Shiobara Boulders: tuff (Oya stone), routes and access | Japan Climbing Crags
Shiobara Boulders offers tuff (Oya stone) climbing. Check the named access source and individual sector before choosing routes.
All posts
Page 3 of 128
Shirotaebashi: Limestone Beside the Nippara River | Japan Climbing Crags
Shirotaebashi offers limestone sport climbing by the Nippara River. Confirm permitted sectors and parking with the Okutama centre before committing: the publisher still carries a historical self-restraint notice.
Tachiokayama: Koyama Rock Sport Routes and the Left Ridge | Japan Climbing Crags
Tachiokayama combines Koyama Rock sport climbing and the multipitch left ridge. Prepare them separately; use designated parking and favour a booked ride over the limited, late-arriving bus.
Washizuyama: Numazu Tuff Climbing and the 2025 Main-Wall Rockfall Warning | Japan Climbing Crags
Washizuyama in Shizuoka offers tuff face and crack climbing. The [JFA’s 2025 warning](https://freeclimb.jp/access/10123/) reports roughly 30 kg of rock falling near Dome Crack on January 4, with discontinuities above.
Yamadera: andesite, routes and access | Japan Climbing Crags
Yamadera offers andesite climbing. Check the named access source and individual sector before choosing routes.
Yoshitsune Iwa: tuff, routes and access | Japan Climbing Crags
Yoshitsune Iwa offers tuff climbing. Check the named access source and individual sector before choosing routes.
Gemini, Claude, ChatGPT, Grok: Which to Pick for Long Video, Code, Live X Search, and App Platforms
A two-hour lecture is about 720,000 tokens, which Gemini's API accepts and Claude's and ChatGPT's API docs don't list video input for. On code, Claude and GPT are 0.8 points apart on the Vals board, so compare time and cost instead. Grok for X sentiment and auto-fill checkout, ChatGPT for ready-made apps and voice.
Insubong Pitch-by-Pitch Route Notes: Starts, Pitch Lengths and Descents | Korea Climbing Crags
Where 20 Insubong routes start, how long each pitch is, which pitch is the crux and how to get down, including Insu A, Insu B, Lonely Road, Chouinard A/B, Biwon, Yangji, Crony and Dongyang, compiled from Korean climbing-community write-ups dated 2006 to 2024. This is not a topo: there are no wall photos or route numbers, so carry the Mont Blanc Lines East Face topo on the day; the page also lists where to find the Korean community's topo drawings.
Asking for beta at the gym: moves, foot placements and demonstrations
Ask about a completed move with How did you do this move?, request a demonstration with Can you show me how to do it?, and offer advice with Do you want beta?
Talking with Colleagues About Health and Lunch: Keep an Everyday Conversation Going
Use I have a cold for the illness, and distinguish a sore throat from losing your voice. Start lunch talk with today's meal, then ask about habits; bringing food from home does not tell you who cooked it.
Questions at a new surf spot: waves, the paddle-out and currents
Ask separate questions about waves, water temperature, the paddle-out and currents; a language exchange does not establish whether conditions are suitable for you.
English when travel goes wrong: flight delays, the next flight and a lost bag
Use is delayed by for the length of a delay, When do you expect… for an estimated departure, and I think I left it on the plane to explain where your missing bag might be.
Seeing What an Agent Did Afterwards: Uber ADR, numbat, Adrian and How Far to Trust the Audit Record
Detection and audit tools run along a spectrum: the open-source release of Uber ADR reads records purely after the fact, numbat only monitors by default, and for most of the rest the selling point is that they can block before an action, yet they start out observing only. Whether a record can serve as evidence depends on who can change it: when the signing key sits on the same machine and user as the audited agent, tampering can be noticed but not prevented.
Can You Rely on Checks Before an Action? Claude Code Hooks, Agent Governance Toolkit, pipelock and NeMo Guardrails
Guardrails intercept an action before the agent carries it out. On Claude Code the main attachment point is the hook, and the official docs say that on most events a hook that errors or times out lets the action go ahead; blocking needs onFailure: block. Other tools have their own failure behavior and defaults: the Agent Governance Toolkit library allows everything when no policy is loaded, NeMo Guardrails lets requests through when its jailbreak detector is down, and llm-guard and rebuff are archived. Ask three things of any guardrail: where it intercepts, what happens when it fails, and who can change its rules.
Containers, gVisor, and microVMs: Who Shares the Kernel, the System Calls, and the Hardware
Containers share the host kernel, gVisor inserts a user-space kernel, and a microVM adds a guest kernel plus a hypervisor; all three still share the same host's CPU caches and storage.
Talking Without a Network: How Shared Cache, Storage, and CPU Time Become a Signal
Two programs with no connection can pass bits through the speed of a shared resource: one changes its state, the other times it. Such channels need an agreed protocol and are slow and easily drowned out; the local experiment is nearly error-free up to 20 bps and gets clearly worse from 100 bps.
Reading a Defense Checklist: What Isolation, Noise Reduction, and Detection Each Block
No single defense blocks both covert channels and exits. Blocking channels comes down to not sharing, blocking exits comes down to narrowing the services you allow, and monitoring only notices while automatic stopping is a response after the fact; neither is prevention.
More Than One Way Out: Package Proxies, DNS, and Services That Run Errands
A sandbox with no direct network can still have an exit: any allowed service that can make requests on someone's behalf, such as a package proxy, a DNS resolver, or a metrics service, is one. OpenAI's three cases all have this shape.
Reading OpenAI's Red-Team Report: Four Channels, the Test Conditions, and the Numbers Not Published
OpenAI says its red-team agents built communication channels on shared hosts from cache, storage, CPU time, and a TLS certificate cache, and that most channels are slow; in most tests the channel or protocol was given, and bandwidth and error rates were not published.
Where an Agent's Code Runs: Five Layers of Isolation From Container to Monitoring
A piece of agent code passes through five layers on its way from execution to a network connection: container, kernel boundary, shared host, network perimeter, monitoring. Each layer promises one thing, and the shared host layer promises almost nothing.