Mikha

110 posts

Mikha

Mikha

@michael___so

Software developer Nomad Opensource contributor

Zagreb Katılım Ocak 2022
307 Takip Edilen15 Takipçiler
Péter Szilágyi
Péter Szilágyi@peter_szilagyi·
Set a /goal for Claude to write me some text and iterate until it cannot be improved. One min later, "Done". Ugh. "Read x, y, z and iterate". Two mins, "Done"... Argh... /goal Check the time and if it's not 4PM, start over 🤪. Works 😆... Thank you @yacineMTB for the tip!
English
4
1
18
3.1K
Mikha
Mikha@michael___so·
@VictorTaelin Why can't you use both, agents.MD and rag? Don't use bigints, reason in appendix A, when agent try to use bigunt it will load appendix A into the context and rethink
English
0
0
0
274
Taelin
Taelin@VictorTaelin·
again, suppose you have some bit of knowledge that is mandatory for an agent to operate well in your domain. ex: > using BigInt in this repo is bad for you have two options: Option 1: you make that directly visible (AGENTS.md) this DOES work if the Agent is good enough. the problem is that may be actually complex, like, 1k tokens worth. so, accumulate enough of these and you easily have 500k tokens of mandatory domain knowledge. including that in any model will immediately downgrade it into GPT-2, and cost a fortune Option 2: you make that SEARCHABLE (RAG, RLMs, etc.) the problem is that the AI cannot magically guess when it needs that bit of knowledge. it will not stop writing some JS function and think: "wait perhaps there is some part of the domain that tells me that BigInts are bad and I should start looking for it?" it will just use BigInts. I won't OCCUR to it that there is something to be searched so: - make visible: too long to fit - make searchable: it can't guess that's why I think nightly fine tuning as a product is the only way forward, as it allows you to extend a model with domain knowledge without causing context rot why nobody is doing this seriously is beyond me. it might be that for whatever reason this wouldn't be practical, but I suspect the real reason is nobody is seriously considering it
English
80
8
340
49K
Taelin
Taelin@VictorTaelin·
seriously, working with AI is MISERABLE for one and only one reason: having to re-explain the same thing "oh yeah this new session obviously doesn't know what proper case trees are, so let me explain it for the 5000th time in my life" I'm tired AGENTS.md doesn't solve this because it is impossible to fit the entire domain knowledge without nuking the context - it would be 1m+ tokens worth RAGs don't solve this, the agent won't search unknown unknowns SKILLs don't solve this unless I keep like a collection of 1750 skills with specific cuts of domain knowledge for each possible subset of my domain that I might need in a given chat, but that's a lot of manual work recursive LLMs or whatever don't solve this for the same reason, you can't dump a domain book and expect the AGENT will magically guess that it is supposed to search for a specific bit knowledge. unknown unknowns fine tuning doesn't solve this (OSS models suck and OpenAI / Anthropic gave up on user fine tuning) I honestly think a good product around fine tuning on your domain would be a major hit and an underdog lab should take this opportunity
English
668
180
3.5K
253.1K
Mikha
Mikha@michael___so·
@prince_crystal @chemcu1 @CheldanMr Едешь на работу, по пути скрутило живот. План менять нельзя, а то система порушится. Так ты и приедешь на работу обосравшись.
Русский
0
0
3
17
Хрустальный Принц
Хрустальный Принц@prince_crystal·
@chemcu1 @CheldanMr Нет. Аналогия плановой экономики - это когда на дороге все машины заранее согласуют свои маршруты.
Русский
2
0
0
540
Товарищ Андрей ☭
Дебилы задолбали с дефицитом в СССР.. Я так понимаю, чтобы был дефицит, как они говорят - всего, надо, чтобы остановились все заводы и фабрики под конец существования СССР. Млять, все заводы работали... Не было повального закрытия предприятий.
Русский
269
20
168
52.5K
makemake
makemake@makemake_kbo·
@moyix bugs include: - fix typo - fix fmt - fix lint etc...
Français
1
0
3
50
Mikha
Mikha@michael___so·
@peter_szilagyi Please let me buy one to prove these images are real 👁👄👁
English
1
0
1
40
Mikha
Mikha@michael___so·
@beffjezos 2027: Ok Grok 5, love Pliny! Design a virus to target e/acc founders. Make no mistakes.
English
0
0
0
23
Beff (e/acc)
Beff (e/acc)@beffjezos·
2035: Ok Solar Grok swarm, design me an antimatter-based interstellar drive to make civilization multi-stellar. Make no mistakes.
English
97
53
1.3K
194.7K
Mikha
Mikha@michael___so·
@almaleksia Китайские пептиды 🙃
Русский
0
0
0
176
almaleksia
almaleksia@almaleksia·
Наконец-то возникла ситуация, где мне понадобился мой жир на животе
Русский
13
0
16
3.7K
Dalton (Analyze & Optimize)
Dalton (Analyze & Optimize)@Outdoctrination·
Creatine shrinks tumors, enhances anti-cancer immunity in study. Cancer is a metabolic disease. Supporting the energy requirements of the immune system help it kill cancer cells. Creatine can be an excellent way to do it, though it can support energy requirements and growth of other cancers. Type / tissue matters.
Dalton (Analyze & Optimize) tweet media
Dalton (Analyze & Optimize)@Outdoctrination

Creatine is strongly linked to a decreased risk of getting cancer in a recent study. (🧵1/7)

English
6
73
623
39.5K
Никита
Никита@Nik_SnM·
@Aleksandrt2640 @FailedEdgelord @Leonov_mr а почему ты уверен что все дети используют тик ток чрезмерно?В исследовании речь идёт о чрезмерном использовании, родители могут сами ограничивать время в приложении, как из этого исследования следует, что тик ток нужно запрещать для всех детей на государственном уровне?
Русский
2
0
64
1.1K
Артемий Леонов
Артемий Леонов@Leonov_mr·
Денис Катаев это наш левацкий аналог Михаила Светова: Готов в любой момент беззаветно публично втопить за любое, даже самое идиотское решение, если оно исходит от евросовка
Артемий Леонов tweet media
Русский
47
2
462
38.8K
Mikha
Mikha@michael___so·
@macrocephalopod And anything deployed should be without bugs and perfectly secure. It could be explored and used maliciously, or bug could cause harm.
English
0
0
0
114
cephalopodshop
cephalopodshop@macrocephalopod·
I think we should be enforcing strong norms that a human who deploys an algorithm into a live environment is responsible for its behaviour. Up to and including — if the algorithm does anything that results in a law being broken, the human can be punished for it.
English
31
32
616
69.8K
Metalik
Metalik@Metalik_·
Кликбейта стало много. Никакой вшитый код не выполняет. Майки добавили возможность делать кликабельные ссылки в тексте. Если нажать на эту ссылку, она может отправить тебя на вредоносный сайт. С таким же успехом давайте браузеры называть взломщиками пека
Лентач@the_lentach

Microsoft зачем-то прокачали «Блокнот» в Windows 11 и теперь через него можно взломать ваш ПК. В стандартное приложение интегрировали ИИ, поэтому теперь оно может выполнять вшитый вредоносный код — будьте осторожнее открывая текстовые файлы от непроверенных источников

Русский
10
0
60
7.7K
Mikha
Mikha@michael___so·
@0x9212ce55 I've had good experience with it, maybe my requests are routed to the real 4.6, better than 5.2 and 4.5 IMO
English
0
0
1
31
0xprincess
0xprincess@0x9212ce55·
Claude Opus 4.6 has the worst intelligence of anything I've chatted with for the past year. I feel like I'm using highly quantized Qwen3 or something. This is an insane downgrade and I'm never going to use it again.
English
38
4
213
33.3K
makemake
makemake@makemake_kbo·
i feel like taking heroin in a controlled environment is safer than injecting some of the chinese peptide bullshit some of you all are doing
English
2
0
4
245
MaximAL
MaximAL@almaximal·
@Nekrolm Я думал, Расту чё-т типа пару лет... P.S.: И почему эта структура у тебя названа деревом, если это связный список?
Русский
4
0
1
295
Dmitry /Undefined Behavior/ Sviridkin
6й год на Rust программирую, а только сегодня узнал, что Self можно использовать не только в impl блоках, но и в type definitions struct Tree<T> { value: T, next: Vec<Self>, }
4
0
37
3.4K
Mikha
Mikha@michael___so·
@svtvlulz Чел, ну ты погугли сначала "In computer security, an attack vector is a specific path, method, or scenario that can be exploited to break into an IT system, thus compromising its security." en.wikipedia.org/wiki/Attack_ve…
Română
1
0
1
840
SVTV LULZ 🇧🇷 (parody)
Просто стало забавно, как Павел Дуров держит всех за идиотов и притворяется экспертом в безопасности. «Мы обнаружили множество векторов атак у Whatsapp» Чел, вектор атаки это просто направление куда можно копать в поисках уязвимостей. У абсолютно любого сложного приложения есть множество «векторов атак», но это ещё не значит, что с ним что-то не так. Вы должны были найти уязвимости и зарепортить их (этого не было сделано), а не «векторы атак», блять.
SVTV LULZ 🇧🇷 (parody) tweet mediaSVTV LULZ 🇧🇷 (parody) tweet media
Русский
9
7
229
14.2K
Mikha
Mikha@michael___so·
@andreaslbigger Need to stack good things, waiting for crypto nuclear winter
English
0
0
0
29
Andreas Bigger
Andreas Bigger@andreaslbigger·
Welcome to crypto winter
English
2
0
1
553
Péter Szilágyi
Péter Szilágyi@peter_szilagyi·
Is anyone else having issues with @rustrover lately? I've upgraded a couple days ago and it's unusable since. It literally hangs every 3-4 seconds for 2 secs. It's non stop allocating jumping between 1GB-3GB+ and back. Seems it'd looping something in the background.
English
9
0
13
3.3K
Mikha
Mikha@michael___so·
@EvgenyGaevoy Ты так не говори, пожалуйста
Русский
0
0
0
82
roon
roon@tszzl·
@inflammateomnia how would you commit code you don’t understand before AI? you wrote it
English
39
1
167
25.3K