HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-10-26

A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis (lynchlab.net)

A 10-year-old CNN beat GPT-6 Astra and Claude Fable 5.1 at diagnosing skin cancer from images.

Most commenters saw the result as expected: specialized models beat general ones on narrow tasks.

614 pts · 259 comments

Dolly Parton's Imagination Library (imaginationlibrary.com)

Dolly Parton's Imagination Library mails free books monthly to children from birth to age five.

Overwhelming praise for the program, with details on its decentralized local-funding model.

596 pts · 99 comments

I refuse to accept a censored Internet. Join Me at FOSDEM 27 (talkweb.eu)

A FOSDEM 2027 devroom aims to build an end-to-end toolchain for resisting censorship.

Debate over whether technical tools can meaningfully resist censorship given state and corporate control of infrastructure.

519 pts · 257 comments

Meta Ignores Brazilian Court Order to End Bolsonaro Favoritism (Panamerican Dispatch)

Meta ignores Brazilian court order to stop favoring Bolsonaro over Lula on its platforms.

Commenters split: some see deliberate algorithmic bias, others say stochastic systems are misinterpreted, with many calling for European alternatives.

489 pts · 334 comments

Oregon Police Are Charging a Local News Outlet $2.5M for Records About One Cop (The Intercept)

Oregon police charged a local news outlet $2.5 million for records about one officer.

Commenters overwhelmingly condemned the $2.5 million fee as an intentional gag order.

486 pts · 133 comments

Lula wins election after final votes tallied (reuters.com)

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

Deep divisions over electoral fairness and court bias dominate the thread.

451 pts · 504 comments

OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why? (Max Woolf)

OpenAI and Anthropic slashed LLM prices up to 80% due to open-weight competition and commoditization.

Competition from open-weight models, not loss-leading, drives the price cuts, many say.

379 pts · 381 comments

Anthropic S-1 (sec.gov)

Anthropic files S-1 for IPO with novel governance structure including LTBT and Class T/F/E/N stock.

Skepticism over $81B loss and novel governance structure dominates, with some calling it a consequential IPO.

359 pts · 297 comments

IBM CEO says open weight models are now good and cheap on-prem (theregister)

IBM CEO says open weight models now deliver 90% of performance at 10% cost.

Two camps: those dismissing Krishna as talking his book, and those running real on-prem setups that validate his claim.

354 pts · 256 comments

Serbia's Vucic concedes election defeat in historic win for student movement (reuters.com)

Serbian President Vucic concedes election defeat after a student-led movement's historic win.

Debate over whether Vucic's regime was truly oppressive, with comparisons to other autocracies.

311 pts · 168 comments

Show HN: TitleArt – Make text like it's 1997 (TitleArt)

A faithful recreation of 1990s WordArt for modern browsers, built from screenshots.

Strong nostalgia and praise; creator added CJK support after a request.

309 pts · 54 comments

Switzerland. Block by block. At 1:1 scale (Blocks)

A 1:1 scale Minecraft map of Switzerland built from open government geodata, available for 15 CHF.

Mixed reactions: impressed by the data use, but critical of the LLM-written copy and charging for automated conversion of open data.

306 pts · 137 comments

Samsung running a TensorFlow lite model on 960x540 image to track you (X (formerly Twitter))

Samsung TV runs AlexNet on 960x540 images every few seconds to identify games.

Comments were moved to a different Hacker News thread.

292 pts · 1 comments

Why I Am a Programmer (alfie.wtf)

A nostalgic origin story traces how 90s demoscene culture inspired a lifelong programming career.

Nostalgic origin stories dominate, with a split over whether AI makes programming jobs obsolete or will be walked back.

276 pts · 86 comments

AI use leads UC Berkeley math professor to see 'no point' in grading homework (Daily Cal | Berkeley news)

UC Berkeley math professor sees homework grading as pointless due to AI.

Split between in-class exam advocates and those questioning math's relevance.

226 pts · 320 comments

JetBrains Thinkrail (GitHub)

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

Skepticism over AI-generated docs, debate on agent isolation vs local runs, and jj vs git worktrees.

223 pts · 165 comments

The State of Context Engineering (Drew Breunig)

Long context windows introduce failure modes that require careful context engineering.

Users confirm context degradation worsens past ~200k tokens, with tips on compaction and handoff files.

219 pts · 56 comments

Show HN: A Truly Modular Keyboard (GitHub)

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

Practical skepticism over cost, rigidity, and I2C scaling dominates the thread.

213 pts · 69 comments

fheroes2: A recreation of Heroes of Might and Magic II game engine (GitHub)

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

Nostalgia for HOMM2/3 dominates, with links to modern engines and LLM gameplay experiments.

211 pts · 73 comments

Moderna and Merck say their mRNA cancer vaccine cut death, recurrence by 42% (Ars Technica)

Moderna and Merck's mRNA cancer vaccine cut melanoma death or recurrence by 42% in a Phase 3 trial.

Broadly positive view of trial results, with a notable correction about blaming cancer meds for a separate death.

199 pts · 49 comments