Source, hn_front
15 cards
Brief
A Hacker News front-page project logs a Claude Opus 5.5 agent that wakes on a heartbeat and keeps self-written notes. It is a creative log with no product release.
Brief
The ELC-based architecture claims to match a current LLM on blind reasoning tests at about 0.25 seconds per question. These are the author's claims.
Hype watch
The free MIT-licensed macOS browser runs automation in isolated Spaces. The 2.5x speedup on complex tasks is a vendor claim, and Windows and Linux are not yet available.
Brief
A Show HN experiment found text-based judges could not tell a fabricated agent autobiography from a real 30-day memory. Only the logs can settle it.
Brief
A solo-author post reports a 64% win rate with free talk versus 42% without, against a max-damage bot. The code is on GitHub.
Brief
A personal essay tying rubber duck debugging to a meta-study of 6,000 participants with moderate benefit. It is not a software release.
Brief
A research preview for US Pro and Max subscribers via waitlist, powered by Opus 5.5. Team and Enterprise plans are excluded.
Brief
Background reading arguing that AI use accelerates decoherence and limits runaway self-improvement. It has no release or tooling change.
Hype watch
Public beta v0.6.0 with security claims that are unverified until audited. Its ZK-SNARK proving is still research.
Brief
The vendor reports equal accuracy on five questions, with 92% fewer input tokens and 66% fewer model calls. The numbers are unverified vendor claims.
Brief
A GitHub-hosted series with 16 chapters that grows a 300-line Python toy harness from first principles.
Brief
The arXiv paper from 13 Oct covers 11 benchmarks. Code and data are promised but not yet released.
Brief
Lets AI agents read and search WhatsApp messages, but not send them. It needs a Workers Paid plan and uses Baileys, which violates WhatsApp terms.
Brief
The framework refines prompts, tools and memory with model weights fixed. It reports gains up to 11.5 points on WorkBuddy-Bench and needs Linux and Python 3.12.
Act on this
Use claude --worktree --tmux, and pass --tmux=classic in iTerm2.
