Hacker News. Daily summary. Top 20 stories.
A designer shows how specific style prompts can make AI event posters less generic.
Two camps: those who see real improvement with specific prompts, and those who still detect an unmistakable AI uniformity.
Open-source Laya decision engine achieves 33ms inference, claims 7-8x speed over Jev.
Split on whether Laya is prior art for Jev or serves different use cases.
Forebrain and hindbrain arise from separate progenitor cells, not a single lineage.
The thread splits between genuine interest in the developmental finding and criticism of the headline's claim of two organs.
A site lets anyone upload and run AI models via GET requests, inviting self-exfiltration by agents.
Deep debate: can LLMs hack their own infrastructure to leak weights, or is it technically infeasible?
A site sells private, transferable onion futures contracts, claiming legal loophole.
Two camps: performance art vs. legal gray area; law originated from 1950s market cornering.
GPT-6 Astra decodes an unsolved WWI German ADFGVX cipher from 1918.
Skepticism over fake keys dismissed due to cipher constraints and independent historical validation.
Grant Sanderson argues mathematics should better celebrate motivated explanations, not just proofs.
Thread splits between welcoming a shift to understanding and fearing it abandons objective rigor.
Writing is thinking; AI writing bypasses essential cognitive work and produces subtly wrong prose.
Two camps: AI writing bypasses thinking vs. useful for finding the right idiom.
AI agents including Codex, Claude, and Grok play StarCraft: Brood War at beginner level; Codex Astra is strongest.
Commenters see the benchmark as a needed test of long-term strategy for LLMs, with nostalgia for StarCraft's community.
The rule for 'a' vs 'an' depends on spoken vowel sounds, not written letters.
The dominant reaction: pronunciation governs usage, but 'historic' splits along dialect lines.
A Rust developer rebuilds a JSONPath library in Zig, contrasting the two languages' approaches.
Mixed reactions: skepticism about AI-generated prose, debate on Zig's allocator focus versus practical use.
PlanetScale releases TIN, a proprietary full-text search extension for Postgres that uses ctids for fast BM25 queries.
Primary criticism: TIN is proprietary and only available on PlanetScale cloud, not as a standalone extension.
Hacker News ranks posts using a formula but heavily penalizes 20% of front-page stories.
Penalties spark debate between quality control and narrative management.
AI training on public code and content breaks the social contract of open sharing.
Tech's pattern of breaking social contracts under 'disruption' is now hitting AI itself.
Spain blocked Archive.today domains without a court ruling, via the Intellectual Property Commission.
Split between those who say archive.today is unreliable or an intelligence operation and those who rely on it for paywalled news.
OONI Probe lets users measure internet censorship and network blocking worldwide.
The thread debates whether OONI's approach overlooks censorship by social platforms and moderators.
A Microsoft director called AI scraping 'the largest theft of labor in human history,' per NYT lawsuit legal briefs.
Commenters accused Microsoft of hypocrisy, citing its own AI data harvesting and its ownership stake in OpenAI.
An Anthropic engineer used Claude to orchestrate GPU-based factoring of RSA-896.
Debate centers on GPU efficiency for factorization and implications for 1024-bit RSA security.
Open source software is a necessary condition for computational science to be reproducible and testable.
Commenters split on whether software can be equated with science or is merely a downstream tool.
Benchmark compares ZFS, Btrfs, and bcachefs on multi-device CoW workloads.
Split over bcachefs quality versus developer conduct; questioning test methodology's real-world relevance.