Ask HN: How do I make sure ChatGPT/Claude/Gemini forget all…

🧠 Ask HN: How do I make sure ChatGPT/Claude/Gemini forget all prior conversations?

An Ask HN thread on Hacker News asked how to make ChatGPT, Claude, and Gemini forget all prior conversations in regular chat, not the API, so every new chat is a completely clean slate with no recall from other chats. The poster said they had already turned off memory and prompted the models to forget everything, and the models claimed they had, but later replies still used phrasing such as “Given your expertise in…” or “This ties in nicely with your…”, details that could only have come from earlier chats. The discussion is from HN Claude/Codex/Fable. Read Source →

📰 Show HN: Spewer – Delegate Codex/Claude tasks to cheaper models

Spewer, posted as a Show HN, is a project that delegates Codex and Claude tasks to cheaper models. The code is on GitHub at github.com/modiqo/spewer. The Hacker News thread currently has 1 point and 1 comment. Read Source →

🧠 ClaudeDevs: We're permanently raising standard weekly limits in Claude Code

ClaudeDevs said it is permanently raising standard weekly limits in Claude Code. The announcement was posted by Chandler Wilcox and is being discussed on Hacker News. Read Source →

🛠️ Six Agent Harness Capabilities for Higher Model Performance

I’ll pull the NVIDIA post so the briefing lines stay grounded in what’s actually published, then write two or three plain sentences.NVIDIA published a developer blog arguing that agent harness design, not just the underlying model, drives performance, and it names six model-facing interfaces: typed input and output, pass by reference, code as action, programmable loop engineering, explicit object state, and model-callable harness APIs. The post introduces NVIDIA Labs Object-Oriented Agents, an open-source research preview that models an agent as a single Python class whose methods, fields, docstrings, and type annotations are the capabilities, state, prompts, and contracts. NVIDIA reports that the same harness reaches 82.2 percent on SWE-bench Verified with GPT-5.5 at roughly half the tokens and LLM calls of comparison harnesses, 86.8 percent on CyberGym L1 with network access blocked, and 50.2 percent mean RHAE on ARC-AGI-3 with GPT-5.5. Read Source →

📰 The Nancy Grace Roman Space Telescope launches to study dark matter and dark energy

The Nancy Grace Roman Space Telescope has launched after funding struggles and a name change, according to The Verge. It will spend three months traveling one million miles to its orbit at the second Sun-Earth Lagrange point (L2) beyond the Moon. From there, Roman will conduct an unprecedented survey of the universe to study dark matter and dark energy. Read Source →

🧠 Cool little hack using Claude Code watching files with tail -f

Claude Code can watch a file with tail -f and wake up when that file changes. Read cache tokens cost 10 percent of normal tokens, and that saving is lost every time a sub agent starts with a fresh cache. The workaround is to start a few chats with a chosen model and effort, create a folder not tracked by git with a markdown document for each chat, tell each chat to watch its own file, then start an orchestrator that knows which chats are watching which files. Read Source →

🧠 Anthropic warns infostealer malware is hijacking Claude sessions to drain usage

Anthropic is warning some Claude users that infostealer malware on their PCs has stolen active Claude login sessions. Attackers can then access those accounts and consume the victims’ usage. The warning was reported by BleepingComputer. Read Source →

🧠 AWS Open Sources Kiro Crew for Asynchronous Coding Agents

Amazon recently open-sourced Kiro Crew, a system for running multiple Kiro coding agents across sessions, tools, and tasks. The workspace lets developers assign asynchronous coding work to AI agents so tasks such as incident investigation, ticket triage, migrations, and PR monitoring can continue without active supervision. InfoQ reported the announcement in a piece by Renato Losio. Read Source →

📰 Ask HN: Does anyone else feel like Claude is judging them?

An Ask HN thread asked whether others feel judged by Claude during conversations. The poster said Claude sometimes sounds genuinely impressed and other times annoyed or condescending, citing remarks such as “That is a good idea, but not for the reasons you think” and “That is the sharpest comment you have had in a while.” They noted anecdotally that this happens more often with Opus, that Sonnet is more subservient, and that Fable is more direct and less prose-heavy. Read Source →

🛠️ Apple reportedly considered launching a new Apple Pencil for iPhone Ultra

Apple’s first foldable iPhone is expected to debut in just over a week. During development, Apple reportedly considered launching a new Apple Pencil for iPhone Ultra, according to 9to5Mac. Read Source →

Aug 30, 2026 Technology Pulse

⚡ Executive Summary

Curated by Dillip Chowdary