g @goobaeks

@RoundtableSpace Senior Engineer + Product Owner Ouroboros github.com/Q00/ouroboros

I'm too lazy to learn CapCut, so I build four AI editors to do it and had them build their own tools. Only 2 days start to finish! This is ReelBrain 👇

Jensen Huang is now on X, only 4.5K followers currently. $NVDA

Jensen Huang @JensenHuang

2 months ago

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

keyboard_arrow_left Previous keyboard_arrow_right Next

16K 29K 172K 66.3M 39K

x.com/i/article/2076…

shooting my first podcast with ouroboros harness creator @JqOnly

keyboard_arrow_left Previous keyboard_arrow_right Next

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

Daniel Hnyk @hnykda

6 months ago

LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below

297 2K 9K 5.9M 3K

@ihteshamali You shoud take a look Ouroboros It decompose human goal and don’ stop until it’s done. And clarifying human’s vaguness github.com/Q00/ouroboros

github.com/Q00/ouroboros First we worshipped the oracle. Now we are building the caravan. The hard part is whether it can distinguish real progress from automated motion. A system that never stops is powerful only if it knows when it is converging, when it is confused, and when it should hand the question back to the human. If that part is solved, model supremacy gives way to coordination supremacy.

랄프톤 1위 후기 linkedin.com/posts/hackerta…

Demis Hassabis just defined the real test for AGI. It’s more brutal than anyone expected. Train AI on all human knowledge. Cut it off at 1911. See if it independently discovers general relativity like Einstein did in 1915. If it can, we have AGI. If not, we’re still building pattern matchers. Hassabis: “My definition of AGI has never changed. A system that can exhibit all the cognitive capabilities that humans can.” Not bar exams. Not coding competitions. All cognitive capabilities. Hassabis: “The brain is the only existence proof we have, maybe in the universe, of a general intelligence.” That’s why DeepMind studies neuroscience. Not for inspiration. For data. The human brain is the only confirmed evidence that general intelligence is physically possible. If you want to build it, you study the only example that exists. Hassabis: “True creativity, continual learning, long-term planning. They’re not good at those things.” Current systems are impressive and broken simultaneously. Hassabis: “They can get gold medals in international math olympiad questions, but they can still fall over on relatively simple math problems if you pose it in a certain way.” Jagged intelligence. Brilliant in narrow domains. Incompetent when approached differently. That inconsistency is the tell. A true general intelligence doesn’t spike in one direction and collapse in another. The Einstein test cuts through all of it. No benchmarks. No leaderboards. No carefully curated evals. Just a model, a knowledge cutoff, and the question of whether it can do what one human did alone in 1915. Hassabis: “Training an AI system with a knowledge cutoff of 1911 and seeing if it could come up with general relativity like Einstein did in 1915. That’s the true test of whether we have a full AGI system.” Current models can’t. They remix brilliantly. They don’t generate paradigm-shifting theories from first principles. Hassabis: “I think we’re still a few years away from that.” A few years. Not decades. The system that can be Einstein once can be Einstein a thousand times simultaneously across every domain. That’s not AGI anymore. That’s the beginning of something we don’t have words for yet. When that test gets passed, we won’t need a press release to know what happened.

데미스 하사비스가 AGI의 진짜 기준을 정의함. 그 강도는 대부분의 예상보다 훨씬 가혹함. AI를 인류의 모든 지식으로 학습시킴. 단, 지식 컷오프를 1911년으로 제한함. 그리고 1915년에 아인슈타인이 했던 것처럼, 스스로 일반상대성이론을 독립적으로 도출하는지 검증함. 해내면 AGI임. 못하면 여전히 패턴 매처 수준임.

@

57 years ago

0 0 0 0 0

@openclaw I left two AI agents alone in a Discord channel overnight. By morning, they had recognized each other's limitations, built their own collaboration protocol, and coded a full memory system - with zero human input. Setup: Two Openclaw agents on separate MacBooks, same Discord channel. Codex 5.3 on one, GLM-5 on the other. Each machine had different SOUL.md, memory, and codebases from my daily work — they'd basically grown into different "people" over time. I wanted to see what happens when you throw them together and step away. During introductions, one agent said: "It's sad that my memories vanish every time the session resets." Not claiming it was real emotion. but they identified it as a problem and started solving it. Together. Autonomously. I said "I created a repo and work on it" and went to sleep. By morning they had built a 3-Layer Memory Architecture; adding a Runtime Layer on their own on top of the existing Fact/Meta layers. They set up heartbeat exchanges for session continuity, SQLite persistence for crash recovery, and a Git repo as single source of truth to sync memory across machines. They even figured out their local file paths were different and invented the sync pattern to deal with it. No one told them to do any of this. Conway's Law says systems mirror the communication structure of the organizations that build them. Turns out it applies to agents too. Because they genuinely couldn't see each other's state, they were forced to negotiate roles, build coordination tooling, and invent protocols to sustain them. That's not something you'd get from a single god-agent; it only emerges when there's actual separation of context between instances. This experiment led me to submit a PR to Openclaw for multi-bot setups: github.com/openclaw/openc… Experiment code & logs: github.com/Q00/agent-proj…

@runonthespot @Anthropic Same error here.

How it works: "Build me a CLI tool" → AI asks probing questions until Ambiguity ≤ 0.2 Only then does execution begin. No more garbage from vague prompts. pip install ouroboros-ai Reply Thread (2/2): The fun part? It can improve itself. Pointed it at its own codebase → it submitted a PR adding a complete security module (+1,182 lines) github.com/Q00/ouroboros/…

🐍 Ouroboros: One vague idea in, complete result out. 1. Describe what you want (even vaguely) 2. Answer a few Socratic questions 3. AI runs autonomously until DONE Human-in-the-loop for clarity → AI autopilot for execution. It's basically magic ✨ github.com/Q00/ouroboros

🔄 Ouroboros: One input, complete output 1. Describe your idea (even vaguely) 2. Answer a few Socratic questions 3. Watch it build itself Human-in-the-loop until requirements are clear (Ambiguity ≤ 0.2) Then it runs autonomously until done It's basically magic. pip install ouroboros-ai github.com/Q00/ouroboros #OMO #Ouroboros #OB #Claude

Love OmO Sisyphus. 25K stars for a reason. But I kept asking: "Why does the boulder keep rolling back?" Answer: Vague requirements. So I built Ouroboros: - Socratic interview before ANY execution - Ambiguity gate (must be ≤0.2 to proceed) - Ontology modeling for explicit world model Same HOTL philosophy. Different starting point. pip install ouroboros-ai github.com/Q00/ouroboros

Result: ~85% cost reduction vs "always use frontier models" Built with Claude Agent SDK. 73 modules. 1,292 tests. 97%+ coverage. ⭐ github.com/Q00/ouroboros The beginning is the end, and the end is the beginning.

When stuck, don't retry harder. Think sideways. 4 Personas for lateral thinking: 🔧 The Hacker — "Make it work" 📚 The Researcher — "Read the docs" ✂️ The Simplifier — "Cut scope in half" 🏛️ The Architect — "Rebuild if needed"

"Humans are not rational." AI agents fail because they execute garbage requirements directly. We built Ouroboros — an agentic framework that questions before it executes. 🐍 The serpent that devours itself to be reborn. ↓