Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-09 · high
Weekly AI users outscored both daily users and those who never use it in the OECD's largest-ever student sample.
Why now: The OECD's PISA 2025 results, covering more than 760,000 students across 91 countries, drew coverage dated September 9, 2026.
Daily AI users scored 28 points lower in science than students who never use it on the OECD's PISA 2025 test, per The Verge. The Verge puts that gap at about a year and a half of teaching. The sample is the largest PISA has run: more than 760,000 fifteen-year-olds across 91 countries.
The gap widens to nearly 30 points on summarization tasks, the ones that ask students to condense a text rather than write one from scratch.
Most coverage skips the shape of that curve. Students who use AI once or twice a week outscore both daily users and those who never touch it at all. Never-users miss whatever benefit occasional practice with the tool provides. Daily users may be swapping out the drafting and revising that builds the tested skills.
The study doesn't say which habits separate weekly users from daily ones, or whether daily users were already behind before they leaned on AI more. A one-time test can't sort out which came first.
Each link below shares sources, entities, or timing with this story.
"Adaptive Adversaries" (arXiv:2607.18063) tests agents against attackers that adapt across turns instead of firing one-shot prompts. Claude Opus 4.6 and GPT-5.4 tied at 5.4% aggregate, but per-scenario variance was extreme, with Opus hitting 60% on one scenario where competito...
The argument is that a structurally compressed model's bfloat16 checkpoint is itself only a distillation-recovered approximation, so training the 4-bit student against it inherits that error. Distilling directly from the original model reaches a comparable peak about 7x faster...
An Ask HN thread titled "Is Traditional SaaS Dying?" has the poster describing their own internal projects to replace HubSpot and Float, reporting mixed results. Three points, two comments, pure anecdote. But it's the only honest data point today against the clean "agents eat...
PR #2205 adds the Z private label on bind mounts referenced by RUN in the fetch, git and time Dockerfiles. Without it SELinux blocks access to the bind-mounted pyproject.toml and uv.lock and the build fails outright. Fifteen lines, citing Red Hat's labeling guidance and the OW...
Announced August 28, backing AI processors, memory, networking, storage, robotics and data centers, led by Raghu Raghuram and Martin Casado. The firm's argument is that every layer of the hardware supply chain is capacity constrained, from chips to memory to power, and that's...
Three passes: scrape the Amazon notebook DOM, read exact character positions from the Mac Kindle app's SQLite store to locate fully hidden highlights, then capture Cloud Reader pages to canvas for local OCR through Apple Vision to reconstruct text truncated at those known posi...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.