D.A.D.: A How-To Guide to Using AI for Archival Research — 6/29
The Daily AI Digest
Your daily briefing on AI
June 29, 2026 · 10 items · ~5 min read
From: OpenAI, Semgrep, NBER, El País, Hacker News, arXiv
D.A.D. Joke of the Day
My AI assistant said it needed more context to help me. So did my last three therapists. At least the AI is cheaper.
What's New
AI developments from the last 24 hours
HP Bets Its Entire Company on a Single AI Vendor: OpenAI
HP Inc. announced it will expand its OpenAI partnership across the company after pilots that began in February 2025. The rollout targets customer solutions, employee productivity, and software development. HP cited early results: one engineer reportedly processed 122 pull requests across 43 projects in weeks using OpenAI models, and a security team remediated bugs in a day that HP estimates would have taken up to a month manually. The company says OpenAI's platform will become its unified AI infrastructure as workflows move from pilots to production.
Why it matters: A major hardware maker standardizing its whole company on one AI vendor shows how fast enterprises are moving from experiment to lock-in. Worth a grain of salt, though: the eye-popping figures—122 pull requests in weeks, a month of bug-fixing compressed to a day—come from OpenAI's own case study and are HP's reported results, not independent benchmarks.
Open-Weight Chinese Model Beats Claude on Security Vulnerability Detection
A Chinese open-weight model just outscored Claude on a real-world security task. Semgrep tested GLM 5.2, released last week by Zhipu AI, against its benchmark for detecting IDOR vulnerabilities—a common flaw where users can access data they shouldn't. GLM 5.2 hit 39% F1 score versus Claude Code's 32%, at roughly $0.17 per vulnerability found. The catch: Semgrep's own multimodal pipeline still leads at 53-61% F1. GLM 5.2 runs about 40 billion parameters per query despite having 750 billion total, and its weights are freely available.
Why it matters: Open-weight models matching or beating proprietary ones on specialized tasks—especially from Chinese labs—signals that the moat around frontier AI may be narrower than the big labs would like.
What's Innovative
Clever new use cases for AI
Patient Feeds Raw MRI to Claude, Gets Different Diagnosis Than His Doctor
A patient with a diagnosed shoulder injury decided to get a second opinion—from an AI. He fed his raw MRI files to Claude Code running Opus 4.8 and asked it to assess his rotator cuff. The result: the AI contradicted his doctor's diagnosis of a significant tear, concluding there was no discrete tear at all. The hour-long analysis used multiple AI "subagents" for cross-checking. He also ran his treatment plan through GPT 5.5 Pro, which flagged two prescribed therapies as unsupported by clinical guidelines—including one registered in Germany as homeopathic "without a therapeutic indication."
Why it matters: This is one patient's experiment, not validated medical practice—but it previews a near-future where patients can interrogate complex diagnostic data directly, shifting the dynamics of expertise and second opinions.
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
Age Verification Laws Draw Scrutiny as Potential Surveillance Infrastructure
An opinion piece circulating in tech circles argues that age verification laws spreading across US states, Europe, Australia—and now Canada—aren't primarily about protecting children, but are infrastructure for linking online speech to real-world identities. The argument: once platforms must verify age via government ID or SSN, authorities gain a ready-made system to attribute anonymous posts to specific people without traditional investigation. Canada is weighing its own sweeping version. On June 10 the government tabled Bill C-34, the Safe Social Media Act, which would lock under-16s out of regulated platforms, require age-verification or age-estimation measures that critics say would pull in every Canadian user, fold in the pornography age-verification rules from an earlier Senate bill (S-209), and create a powerful new internet regulator—the Digital Safety Commission of Canada. Reaction has been sharp: law professor Michael Geist calls C-34 a "Trojan Horse," and the advocacy group OpenMedia warns it repeats the privacy mistakes of the UK's age-verification regime. The original piece offers analysis rather than documented evidence, and community reaction is divided—some see chilling implications for anonymous speech, others counter that directly advocating for speech protections would be more effective.
Why it matters: As age-verification mandates spread—now reaching Canada's Parliament—the framing fight will shape how platforms and users respond: child-safety measure, or identity-surveillance infrastructure? The worry isn't the stated goal but the machinery it builds. Once every user must prove who they are to log on, that proof exists to be used for other things—and the system outlasts the government that built it.
Discuss on Hacker News · Source: nonogra.ph · Bill C-34 (Parliament of Canada) · Michael Geist
Brown Professor Alleges 50 Students Used AI to Cheat, Calls It Largest Ivy League Fraud Case
A Brown University economics professor alleges he has detected mass AI-assisted cheating involving at least 50 students on a March midterm in his advanced mathematical economics course. Professor Roberto Serrano claims he has "conclusive evidence" of the fraud and says university administration initially met his report with silence until he escalated to the Academic Code Committee. He's pushing for broader public debate about AI's threat to academic integrity.
Why it matters: Universities are still improvising their response to AI tools that can solve problem sets and write essays—this case, if substantiated, signals how quickly the cheating calculus has changed and may force institutions to accelerate policy decisions on AI use, detection, and assessment redesign.
What's in the Lab
New announcements from major AI labs
Quiet day in what's in the lab.
What's in Academe
New papers on AI and its effects from researchers
AI Summaries May Starve Publishers of the Traffic They Need to Survive
In a new working paper distributed by the NBER, Harvard Business School economist Alex Chan argues that AI answer systems may be undermining the economic model that sustains open-web publishing. The core mechanism: when AI platforms summarize content instead of sending users to original sources, they retain traffic that publishers need to survive. Chan proposes this creates a self-reinforcing cycle—as publishers struggle, less quality content gets created, which makes users more dependent on AI aggregators. The theoretical paper suggests policy interventions including royalties for displaced visits and compensation for AI systems that rely heavily on specific content domains.
Why it matters: This frames the AI-versus-publishers tension not as a copyright dispute but as a market design problem—the kind of framing that could shape how regulators and platform companies approach content compensation.
Economist's Guide Shows Historians How to Analyze Archives Without Coding Skills
A new working paper by University of Pittsburgh economic historian Andreas Ferrara, distributed by the NBER, offers a step-by-step guide for economic historians to use LLMs in their research, aimed at scholars without programming backgrounds. Ferrara argues these tools are lowering barriers to working with messy historical sources—handwritten ledgers, old photographs, audio recordings—that previously demanded serious data science skills. The paper includes four worked examples with replication files: classifying emotions in paintings, linking census records without names, measuring newspaper sentiment around the 1882 Chinese Exclusion Act, and scoring the emotional delivery of FDR's wartime speeches.
Why it matters: This signals how AI is reshaping academic workflows—historians can now tackle archival sources that would have required a technical collaborator or months of coding, potentially accelerating research and changing what questions get asked.
Google Builds Tool to Catch Errors in Scientific Papers Before Peer Review
Google researchers unveiled the Paper Assistant Tool (PAT), an AI framework designed to catch errors and suggest improvements in scientific manuscripts before peer review. Piloted at two major computer science conferences (STOC and ICML), the system checks theoretical results, validates experiments, and flags potential flaws. Google reports PAT achieved 34% better recall on mathematical errors compared to baseline AI methods when tested on a benchmark dataset. The tool is positioned as a pre-submission aid for authors, not a replacement for human reviewers.
Why it matters: If reliable at scale, automated pre-screening could accelerate peer review—a process notorious for delays and overburdened referees—while catching errors earlier, when they're cheaper to fix.
Weather Forecasting Offers a Preview of AI-Driven Institutional Disruption
A new arXiv paper argues that machine learning's proven success in weather prediction is just the beginning—the harder transformation lies ahead. Weather and climate centers must now rethink their entire operational model: how they develop forecasts, exploit data, verify accuracy, and deliver services. The authors contend that institutions need new infrastructure, new skills, and new frameworks while somehow preserving the scientific rigor that keeps forecasts reliable. It's less a technical paper than a strategic roadmap for an industry facing AI-driven disruption.
Why it matters: For any organization watching AI reshape a specialized field, weather forecasting offers a preview: the technology works, but adapting institutions, workflows, and professional identities is the real challenge.
AI Helps Struggling Young Writers Catch Up but May Limit Top Performers
A study of 40 children ages 7-12 found that AI-assisted storytelling dramatically narrowed the quality gap between students—by 83.5%—but did so by lifting weaker performers rather than enhancing top ones. The researchers call this a "floor-raising convergence pattern": children who struggled with traditional storyboarding improved significantly with AI help, while higher-performing children saw their creative ceiling constrained. The effect was selective—AI boosted creativity and story richness but didn't improve coherence or narrative structure, which remained tied to each child's baseline abilities.
Why it matters: This is early evidence that generative AI in education may function as an equalizer rather than an amplifier—useful for closing gaps, but potentially limiting for students who would otherwise excel.
What's On The Pod
Some new podcast episodes
The Cognitive Revolution — AI:AM #4: Cameron on Model Consciousness, Duvenaud's Gradual Disempowerment, swyx's AI-Eng Alpha