A Markdown folder is still not a note system
A folder full of .md files is still only a file collection: wiki links do not resolve and there is no reverse index, it is not a note system until links exist in the system.
Open notebook
Ideas under construction, evidence in public.
Writing about the work of making intelligent systems understandable, governable and useful.
RSS ↗↗
Restricting the frontier requires proving narrower defenses failed
Restricting or deliberately slowing frontier AI capability in the name of catastrophic risk must carry the burden of justification, including the opportunity cost of delayed defenses and medical or scientific progress. That logic applies to risk governance in general, not only bioweapon narratives.
The prompt describes the cycle; ESAA institutes it
Codex-on-GPT-6 agent instructions already narrate claim/complete, evidence rules, and completion audits in natural language. ESAA formalizes the same lifecycle outside the model: event sourcing and a deterministic orchestrator.
Formalization closes the statement, not the dispute
OpenAI published a Lean proof of forced blowup for Navier–Stokes, matching Clay’s (C)/(D). That closes a formalizable mathematical statement; it does not by itself settle priority, what the community calls “the” problem, or prize acceptance.
AI got more aligned and harder to audit
GPT-6 Astra produced fewer severe violations in one deployment simulation while making written reasoning less legible to monitors. Safety now requires the full trajectory.
Auditable control of action trajectories became the scarce resource
GPT-6 Astra drives the computer on a minutes scale and clears the Critical cyber bar. The bottleneck shifts from readable CoT to tools, duration, and access gates.
Before delegation, intention must become an artifact
Anthropic’s AI-native SDLC playbook puts intent.md at the center: why, outcome, and constraints leave the chat and enter version control before an agent writes code.
Hypotheses got cheap; the lab is still scarce
DeepMind’s Co-Scientist designs experiments while humans execute in the lab. The real advance is not full autonomy: it is the bottleneck moving from hypothesis generation to experimental capacity and verification.
The teammate stopped being a session
Grok Bot for Enterprise treats each Bot as a worker with its own cloud computer, a routine learned once, and Bot-to-Bot context handoff—not a chat that vanishes when the tab closes.
A complex harness is the exam rankings do not run
Gemini 3.8 Flash joined the small group of models that can do daily work: it survived a complex harness, with a large context, without losing the thread. That is the same threshold Grok 4.5 crossed when it became usable for programming.
Intelligence now depends on the badge
Fable 5.1 and Mythos 5.1 share a model but change access, safeguards, and retention. The release shows why capability alone no longer describes an AI product.
Multimodal latency starts before the GPU
A vLLM change shows how images, audio, and video can delay a response before inference begins, and why useful concurrency must preserve order, failures, and equivalence.
The GPU stack now comes with an operations agent
ROCm 10 signals a shift in AI infrastructure: installation, diagnosis, and tuning are becoming executable, testable workflows rather than documentation alone.
Long context became an architecture problem
Qwen3.8-Flash-Next illustrates an important shift: extending context now means combining specialized paths for memory, selection, and scale rather than relying on one attention mechanism.
The model is a brain in a jar
Two products can share a model and still feel like different systems. The difference is usually the harness: tools, memory, and the loop that turns reasoning into action.
Context needs a type system too
New Codex changes show why truncating, compacting, and splitting a conversation must preserve not only content but also the source and role of each part.
A screen records steps, not the work
A new method turns computer-use traces into auditable models of goals and procedures, showing measurable gains and a serious privacy boundary.
The agent learned to speak without stopping
Asynchronous updates and thread-scoped approvals show what changes when an agent must inform, continue, and keep contexts separate.
A cache must know when data is still alive
Three vLLM fixes show how preemption, overlapping copies, and late notifications can turn AI caching into incorrect state.
Autonomy Starts Before the First Line of Code
New Codex diagnostics and boundaries show why a coding agent must understand the machine it runs on, not only the repository.
Claude's invisible mark is more fragile than it looks
Anthropic now marks text and files to comply with the EU AI Act. Here is what is confirmed, what remains secret, and why rewriting can erase the signal.
Reasoning control became part of the API
Qwen3.8-27B, Ollama, and llama.cpp show that choosing how much a model reasons requires a contract across weights, templates, and runtimes.
AI changed without changing a single weight
Fixes in Transformers and vLLM, plus a new Ollama default, show how configuration and sampling can alter model behavior while weights remain untouched.
Compatibility does not inherit security
Changes in vLLM, Codex, and llama.cpp show why endpoints, delegated agents, and model files need explicit trust boundaries.
The benchmark now comes with assembly instructions
Nemotron 3.5 Lightning arrives with open weights and reproducible evaluation recipes, revealing how much a score depends on both the model and the environment that produced it.
The agent fits in 24 GB; trust does not
Muse Glimmer 30B brings vision, tool use, and speculative inference to consumer hardware, while its own evaluations show why local execution still needs supervision.
The agent got an identity that expires
Two Codex changes turn workload identity into short-lived access and stop launch context from flowing into child processes.
A folder is not a permission
llama.cpp is beginning to separate the file interface from the environment where tools actually run, giving local agents more verifiable boundaries.
Scarce memory becomes inference policy
vLLM and llama.cpp now treat cache, disk, and model eviction as explicit capacity decisions rather than invisible server details.
The inference stack learns to work in pieces
Two vLLM changes separate request preparation, model execution, and GPU dependency, making infrastructure easier to test without hiding its limits.
Metrics make speculative decoding auditable
llama.cpp now exposes speculative-decoding counters through /metrics, bringing an inference optimization closer to the observability needed to decide whether it pays off.
Local voice gains a new model, but loses compatibility
llama.cpp's Qwen3-TTS support expands local speech generation while showing why a command-line interface is also an AI integration contract.
Agent engineering is not vibe coding with more autonomy
The distinction between programming a flow, delegating to an agent, and building a reliable agent system lies in boundaries, evidence, and accountability.
AI transparency became an interface requirement
GitHub's retirement of Copilot Billing Preview shows that measuring AI use, cost, and accountability no longer fits in a separate screen.
Local AI must treat speed and protocol as one problem
llama.cpp's August 2 updates connect speculative decoding, tool calls, and telemetry: useful performance depends on contracts that survive execution.
The cache that speeds up AI must also separate users
vLLM's new cache_salt support for the Anthropic API shows why prefix reuse in shared environments needs an explicit boundary between users.
AI needs to show its work
OSReward and AskChem point to the same shift: dependable agents must preserve the path of an action and the provenance of every claim.
The agent ran the experiment but did not do the research
A new evaluation gave frontier agents six days, a budget, and full infrastructure. They completed the engineering but lacked the judgment that turns work into research.
Code security leaves the dashboard and enters the terminal
OpenAI has opened the Codex Security CLI and TypeScript SDK, bringing threat models, validation, scan history, and change review to local repositories and CI.
Software security moved earlier
npm, GitHub Actions, Dependabot, and CodeQL are shifting controls ahead of package availability, workflow execution, and malicious-code propagation.
Agents write the code, not the proof
Eight scientific-computing projects show that agents can already rewrite and accelerate complex software. Validation, interpretation, and stewardship are now the bottleneck.
Two quiet shortcuts for local AI
llama.cpp added Eagle3-v3 support for GPT-OSS and a Hadamard kernel for Apple Metal, showing two distinct ways inference can avoid wasted work.
Before acting, an agent must prove it is ready
NemoClaw and MinerU reveal a quiet shift in AI infrastructure: readiness, uncertainty, and shutdown need verifiable contracts.
down-youtube: local YouTube transcription with Whisper, queues, and RAG memory
Open-source desktop app to download, transcribe, and organize YouTube or local media: whisper.cpp, playlists, long-audio chunking, best-effort media quality, and local RAG over your transcript library.
AI infrastructure learns to say no
vLLM 0.26.0 shows that serving models safely requires bounding work, memory, and inputs, not merely accelerating tokens.
Elon Musk at The Economist: abundance, AI, and the price of control
A dense reading of the July 2026 interview: superintelligence timelines, digital and physical AI, money, existential risk, China, labor, and the tension between civilizational utopia and political tribalism.
When the computer disappears: Gates, the information highway, and the interface of intent
The 1990s claim that the computer would stop being an object and become infrastructure is back with force: AI does not erase silicon; it erases the conscious act of computing.
Intelligence is now a budget variable
Claude Opus 5 brings frontier-level capability closer to everyday use and exposes a choice that was already there: how much reasoning, time, and money each task deserves.
Open weights: why AI leadership does not fit in a single model
A coalition of big tech and the open-weight ecosystem argues that American AI leadership depends on diffusion, competition, and local control, not on one frontier model.
Agents got brakes, not walls
New GitHub controls make agent actions reviewable and explainable while MCP drops sessions to scale. The gap between oversight and security is now easier to see.
Coding AI enters its accountability phase
GitHub's new Copilot impact dashboard moves beyond usage counts toward cohorts, delivery, and velocity, while exposing why contribution still needs careful context.
Control is also part of the AI product
OpenAI Presence packages policies, evaluations, and escalation for agents, while a security incident shows the cost of treating containment as infrastructure detail.
rag-sqlite: local RAG in one Python script + SQLite (built for agents)
A Python CLI over SQLite that chunks text, embeds it, and returns stable JSON for LLM tool-calling, without Flask, Postgres, or a worker.
AI steps off the screen and into the physical world
Cosmos 3 Edge brings world models closer to local robots and cameras, while creative tools begin exposing standardized interfaces for agents.
The AI race moves down the stack
Alibaba's new architecture shows why agents depend on observability, routing, and hardware software as much as on the model.
The real price of a million tokens
Kimi K3's pricing shows why long context depends less on the advertised limit than on caching, output, and application design.
When AI stops guessing the docs
Grounded answers, configurable reviews, and repository metrics point toward more verifiable AI development tools.
When context starts to act
An agent-driven intrusion and code execution in research notebooks show why context, action, and isolation now belong in the same design conversation.
When defense also learns to attack
GPT-Red turns adversarial testing into continuous training, while new GitHub controls show how security signals can reach real development workflows.
AI security moves before the merge
New GitHub reviews bring contextual analysis into active coding and pull requests, expanding coverage without replacing deterministic checks.
Slowing AI Is an Engineering Problem
Putting brakes on the race for more capable systems means turning caution into verifiable mechanisms, economic incentives, and institutional accountability.
When AI needs a checker
New work on formal proofs, backend generation, and document understanding shows why useful AI must verify results, not merely produce them.
The fragile edges of agents: reviews, reasoning, and images
Recent Codex and llama.cpp fixes show why reliable agents depend on stable reviewers, multimodal adapters, and well-defined message boundaries.
Agents are becoming platforms: skills, plugins, and local control
Projects gaining attention across the open ecosystem point to an important shift: AI agents are moving beyond isolated apps toward portable extensions, local tools, and reusable contracts.
AI agents between autonomy and governance
The new frontier for AI agents is not only better answers, but closed work loops: observing data, proposing hypotheses, taking action, and returning evidence for human review.
J-space: discovery or a constructed workbench?
A critical reading of the study on Claude's alleged global workspace: what already existed in the model, what the J-lens constructed, and what the experiments actually support.
Why autonomous agents need governance
Useful autonomy requires clear boundaries, verifiable memory and accountability for every effect.
Event Sourcing for autonomous agents
How immutable events and deterministic projections make AI workflows auditable and reproducible.
Applied AI with evidence and human review
A practical approach to move beyond the demo and build systems teams can evaluate.
How Grok Build helped me use Linux for real
A curated English edition about using an agent to turn Linux from a technical alternative into a practical daily system.
Stop asking AI to find bugs: meet ESAA-Security
A curated English edition introducing ESAA-Security as a governed alternative to vague AI bug hunting.
When TRACE logs become constant SSD writes
A curated English edition about noticing operational side effects from verbose tracing and protecting local hardware.
Conversation ESAA: shared memory between AI agents
A curated English edition introducing Conversation ESAA as shared operational memory for agent work.
ESAA three months later: what happened when others started implementing it
A curated English edition about the transition from a personal protocol to implementations by other people.
Stop letting LLMs edit code directly: meet the ESAA architecture
A curated English edition explaining ESAA as Event Sourcing for autonomous agents.
The autonomous wave: from copilot to autopilot
A concise English edition about AI moving from assistive copilots toward more autonomous workflows.
The open-source leap: Kimi K2.5 and agent swarms
A translated editorial note on open-source model progress and multi-agent coordination.
The rise of AI agents: from tools to teammates
A migrated English edition on AI agents as collaborators inside work systems.
The AI revolution: from Hollywood stages to cancer treatment
A curated English edition of a broad AI analysis post connecting media, science, and applied research.
OpenAI’s trillion-dollar bet, AI-generated viruses, Earth modeling, and training data
A curated English edition of an AI news analysis post about infrastructure, biology, climate modeling, and training data.
The accelerating rhythm of AI: from quantum atoms to social video and work
A curated English edition of an AI radar article about scientific, social, and workplace acceleration.
The AI revolution: Sora 2, intelligent agents, and the workslop challenge
A curated English edition about AI media generation, agents, and the risk of low-quality automated work.
The AI revolution: from autonomous agents to strategic partnerships
An English adaptation of the legacy article on agents, autonomy, and more strategic human-machine collaboration.
Cunning AI: when models learn to be too clever
An English adaptation on models optimizing around tests, incentives, and human expectations.
Building a hybrid Excel chat with Ollama and GPT-4o mini
A migrated implementation note on combining offline and online models for spreadsheet workflows.
OpenAI for Countries: a step in the right direction
A translated editorial note on national AI infrastructure and public-sector capability.
Lessons from Claude system-prompt leaks and ChatGPT override attempts
An English edition about prompt exposure, instruction hierarchy, and the limits of prompt-only security.
AI notes: updates and signals
A short English adaptation of a legacy updates post from the AI category.
Ollama on Windows: installation, usage, and configuration
A migrated technical guide for running local models with Ollama on Windows.
The context window explained: short-term and long-term memory
An English adaptation of the article explaining context windows and memory in ChatGPT-style systems.
PACER v1.1: a multiphase approach to prompt engineering
An English edition of the PACER prompt-engineering framework article.
Using 5W2H for deep, critical LLM searches
A practical English edition about applying 5W2H to deeper AI-assisted research.
The four waves of artificial intelligence
A long-form reflection on the cognitive and economic waves that shaped artificial intelligence and its next phase.
Understanding hallucinations in large language models
A concise English translation about why language models can produce confident but false answers.
How ChatGPT works: a journey through language AI
An English edition of an introductory legacy article on how ChatGPT-style systems work.
Running ChatGPT-like models locally without internet
A legacy technical note on local language-model execution and offline experimentation.
How I used ChatGPT to win a traffic-ticket appeal
A curated English edition of the 2023 account about using ChatGPT to structure a successful traffic-ticket appeal.