Turn Hacker News into a personalized brief

Open-weight GLM 5.2 reportedly outperforms Claude Opus 4.8 on vuln-finding benchmarks and is cheap enough for local or provider-based use.

🗣️ HN commenters add:

• “Claude Code” is just an agent harness; $/vuln metric is called meaningless without baselines
• Methodology and randomness disputed; GLM-5.2 (~753B params, see HF card) seen as strong, export-control risk noted

Age checks in US/EU/AU sites log IDs under “protect kids,” enabling future linkage of online speech to real identities.

🗣️ HN commenters add:

• AV data can let states or platforms weaponize “everything you say” against you
• “Protect minors” seen as pretext for speech control and real-person/anti-AI verification

ATS scoring swings widely; high-volume funnels (~100+ applicants/hour) plus auto-ranking make interviews largely random.

🗣️ HN commenters add:

• Human screeners can’t keep up; ATS flow-rate decides who’s seen
• LLM outputs aren’t truly deterministic even at temp=0; ranking remains noisy

EU leaders push “chat control” mass scanning in closed-door talks; EP president Metsola allegedly revives rejected plan while “2.0” mandate is negotiated. See fightchatcontrol.eu.

🗣️ HN commenters add:

• EP earlier rejected chat control; Metsola now seen as bypassing that decision
• Critics say child-safety pretext masks broad surveillance and privacy erosion

CEO Callum Negus-Fancey and CTO Luke Wright allegedly used bogus DMCA claims to hide reporting on Pollen’s collapse; Google delisted it from search.

🗣️ HN commenters add:

• Abuse of “under penalty of perjury” DMCA with no real penalties; contesting requires lawyers and money
• Streisand effect: takedown attempt pushed story to HN front page; background on CEO in Independent profile

LLMs can comment on single MRI slices but lack full 3D context; radiologists warn this risks missed findings and “mirage reasoning” hallucinations.

🗣️ HN commenters add:

• MRI is inherently 3D; 2D screenshots are insufficient for safe diagnosis.
• Modality limits matter (e.g., ultrasound vs MRI); AI may overconfidently ignore such gaps. Related: Futurism, Nature, NYT.

Brown econ prof alleges “overwhelming” AI cheating on new take-home, closed-book exams; fears LLMs erode assessment. See htmx essay.

🗣️ HN commenters add:

• Take-home cheating long predates AI; LLMs just scale it.
• Core problem is incentives: degrees as job signals, honor codes collapsing.

Open-source tool reimplements Apple’s AirPods protocol so Linux/Android get noise control, ear detection, and precise battery, no Apple device needed.

🗣️ HN commenters add:

• Android 17 removes need for root; earlier Android needed rooting for full features
• Prior work and context in HN thread and reverse-engineering repo ida-pro-mcp

5k NYPL menus (1880–1920) digitized and hand-labeled, now explorable via interactive viz to compare dishes, prep styles, and historical prices.

🗣️ HN commenters add:

• NYPL dataset is manually verified at menus.nypl.org
• New pudding.cool UI is insightful but buggy on mobile and hard to deep-link

DRAM, HBM, NAND prices over 1960–2026, shown as $/GB on a log scale, without inflation adjustment.

🗣️ HN commenters add:

• $/GB pre-1990 is misleading; nobody bought GBs of RAM then
• Better metric is $/RAM for typical tasks; volatility tied to crypto, AI, Moore’s law slowdown

Daily word grid where valid words shrink the board; definitions shown; bonus words optional; min 4 letters, singular/plural allowed. Similar to Puzzmo Ribbit.

🗣️ HN commenters add:

• Very close to Puzzmo’s Ribbit; see zanagrams.com/puzzles/5.json, puzzmo.com
• Suggest scoring on bonus-word ratio, not just time, to deter brute-force tries

Bill 948 would fine MI employers or force overtime pay if workers must answer job messages off-hours; scope may include exempt staff.

🗣️ HN commenters add:

• Targets “uncontracted on-call” duties, not casual shift swaps in group chats
• Split between seeing it as basic labor protection vs red tape that prompts workarounds or job moves

Springer Nature pulled two 1940s Planck papers as “copyright violations”; one likely flagged only for duplicate title, leaving blank pages and empty PDFs.

🗣️ HN commenters add:

• Retraction tied to alleged self-plagiarism across two journals
• Second removal may be an automated error; prior thread: HN discussion

No article summary provided; only HN discussion available.

Forced “AI or fired” mandates drove tokenmaxxing; now firms cut spend after using metrics/leaderboards to rank or fire staff.

🗣️ HN commenters add:

• 2025 “AI or fired” rules pushed artificial token usage, then budgets snapped back.
• Some say token metrics were rare; others cite leaderboard anxiety and firings, e.g. Dan Luu.

Kyoto foresters prune cryptomeria (“sugi”) into living platforms that sprout many straight, knotless poles, yielding dense, flexible roof timber from minimal land.

🗣️ HN commenters add:

• Notes it’s intensive pollarding/coppicing of sugi, not generic “Japanese cedar”
• Calls out rehashed 2020 content, broken sources, dubious “140% flexible/200% stronger” claims

Central banks’ 2026 BIS report flags AI-driven asset bubbles and systemic risk from overhyped automation and rent-extraction assumptions.

🗣️ HN commenters add:

• Link to BIS reports: BIS annual economic reports
• Bubble thesis: if “total automation” fails, valuations crash and labor markets take the hit

US import rules (25-year wait, costly certification) and FCC-style bans block Chinese EVs like Xiaomi’s SU7 despite strong reviews and US exec praise.

🗣️ HN commenters add:

• US has long labeled foreign tech “weapons” to justify control; some want open firmware and local control for Chinese cars.
• Others distrust US gov too; AI export/buy bans may hurt US firms’ competitiveness and security.

CA budget deal funds uploading all DL/ID data to SPEXS, a private AAMVA-run national DB; DHS pressure via REAL ID; “guardrails” seen as weak. More

🗣️ HN commenters add:

• S2S/pointer use is starting now; noncompliant IDs may be forced in later
• Tradeoff: cross-state fraud checks vs national ID, privacy, immigration tracking fears

Ford admits AI alone didn’t ensure quality; rehires veteran engineers after “mistakenly” assuming AI would yield high-quality products. HN discussion missing.