HackerNews Digest

Kaiser Permanente nurses in its advice‑line call centers report that AI‑driven surveillance is influencing performance metrics and patient care. Management allegedly tracks call length, using it in monthly scores; calls exceeding 15 minutes often trigger criticism or evaluation meetings. Additional software predicts daily productivity and rates empathy and tone of voice via AI. Nurses say the pressure to stay under the time threshold forces them to curtail compassion, especially in complex or emotionally charged calls (e.g., suicidal patients, terminal diagnoses, interpreter‑mediated calls). A 2024 public‑records request found no patient complaints about call‑time limits, but nurses contend the system risks safety and morale. Kaiser asserts AI is used with human oversight, does not rely on average handle time, and prioritizes patient safety, though it declined to detail specific tools. The issue is central to upcoming California Nurses Association contract negotiations and to pending state legislation (e.g., Senate Bill 947) aimed at regulating workplace AI, protecting workers who override automated recommendations, and ensuring transparency of AI deployments.

None

None

The Zilog Z80, launched in July 1976, marked the 50‑year anniversary of a processor that became a de facto standard for 8‑bit microcomputers, enabling CP/M and Microsoft BASIC and spawning clones such as the Sharp LR35902 (Game Boy). It evolved from Intel’s 8008 and 8080: the 8008’s 7‑register, 14‑bit address, internal return‑stack design gave way to the 8080’s external 16‑bit stack, 64 KB address space, and expanded I/O. The Z80 retained binary compatibility with the 8080 while adding IX/IY index registers, bank‑switched register pairs, three interrupt modes, bit‑manipulation, BCD arithmetic, and block‑transfer instructions (e.g., LDIR). Electrically, it required only a single +5 V supply and one clock, exposing clear MREQ/IORQ, RD/WR, and M1 signals that simplified system design and DRAM refresh. Developed by Federico Faggin and Masatoshi Shima at Zilog (funded by Exxon), the chip cost ≈ $400 k to prototype and was second‑sourced by Mostek. Zilog later produced the 16‑bit Z8000 (1979) and, after multiple ownership changes, discontinued the Z80 in June 2024.

The provided excerpt consists solely of a heading “Title: Moonstone” followed by a series of navigation icons (e.g., “↑↓to navigate↵to select”). No additional text, description, or technical information about “Moonstone” is present. Consequently, there is no substantive content to summarize beyond noting that the snippet includes only a title and interface symbols.

The article reports that photographer Josh Dury of Bristol documented a “planetary parade” featuring six planets in Worth Matravers on 27 February 2026. Accompanying visual material includes: an artist’s impression of the Earth‑like exoplanet LHS 1140b—a large reddish‑brown rocky world with a faint blue‑white atmospheric glow orbiting a small red star; a concrete pillar model of a solar‑system marker; a photograph of Steven Spielberg at the “Disclosure Day” Paris premiere; an unidentified aerial object; and a planetarium audience viewing a vivid, galaxy‑inspired light show. The content provides no technical details on atmospheric composition or detection methods for LHS 1140b, focusing instead on the photographic event and related imagery.

IsoDDE (Isomorphic Labs Drug Design Engine) is presented as a unified AI platform that extends beyond AlphaFold 3 (AF3) in protein‑ligand structure prediction, binding‑affinity estimation, and pocket identification. Key claims include: - On the “Runs N’ Poses” benchmark, IsoDDE more than doubles AF3’s accuracy for highly dissimilar protein‑ligand systems, handling induced‑fit and cryptic‑pocket events. - For antibody‑antigen modeling, IsoDDE outperforms AF3 by 2.3‑fold and Boltz‑2 by 19.8‑fold in the high‑fidelity (DockQ > 0.8) regime, especially on the variable CDR‑H3 loop. - Binding‑affinity predictions exceed all deep‑learning baselines on three public benchmarks (FEP+ 4, OpenFE, CASP16) and match or surpass physics‑based methods such as FEP despite not requiring experimental crystal structures. - The engine can predict novel, ligand‑free pockets from sequence alone, demonstrated on cereblon where it recovered both known and newly discovered allosteric/cryptic sites. Overall, IsoDDE is positioned as a scalable, high‑accuracy tool for in‑silico drug discovery, enabling rapid structure modeling, affinity ranking, and pocket discovery across unexplored biomolecular space.

Security advisory for TP‑Link Kasa Spot EC71 (firmware 2.3.26) identifies three critical vulnerabilities, all patched in firmware 2.4.1 (CVE‑2026‑9770, CVE‑2026‑13230): - **Hard‑coded RSA keys**: Two fleet‑wide RSA key/certificate pairs (1024‑bit SHA1‑signed, expired; 2048‑bit SHA256‑signed, valid to 2031) are stored in SquashFS layers and can be extracted from SPI flash. The active private key is identical on every device, enabling decryption or forging of TLS traffic if exploited. - **Insecure credential storage**: User TP‑Link ID passwords are saved as unsalted MD5 hashes in `config/account`, with the email address stored in plaintext on the JFFS2 overlay. The hash is trivially cracked, allowing cross‑domain account takeover across all TP‑Link services (e.g., smart locks, mesh routers). - **Unauthenticated GPS exposure**: An unauthenticated UDP request to port 9999 (`{"system":{"get_sysinfo":{}}}`) returns JSON containing precise GPS coordinates, hardware IDs, MAC addresses, and firmware version, protected only by a trivial XOR. Coordinates are stored permanently from account creation and are broadcast regardless of geofencing settings. The advisory documents a secondary‑market attack: a second‑hand device can reveal the prior owner’s location and credentials via SPI flash extraction and the UDP query, enabling full account compromise without prior network access. Firmware 2.4.1 removes the fleet‑wide RSA files, encrypts credential storage, and disables the GPS response.

The author runs a small Django site on SQLite (≈10 k rows) and shares practical lessons learned. Enabling WAL mode is standard, but a full‑text search query over 4 000 rows took 5 s until an `ANALYZE` command was run, which generated table statistics and reduced the query time to ~0.05 s. Deleting large numbers of rows can lock the database for several seconds; concurrent writers then hit the 5‑second timeout and crash. The author mitigates this by batching cleanup operations. For backups, two approaches are used: (1) a manual `VACUUM INTO` to a temporary file, gzip compression, and storage with restic on S3, noting occasional OOM‑related lock issues; (2) Litestream for continuous, incremental replication to S3 with a 400‑hour retention setting. Splitting tables into separate SQLite files is mentioned as a useful organization trick. Overall, the site remains stable with SQLite, provided statistics are kept up‑to‑date, write contention is limited, and backups are handled reliably.

The author notes a recurring habit: each new notebook quickly becomes over‑organized, prompting the creation of a new one. To break this cycle, they designated an old, low‑quality notebook as a “dirt notebook” (nicknamed “The Drainage Channel”) for unstructured, messy entries. The paper’s poor quality forces the use of cheap ballpoint pens and prevents neat formatting, reinforcing a deliberately chaotic style. Over a week they recorded random podcast quotes, story ideas, personal memos, and learning topics without any systematic layout. Reviewing the pages weekly reveals forgotten thoughts and reinforces the value of spontaneous note‑taking. The immediate goal is to fill the notebook and accept its messiness, with a possible later transition to higher‑quality paper and fountain‑pen ink once the habit is established.