Hacker News Digest

Friday, June 19, 2026

In This Issue

Hacker News
I found 10k GitHub repositories distributing Trojan malware
Microsoft new Outlook takes 10 seconds to do what Outlook Classic does instantly
Emacs 31 Is Around the Corner: The Changes I'm Daily Driving
.gitignore Isn't the Only Way to Ignore Files in Git
A website that lists websites to submit your website to
I told them forced consent was unlawful. 5 years later it cost Elkjop €1.8M
GLM-5.2: The Most Powerful Open Model yet and the Brutal Reality of Running It
The AI Hate Progression
We built a persistent agent memory layer on Elasticsearch with 0.89 recall

Zipper Data Brief

June 19, 2026

Your daily digest of the best from Hacker News

824 points · theorchid · comments

GitHub is being flooded with trojanized repositories that impersonate legitimate projects and target users via SEO manipulation and dependency injection, while GitHub's moderation response is inconsistent and slow, leaving developers vulnerable to account theft and malware infection.

Microsoft new Outlook takes 10 seconds to do what Outlook Classic does instantly

679 points · Adam-Hincu · comments

# Summary Microsoft's new Outlook is significantly slower than its classic predecessor because it's built as a WebView2 browser wrapper rather than a native app, consuming 4x more RAM while delivering no functional benefits—a symptom of broader Microsoft bloat that has made even simple utilities like Calculator and Notepad slow to launch, driving users toward alternatives like Linux and Thunderbird.

Emacs 31 Is Around the Corner: The Changes I'm Daily Driving

444 points · frou_dh · comments

# Summary Emacs 31 is generating excitement among long-time users for practical improvements like built-in tree-sitter grammar installation, better xref editing, and native window transposition—features that reduce configuration burden and catch Emacs up to modern editors. The discussion reflects a devoted user base that values Emacs's configurability, especially its synergy with AI tools, though some note it still requires more setup than alternatives.

.gitignore Isn't the Only Way to Ignore Files in Git

439 points · FergusArgyll · comments

# Summary The discussion highlights lesser-known Git ignore mechanisms beyond `.gitignore`: `.git/info/exclude` for local-only ignores, `~/.config/git/ignore` for user-wide ignores, `.gitattributes` for hiding diffs, and `git update-index` for tracking modifications—with many developers expressing surprise at not knowing these options existed.

A website that lists websites to submit your website to

415 points · azeemkafridi · comments

# Summary The discussion highlights how submitting websites to aggregator lists is a nostalgic return to 90s internet practices (like Yahoo directories and web rings), with commenters noting that while such directories can drive traffic and backlinks, genuine community engagement matters more than drive-by submissions, and spam remains a persistent problem across these platforms.

I told them forced consent was unlawful. 5 years later it cost Elkjop €1.8M

410 points · speckx · comments

# Summary A person successfully challenged Elkjøp's forced consent requirement for marketing, resulting in a €1.8M fine from Norwegian authorities after a 5-year process. The discussion reflects broader frustration with companies illegally tying optional services (like marketing) to membership and the slow enforcement of privacy rights, with many users expressing appreciation for strong EU/Norwegian privacy laws while lamenting their absence in the US.

AI / Machine Learning

GLM-5.2: The Most Powerful Open Model yet and the Brutal Reality of Running It

41 points · ermantrout · comments

# Summary Commenters criticize the article as poorly researched and written (likely by an LLM), arguing that running massive models locally remains impractical despite being technically possible, while questioning why anyone would do so when cheaper cloud alternatives exist with better economies of scale.

The AI Hate Progression

113 points · gpi · comments

# Summary The discussion reveals deep polarization: critics argue AI was deployed unethically without consent and threatens creative work, while supporters counter that AI is genuinely useful, consent issues predate AI, and opposition amounts to uninformed nostalgia. Both sides largely talk past each other, with little common ground.

We built a persistent agent memory layer on Elasticsearch with 0.89 recall

109 points · showmypost · comments

The community is skeptical of using Elasticsearch for agent memory—suggesting simpler alternatives like SQLite or LanceDB are more appropriate—while acknowledging the article's techniques (hybrid search, reranking, decay scoring) are solid but questioning whether the added complexity actually improves agent performance and whether users notice the 11% miss rate.

Zen and the Art of Machine Learning Research

86 points · jxmorris12 · comments

# Summary The discussion explores how temperament, feedback frequency, and experimental intuition—rather than pure theory—determine success in ML research, while highlighting the tension between deep intellectual rigor and pragmatic incremental progress.

Generative AI Is Having Its Herbalife Moment

53 points · watermelon0 · comments

# Summary Commenters debate whether AI coding tools are genuinely problematic or unfairly criticized, with disagreement over whether the MLM comparison holds, whether vibe-coded software quality actually matters for business success, and whether concerns are legitimate critiques or gatekeeping by professional developers.

Startups / Business

The Korean telecom giant at the center of Anthropic's Mythos controversy

115 points · dstala · comments

# Summary The discussion centers on whether the White House's blocking of Anthropic's Mythos model over SK Telecom's involvement is genuinely about security concerns or political favoritism toward competing AI companies like OpenAI, with commenters debating whether the SK Telecom issue was a pretext and questioning the precedent this sets for foreign investment in U.S. AI companies.

Hello World: A new category of workplace for the next generation of builders

17 points · dedpool · comments

The discussion shows enthusiastic support for a new coworking space called "Hello World," with commenters expressing excitement about visiting and joining the community.

Launch HN: TesterArmy (YC P26) – Agents that test web and mobile apps

121 points · okwasniewski · comments

# Summary The discussion expresses skepticism about TesterArmy's value proposition, with commenters questioning whether LLM-powered E2E testing truly solves a real problem, noting concerns about nondeterminism, token costs, security risks of third-party SaaS, and whether this is genuinely novel compared to existing solutions and local LLM alternatives.

Peptides Fuel a Multibillion-Dollar Startup Race

8 points · Anon84 · comments

No comments available.

Ask HN: Conflicted about founding engineer role

6 points · gondolin1683 · comments

# Summary Most commenters advise against the startup offer due to low equity (3%), reduced salary ($160K vs $250K), high failure risk, and poor risk-reward tradeoff at this early career stage—unless the person prioritizes learning and adventure over financial security and stability.

Hacker News Digest

In This Issue

Zipper Data Brief

Top 6 Trending

AI / Machine Learning

Startups / Business

More Stories (34)