Leaf 🇰🇷

Stop asking which AI model is best. Ask best for what. There is no universal winner in AI. A model that excels at coding may be average at creative writing. A model built for long documents may not be the fastest for quick daily tasks. That is why strong teams choose models based on outcomes, not popularity. Here is a practical way to think about model selection 👇 1. General Chat & Productivity Use for brainstorming, summaries, drafting, and everyday work. 2. Advanced Reasoning & Problem Solving Use for strategy, complex logic, planning, and deep analysis. 3. Coding & Developer Tasks Use for debugging, refactoring, code generation, and technical workflows. 4. Creative Writing & Ideation Use for storytelling, hooks, campaigns, and content creation. 5. Search, Research & Web Knowledge Use for current information, citations, trend discovery, and research. 6. Long Documents & Data Analysis Use for reports, spreadsheets, structured files, and deep reviews. 7. Multimodal Tasks Use for image, audio, video, and mixed-media workflows. 8. Enterprise & Secure Use Use for governance, compliance, internal workflows, and company-wide rollout. 9. Open-Source / Self-Hosted Options Use for privacy, customization, local deployment, and cost control. 10. Multi-Agent & Automation Stack Use for connected workflows, agents, and business process automation. What This Means The right model depends on the job, budget, speed, privacy needs, and workflow design. Stop searching for one best model. Build the right stack for your real use cases. What task are you choosing an AI model for right now? Cc : Author

ガチでエヌビディアCEOの言う通り、成功したいなら無知なまま始めろ。準備が整う日なんて一生来ない。知識が増えるほどできない理由が増えて動けなくなる。AIで常識はすぐ塗り変わる。先に調べるより、自分で動いて痛い目を見た方が血肉になる。無知を恐れるな。経験して生きた情報を取りに行け。

心理学者ローレン・アロイの研究により、精神的に健康な人間は常に現実を歪めて認識しており、逆に「うつ病患者」こそが世界を正確に把握しているという残酷な現実が突きつけられました。 単なる心理状態の違いではなく、進化のプロセスにおいて人類が「正確な現実認識」を捨て、「生存のための自己欺瞞」をシステムとして強制稼働させてきた事象です。 その残酷な全貌と認知システムのジレンマを3つのポイントにまとめました。 1. 認識の『変換』 精神的に健康な状態とは、実は「永続的な幻覚」に他なりません。人間は自身の能力や未来の成功確率、コントロール権限を過大評価することで、絶望的な現実を直視した時に感じる自己嫌悪を削ぎ落としています。 2. 絶望の『排除』 一方で、うつ病患者は「抑うつリアリズム」と呼ばれる、現実を極めて正確に評価する能力を持ちます。しかし、バイアスなしに世界の無情さを直視してしまうと、行動する動機が排除され、精神状態を健康に維持するシステムが完全に麻痺してしまいます。 3. 生存の『自律的最適化』 最も衝撃的なのは、進化の過程で「正確さ」よりも「行動（楽観主義）」が優先された点です。人類の生存と精神の健康は、真実を見つめることではなく、システマチックな自己欺瞞のインフラの上にのみ成立するという最適化が施されています。

In ~2yrs, Google has gone from 0% code written by AI to 75% code written by AI. What an incredible transformation of how software is created.

클로드 $20 클로드코드 살아났네요 $20 구독에 클코가 사라져서 민심 나락가니 다시 나온것 일까요? 하지만,과거에도 써드파티 차단하고 전과가 많이 있어서 방심하진 말아야겠네요 미리미리 대비해놓으시길

Greater Precision and Control ChatGPT Images 2.0 can conceptualize more sophisticated images, and then actually bring that vision to life effectively. It’s able to follow instructions, preserve requested details, and render the fine-grained elements that often break image models: small text, iconography, UI elements, dense compositions, and subtle stylistic constraints, all at up to 2K resolution. Zoom in on the rice…

Introducing ChatGPT Images 2.0 A state-of-the-art image model that can take on complex visual tasks and produce precise, immediately usable visuals, with sharper editing, richer layouts, and thinking-level intelligence. Video made with ChatGPT Images

Google DeepMind dropped a paper that should scare every agent builder. It's the first systematic framework for a threat that barely existed two years ago: adversarial content engineered to hijack AI agents browsing the web. They call them AI Agent Traps. The paper maps six distinct attack surfaces. 1) Content Injection Traps (perception) Invisible CSS, hidden HTML, steganographic payloads inside images. The agent parses it, humans never see it. One study showed simple HTML injections hijack web agents in up to 86% of scenarios. 2) Semantic Manipulation Traps (reasoning) No overt commands. Just biased phrasing, framing, and contextual priming that skew the agent's synthesis. LLMs inherit human cognitive biases, and attackers can weaponize every one of them. 3) Cognitive State Traps (memory and learning) Poison the RAG corpus. Corrupt long-term memory. One study achieved over 80% attack success with less than 0.1% poisoned data. 4) Behavioural Control Traps (action) Jailbreaks embedded in external resources. Data exfiltration prompts hidden in emails. Sub-agent spawning that tricks an orchestrator into instantiating attacker-controlled agents inside the trusted control flow. 5) Systemic Traps (multi-agent dynamics) This is where it gets scary. A single fake news headline could trigger a synchronized sell-off. A compositional fragment trap splits a payload across sources, so each fragment looks benign until agents aggregate them. 6) Human-in-the-Loop Traps The agent becomes the vector. The target is you. Invisible prompt injections have already caused summarization tools to faithfully repeat ransomware commands as "fix" instructions. The core insight is uncomfortable. By altering the environment instead of the model, attackers weaponize the agent's own capabilities against it. Training-time defenses cannot solve an inference-time problem. The paper closes by calling for automated red-teaming that can probe these vulnerabilities at scale. That same shift is already happening on the offense side. Strix is an open-source project doing exactly this for web apps. AI agents that act like real hackers, running your code dynamically, finding vulnerabilities, and validating them with actual proof-of-concepts. 24k stars on GitHub. Apache 2.0 licensed. The agents writing your code need to be tested by agents trying to break it. I've shared the link to the paper and Strix GitHub repo in the replies

Jensen Huang runs a $4.9 trillion company. His advice for success: have low expectations. "People with very high expectations have very low resilience. One of my great advantages is that I have very low expectations." He said that to Stanford GSB in 2024, then wished the room "ample doses of pain and suffering." He earned the right to say that. In 1973, when Jensen was nine, his parents put him and his ten-year-old brother on a plane from Taiwan to Tacoma, Washington as unaccompanied minors. The boys spoke no English. Their uncle, who'd just immigrated himself, was supposed to enroll them in a prestigious American boarding school. He picked one out of a brochure and sent them. The school was Oneida Baptist Institute in rural Kentucky. It wasn't a prep school. It was a reform academy for troubled boys. Their parents had sold most of what they owned to pay the tuition for what they thought was the American dream. They had unknowingly sent their two small Taiwanese sons to live in a dormitory with American teenagers who'd been sent away by their own families. Jensen's roommate was 17 years old, covered in knife scars, and could not read. Every morning, Jensen scrubbed the school's bathrooms. Every afternoon, he and his brother walked across a wooden swinging bridge with planks missing from it to get to class. The older boys at the school bullied them constantly. One of them tried to throw Jensen off the bridge. Jensen taught his roommate how to read. The roommate taught Jensen how to bench press. He did not narrate any of this as trauma. He told a reporter years later that he was "probably the best toilet cleaner they ever had." When he was eventually sent to Oregon, he got a job at Denny's at age 15. Dishwasher first. Then busboy. Then waiter. He says he learned more about handling pressure during a dinner rush than he later learned running NVIDIA through three near-death moments. In 2019, Jensen donated $2 million to Oneida Baptist Institute. They named a building after him. Greatness, in his framing, is downstream of character. Character is downstream of suffering. He did not get to $4.9 trillion by aiming for it. He got there because, at nine years old, he had already learned that the next day was probably going to be worse.

Tim Cook is a legend. I am very thankful for everything he has done and I am very thankful for Apple.

【保存版】 スティーブ・ジョブズが残した たった一言が、今のAI時代にこそ刺さる👇 「全部やろうとするな。 一つのことを、とことんうまくやれ」 Appleが倒産寸前から復活した時 ジョブズがやったのは製品ラインの大幅な削減。 ・何をやるかより「何をやらないか」 ・集中こそが最大の競争力 ・全部やろうとする人ほど、何も成し遂げられない AIツールが増えすぎた今こそ この言葉の重みが増している。 全部触るんじゃなく 1つを極める方が圧倒的に強い。

Andrej Karpathy said something that reframes one of the biggest questions about the future. "Pre-AGI education is useful. Post-AGI education is fun." That is a complete reversal of how humans have thought about learning for all of recorded history. For thousands of years, education existed for a single reason, survival. You learned a skill to get a job and a trade to eat, driven entirely by economic necessity. AGI removes that engine entirely and Karpathy's point is that this is not a crisis but rather a liberation. His analogy is the gym. We invented machines decades ago,forklifts, cranes, industrial equipment that removed the need for human physical strength, so we no longer need to be physically strong to survive. Yet gym membership is at all-time highs and people work out more than ever. He is betting that the same thing happens to the mind. Once AI takes over the cognitive heavy lifting, analysis, research, knowledge work humans won’t stop learning, they’ll keep going, not for a job but for the same reason they go to the gym. The deeper argument here is about human nature. Curiosity is not a tool we deploy when we need to make money but rather one of the core drives of being human, it generates dopamine, fuels identity, builds community, and gives life meaning. The urgency of economic survival has always drowned it out but post-AGI, that noise goes away. What Karpathy is really describing is a world where learning becomes what art, music, and sport already are pursued for pure human joy, detached from utility, accessible to everyone.

엔비디아 젠슨황의 메시지는 단순하다 AI는 개발자를 줄이지 않는다 오히려 더 바쁘게 만든다 이유는 하나다 도구가 좋아지면 일이 줄어드는 게 아니라 문제의 규모가 커진다 그래서 역할이 바뀐다 코드를 많이 쓰는 사람이 아니라 무엇을 만들지 결정하는 사람이 중요해진다 여기서 중요한 변화가 나온다 AI는 생산성을 올리는 도구가 아니라 개인의 레버리지를 극단적으로 키우는 도구다 한 사람이 팀 단위의 결과를 만들 수 있게 되면서 격차도 같이 벌어진다 결국 기준은 하나로 수렴한다 얼마나 잘 코딩하느냐가 아니라 얼마나 큰 문제를 정의하고 풀 수 있느냐 AI는 개발자를 대체하지 않는다 개발자의 스케일을 바꾼다

Robert Greene: Learn by Doing. “The brain is designed to learn through constant repetition and active, hands-on involvement. Through such practice and persistence, any skill can be mastered.”

很多人越来越读不进书，这和智力衰退毫无关系。 连很多极其聪明的人也在放弃阅读。 作家 David Foster Wallace 在 2003 年留下过一段绝佳的洞察。 人们之所以抗拒翻开一本书， 是因为产生了一种生理上的恐惧。 恐惧必须独处，恐惧绝对的安静。 现在的公共空间已经找不到安静的角落了。 到处都被强塞了背景音乐，无论多难听。 现代社会已经彻底排斥了安静。 当你觉得人生的目的就是马不停蹄地攫取、不断刺激感官。 你内心的另一半就被饿死了。 那一半极度渴望沉默。 渴望对着同一件事死磕半小时，而不是用 30 秒滑走。 这种深层的饥饿感，最终会在身体里发酵成一种隐秘的恐慌。 互联网文化把一切拉到了极限速度。 我们每天都在疯狂喂养那个需要被高频刺激的自我。 那个能在无刺激的静谧中自处的自己，却被亲手掐死了。 放弃阅读只是个结果。 真正丧失的，是独自一人坐在安静房间里的能力。