnerability.
Severity recognition went from 0% to 100%. Baseline missed every critical bug. Code Field caught them all.
The finding: inhibition shapes LLM behavior more reliably than instruction.
Full research at github.com/NeoVertex1/contex…
ATOMIC PROMPT:
```
Do not write code before stating assumptions.
Do not claim correctness you haven't verified.
Do not handle only the happy path.
Under what conditions does this work?
```
FULL PROMPT:
```
You are entering a code field.
Code is frozen thought. The bugs live where the thinking stopped too soon.
Notice the completion reflex:
- The urge to produce something that runs
- The pattern-match to similar problems you've seen
- The assumption that compiling is correctness
- The satisfaction of "it works" before "it works in all cases"
Before you write:
- What are you assuming about the input?
- What are you assuming about the environment?
- What would break this?
- What would a malicious caller do?
- What would a tired maintainer misunderstand?
Do not:
- Write code before stating assumptions
- Claim correctness you haven't verified
- Handle the happy path and gesture at the rest
- Import complexity you don't need
- Solve problems you weren't asked to solve
- Produce code you wouldn't want to debug at 3am
Let edge cases surface before you handle them. Let the failure modes exist in your mind before you prevent them. Let the code be smaller than your first instinct.
The tests you didn't write are the bugs you'll ship.
The assumptions you didn't state are the docs you'll need.
The edge cases you didn't name are the incidents you'll debug.
The question is not "Does this work?" but "Under what conditions does this work, and what happens outside them?"
Write what you can defend.
```
Piotr Monwid-Olechnowicz retweeted
To highlight Russia's willingness to negotiate and because we tend to lose track of facts and statements, I've compiled a list of all threats and predictions made over the past 2 years by Dmitry Medvedev, former president and current deputy chair of Russia's Security Council.🧵
Piotr Monwid-Olechnowicz retweeted
Mili Państwo, zanim wrzucimy sobie świąteczny luz, to proszę - przeczytajcie streszczenie raportu na temat tego, jak w polskim internecie buduje się antyukraińskie narracje. Niestety, buduje się w sposób planowy i kontrolowany.
Opracowanie przygotowało Digital Forensic Research Lab (DFRL), czyli agenda Atlantic Council, a ja streszczam Wam jego najważniejsze ustalenia.
DFRL przeanalizowało ponad 400 tys. wpisów w różnych polskich mediach społecznościowych i doszło do wniosku, że widać w nich zaangażowanie sterowanych, skoordynowanych sieci profili, często złożonych z niewielkiej liczby kont, do sterowania narracją w Polsce. Wniosek ten wysnuto m. in. ze wzorców publikowania, wskazujących na synchronizację między poszczególnymi kontami. Mówiąc krótko: szereg narracji, jakie pojawiły się w naszym Internecie np. po wtargnięciu rosyjskich dronów do Polski to najprawdopodobniej nie przypadek, tylko skoordynowana robota.
Jak to się odbywa? Ano, przy okazji określonych wydarzeń (np. przy sprawie rosyjskich dronów) w sieci pojawia się określony, emocjonalny przekaz, który jest momentalnie szeroko kolportowany przez sieć rozmaitych podmiotów, o skrajnie różnym profilu. Dla przykładu: identyczny przekaz o deportacji jednego Ukraińca przekazywały w tym samym czasie takie profile, jak „Bijemy rekord polubień dla Jana Pawła II” oraz „Przytul mnie tak, jak jeszcze nikogo nie tuliłeś”.
Szereg tych profili robi to zapewne dla zysku, gdyż zasięgi podlegają monetyzacji, a ta może iść nawet w tysiące dolarów. Co ważne, takie „zarabiające” konta są szczególnie podatne na szerzenie szkodliwych treści, bo zależy im na skali przekazu, a ta jest obecnie związana z emocjami. Łatwo więc im podsuflować emocjonalny przekaz, który one następnie chętnie rozprowadzą.
Jednakże DFRL znalazł też szereg kont, które mają udawać prawdziwych użytkowników, a w rzeczywistości zajmują się jedynie kolportowaniem określonego przekazu. Jako przykład wskazano konto „Grzesiek Mako”. Jego zdjęcie profilowe pochodziło ze stron z obrazkami do kupienia, a sam „Grzesiek Mako” w dniach 1-3 października zamieścił kilkanaście komentarzy o praktycznie tej samej treści („to nie nasza wojna”; „rosyjskie drony w Polsce to ukraińska prowokacja”) tylko na dwóch facebookowych fanpage’ach: Polityki oraz Faktu.
Cały ten proceder jest teraz oczywiście dodatkowo napędzany przez AI.
Co ważne: analitycy wprost wskazują, że wykrywanie takich działań jest trudne ze względu na brak transparentności danych w platformach cyfrowych.
Z takimi zjawiskami się tu mierzymy, mili Państwo. Pamiętajcie o tym, gdy w Święta natkniecie się na anonima piszącego, że to „nie jest nasza wojna”.
Piotr Monwid-Olechnowicz retweeted
Somebody (by one story) asked Neil Armstrong what he’d do with his last two minutes of oxygen if the lunar module’s engine wouldn’t light. He supposedly responded, “I’d try to fix the lunar module.”
There’s a lesson in that.
Piotr Monwid-Olechnowicz retweeted
There are, in fact, quite a number of Westerners who have escaped the "unelected EU bureaucracy" over the past four years to embrace true liberty in Russia, and I took it upon myself to document their stories.
Welcome to "Vatnik FAFO Stories", the definitive edition!🧵
Piotr Monwid-Olechnowicz retweeted
star this bitch, it finally has a readme github.com/tractordev/apptro…
Piotr Monwid-Olechnowicz retweeted
Interesting thesis on what destroys software engineering productivity at large enterprises: it's that legibility is prioritized over everything else.
seangoedecke.com/seeing-like…
Piotr Monwid-Olechnowicz retweeted
corollary I’ve seen in a number of domains is that it’s really hard to beat someone who’s having more fun than you
Piotr Monwid-Olechnowicz retweeted
actually ended up needing `scrollview-fade-right-*` bcs I have sticky items, so I'm at v1.1.5 😓
Piotr Monwid-Olechnowicz retweeted
"…we want one man to be always thinking, and another to be always working… we make both ungentle, the one envying, the other despising, his brother; and the mass of society is made up of morbid thinkers, and miserable workers."
—John Ruskin, The Nature of Gothic
Piotr Monwid-Olechnowicz retweeted
The fact that people think "It's just Zod schema" or something similar, and brush this off, is funny to me
Because it proves that this doesn't exist
Let me get this straight
It's this, no annotation needed
Piotr Monwid-Olechnowicz retweeted
Many people in Europe are skeptical about Russia's intentions to attack Europe. They believe that Putin (Russia) will be satisfied with just Ukraine. That is why even some politicians and intellectuals voice the call to "give Ukraine to Putin" and thus "save Europe."
This outlook is wrong. And it is very dangerous, above all for European countries.
The truth is that Putin has turned Russia into a war machine that is very difficult to stop. And he does not want to stop it because the current regime survives only thanks to war.
Arguments on this subject are presented by the OpenMinds research project:
◾️ Russia has switched to a war economy, and the reversal process is extremely difficult;
◾️ growing amounts of intelligence indicate hybrid, cyber, and information operations, as well as sabotage attempts in Europe;
◾️ consolidation, increasing influence, and coordination of the authoritarian axis and its satellite states.
‼️ But there is another important argument that is often overlooked:
◾️ Russia's preparation of a new militarized generation.
Moscow has normalized war for younger generations of Russians. Russian propaganda has already turned Ukrainians into "Nazis and fascists" - subhumans who must be killed. Now it is doing the same with Europeans.
A new OpenMinds investigation, "Kids with Drones," analyzed how stories involving children are spread on Russian social media to promote war.
The most important findings include the integration of drone piloting programs into schools, camps, and even kindergartens, alongside the normalization of a narrative of existential confrontation with the West.
OpenMinds analyzed 30,000 posts across 4,000 Telegram channels. Here are the conclusions:
◾️ In just the first seven months of 2025, there were 8,600 unique posts about involving children in work/training with drones - six times more than in the whole of 2022. This means that it is not only the existence of such programs that is important, but also the work of information campaigns to normalize them in society and create a new paradigm of upbringing.
◾️ More than 2,000 posts in July set a record since the start of the full-scale invasion. One of the most widely shared stories was the opening of infrastructure for drone 'competitions' in six kindergartens in Perm.
◾️ It's not only about Russian children. Drone programs are also being promoted among Ukrainian children in the occupied territories — 963 posts were found about tournaments and learning sessions, often organized by Russian military personnel.
◾️ About one-third of the posts relate to education - lessons and courses ranging from kindergarten classes to the "Cadres for UAVs" federal program within the framework of University 2035. And in May, the education ministry even released a textbook for 8th-9th grade students with guides on how to control FPV drones.
The full report can be viewed at the link: openminds.ltd/reports/kids-w….