HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-07-17

Kimi K3: Open Frontier Intelligence (Kimi)

Kimi releases K3, a 2.8T-parameter open-source model rivaling proprietary frontiers.

Pricing and reasoning efficiency dominate; high per-token cost may not reflect real-world value.

1822 pts · 1069 comments

The lost joy of music piracy (pigeonsandplanes.com)

Nostalgic look at music piracy communities like Oink and What.CD and their loss to streaming.

Piracy is alive via Soulseek and RED; streaming is a flawed compromise.

797 pts · 564 comments

Microsoft Comic Chat is now open source (Microsoft Open Source Blog)

Microsoft open-sourced Comic Chat, its 1996 IRC client with comic-style chat.

Nostalgic but with a reminder that Comic Chat annoyed IRC users with its spammy protocol extensions.

739 pts · 159 comments

Sony deletes more movies from the accounts of people who ‘bought’ them (Techdirt)

Sony is deleting hundreds of movies and TV shows from PlayStation Store accounts due to lapsed licensing agreements.

Commenters call it theft and argue the misleading "buy" button justifies piracy in response.

665 pts · 411 comments

Decoy Font (mixfont.com)

A TTF font hides typed messages in plain sight using hybrid image spatial frequency illusions.

Mixed reception: cool art project, but many models can read it and it breaks accessibility.

619 pts · 142 comments

OnePlus halts operations in USA and Europe (OnePlus Community)

OnePlus will stop selling new phones in Europe and North America.

Many commenters saw the exit as inevitable after OnePlus lost its price edge and brand identity.

575 pts · 353 comments

How Our Rust-to-Zig Rewrite Is Going (rtfeldman.com)

Roc's compiler rewrite from Rust to Zig achieved feature parity after 487 days.

Two camps: fast builds vs. safety, with debate over whether compilers need unsafe code.

507 pts · 269 comments

Blatant AI slop just won a 25k USD DeepMind Kaggle Grand Prize (kaggle.com)

A $25k DeepMind Kaggle prize winner faces accusations of flawed methodology and misinterpreted data.

Critics argue the winner exemplifies AI slop, with flawed methodology and self-promotion overriding quality.

342 pts · 196 comments

NotebookLM is now Gemini Notebook (Google)

Google renamed NotebookLM to Gemini Notebook, adding code execution and cloud sync.

Skepticism dominates, with many predicting Google will eventually kill this rebranded product.

339 pts · 164 comments

$100 AI Music Video: Claude Fable 5 vs. GPT-5.6 Sol (TryAI)

An agentic harness gave Claude Fable 5 and GPT-5.6 Sol $25-$100 to autonomously produce music videos.

Most commenters found the AI videos unwatchable, citing overly literal lyric interpretation and poor quality.

322 pts · 434 comments

SpaceX stock erases all its gains and slides below IPO price in intraday trading (Los Angeles Times)

SpaceX stock fell below its $135 IPO price on Wednesday, erasing all gains since listing.

Historical data shows most IPOs underperform the market, with SpaceX's small free float amplifying volatility.

300 pts · 265 comments

At least 105 past YC founders have worked at OpenAI and Anthropic (joinedanthropic.com)

Site profiles 105 YC founders who later worked at OpenAI or Anthropic.

Skepticism over the stat's significance given YC's ~13,000 total founders; many attribute hiring to Altman's network.

299 pts · 216 comments

LM Studio Bionic: the AI agent for open models (LM Studio)

LM Studio launches Bionic, an AI agent for coding and document work with local and cloud open models.

Skeptics question Bionic's differentiation from open-source agents and note LM Studio is closed-source, despite claiming open-model support.

288 pts · 105 comments

The human-in-the-loop is tired (Pydantic)

LLM-assisted programming boosts productivity but creates a new fatigue of supervision and review.

Suspicion the article is AI-generated splits the thread; practical advice advocates treating LLMs as assistants, not agents.

274 pts · 166 comments

Ente – Opening Our Books (ente)

Ente made its business metrics public, including revenue, customers, and accounts registered.

Mixed reception: praised for transparency but criticized for showing revenue alone without operating costs.

266 pts · 102 comments

The LLM Critics Are Right. I Use LLMs Anyway (Jeremy Theocharis)

A software engineer admits LLM critics are right but still spends $10k/month on tokens.

Two camps: skill atrophy vs. productivity gains, with regex as a cautionary example.

262 pts · 275 comments

Immersive Linear Algebra Book with Interactive Figures (2015) (immersivemath.com)

An interactive online book teaching linear algebra with embedded visual figures.

Mostly enthusiastic praise, with a split on whether interactive books sacrifice mathematical rigor.

256 pts · 28 comments

1,300 Beautiful Wildlife Illustrations from the 19th Century Now Restored (Open Culture)

Article body wasn't reachable. HN discussion still summarized.

AI restoration splits commenters between purists and pragmatists.

242 pts · 47 comments

If you want to create a button from scratch, you must first create the universe (madcampos.dev)

Recreating a native HTML button from scratch requires implementing dozens of behaviors.

Two camps: those who agree native elements should be used, and those who say missing native components force custom implementations.

239 pts · 124 comments

Detecting LLM-Generated Texts with “Classical” Machine Learning (Lyc8503's blog)

Traditional ML (SVM+TF-IDF) detects LLM-generated web fiction with up to 85% sentence-level accuracy.

Two camps: statistical patterns are real and detectable vs. adversarial methods will always break classifiers.

218 pts · 160 comments