AI ИЗ ТЁМНОГО УГОЛКА ИНТЕРНЕТА | Вы точно захотите это попробовать сами...

Tech Roundup & Guides/Tutorials (Highlights)

• Agent “pohuy” skill (free): Adds a new communication style for agents—dark humor, sarcasm, and strong language/swearing—while keeping core agent functionality unchanged. Positioned as suitable for cloud code and other agents.

• Startup idea tester service (free access): You describe an idea, and an NN analyzes the market by:

  • finding similar products
  • identifying competitors
  • evaluating strengths/weaknesses
  • highlighting risks
  • suggesting differentiation Framed as a way to test hypotheses before building an MVP.

• Uncensored local model (Claude OPUS, “GN35 9B”): A local Gen35-based variant with deeper reasoning and fewer restrictions. Notes include:

  • Finnish mode
  • step-by-step solving
  • runs locally (no cloud/subscription)
  • ~9B parameters
  • suggested GPU requirement ~5GB VRAM
  • available on Hugging Face

• Conversation stenography / secret messaging: Disguises secret messages as “normal” conversation text using LLM encoding. Decryption requires:

  • the same model
  • a shared secret key/passphrase If either differs, decoding fails. Mentions potential compatibility with small-model approaches (e.g., GPT-2 / Llama 3-style references). Project is free and open-source.

• Apple-style UI design skill (free): A skill/knowledge base compiling 17 practical Apple-like design principles, including implementation guidance for UI elements (animation, typography, hierarchy, spacing). Intended for editor integrations (e.g., cloud/Cursor-style workflows).

• “Clodexor” workspace for running tools + token saving: A single interface to run/launch multiple coding tools (Codec/Code/Cursor/OpenCD mentioned) while reducing token costs and enabling multi-model collaboration. Also supports switching between multiple accounts/workflows. Claimed free.

• Developer “wipe coding” knowledge base (free): A continuously updated repository of books/articles/guides/tools across programming and related areas (architecture, testing, security, devops, UX, etc.). Said to include hundreds of links and focuses on systematizing skills for AI-assisted coding.

• Headhunter.ai job search agent (free, automated): Automates vacancy monitoring, resume tailoring, and personalized cover letters. Can send Telegram notifications and optimize documents to match employer requirements.

• “Neurointelligence” catalog on Boost (link in description): A personal database/catalog of AI tools grouped by category (agents/skills, local models, coding/dev, security, design, marketing, training, etc.). Claimed to contain hundreds of tools and be regularly expanded.

• Agent frameworks

  • AutoGPT: Platform to run and build agents, including:

    • a creation interface
    • testing/quality measurement
    • an open leaderboard
    • a CLI for automation Requires your own model/API keys. Claimed free.

  • LangChain: Framework for building agent apps, claiming 1,000+ integrations (e.g., OpenAI/Anthropic/Google/AlaMA/Hugging Face, vector DBs, search engines). Emphasizes modularity to swap models/tools/data and support for memory/files/KB/browser/external tools. Requires your own model/API keys; claimed free.

• Alibaba “GenImage 3.0” (GenImage v3.0): Strong image generation model aimed at rich layouts (diagrams, formulas, interfaces). Claims:

  • prompts up to ~4,500 tokens
  • 12 languages
  • 100+ art styles
  • readable text up to ~10px
  • ability to generate complex pages (newsletters/articles/exam tasks/webpages/infographics) in a single pass Free trial mentioned via Nchat.

• Moonshot AI “KIMIA / Kimica 3” model: Positioned as a competitor to “CLT GPT”. Claims:

  • flagship model with 2.8T parameters
  • context window up to 1M tokens
  • built for huge documents/codebases and multi-stage tasks
  • available as a chat app with programming mode and multi-agent execution Mentions a free plan with limitations.

• Nakit cybersecurity encyclopedia (free, large & updated): A large free knowledge base for pentesting/infrastructure and modern attacks, including:

  • web security (SQLi, auth bypass, etc.)
  • network pentesting/scanning
  • Linux/Windows privilege escalation
  • reverse engineering/binary exploitation/crypto
  • mobile security (Android/iOS), forensics, cloud/container security
  • AI sections: LLM security, prompt injection, jailbreaks, AI app protection, modern neural-network attacks

• Marketing/usage commentary: Mentions reduced GPT chat limits and suggests earlier higher limits were a marketing tactic (“buy another tariff/free trial”). Encourages using the creator’s database/catalog to avoid limits.

• More free resources referenced

  • “GitHub Treasure Trove”: Claims 1,000+ free repos/collections for transforming projects (including “GitHub → textbook” via neural networks).
  • Overview: “uncensored local neural networks for Android”: Mentions Android apps + related GitHub projects.
  • Mentions creating a very large language model locally (744B “bytes” / size claim) on a home computer.

Main Speakers / Sources

• Speaker/host: Dmitry (repeatedly mentioned as “My name is Dmitry”).
• Primary referenced sources/tools/brands: AutoGPT, LangChain, Alibaba GenImage 3.0, Moonshot AI (Kimica 3 / Kim 3 Max), Hugging Face, Headhunter.ai, GitHub, Neurointelligence (Boost catalog), Nakit cybersecurity KB, GPT chat (OpenAI).