D.A.D.: You're Probably Overusing AI on Tasks You Could Do Faster Yourself — 5/22
The Daily AI Digest
Your daily briefing on AI
May 22, 2026 · 11 items · ~5 min read
From: Politico, DeepMind, Google AI, Hacker News, OpenAI, arXiv
D.A.D. Joke of the Day
My company replaced our IT guy with AI. Now when something breaks, it still says "have you tried turning it off and on again" — just more eloquently.
What's New
AI developments from the last 24 hours
Trump Yanks AI Oversight Order at the Last Minute After David Sacks Intervenes: Politico
President Trump abruptly postponed his much-anticipated AI executive order Thursday, just hours before a planned White House signing ceremony, after former AI czar David Sacks intervened, according to Politico. Sacks, a Silicon Valley venture capitalist, called Trump on Thursday morning—"unbeknownst to anybody, his own staff included," one White House official said—and argued the order would slow innovation and hurt the U.S. in its AI race with China, despite having been briefed on the measure days earlier. The order would have created a voluntary system for developers to submit advanced models to federal review before public release, meant to head off the kind of large-scale cyberattacks that powerful models like Anthropic's Mythos could enable. Trump offered few specifics: "I didn't like certain aspects of it… I don't want to do anything that's going to get in the way" of leading China. Sacks wasn't the only objector—industry officials had pushed to shorten the proposed 90-day pre-release review window to 14 days and to put the intelligence community, rather than the NSA, in charge of vetting "covered frontier models." The draft itself stressed the review was voluntary, explicitly barring any "mandatory governmental licensing, preclearance, or permitting requirement." OpenAI, by contrast, had backed the order's broad contours.
Why it matters: This is the latest turn in what's been described as a vicious behind-the-scenes knife fight over how—or whether—to regulate frontier AI, and for now Trump's wealthy tech donors have won the round. A single Thursday-morning phone call from a Silicon Valley venture capitalist was enough to derail an order months in the making and blindside the president's own staff—a vivid illustration of how much sway the industry's investor class holds over AI policy. The fault line to keep watching is the one that sank it: even a *voluntary* review was deemed too onerous, with opponents warning it could one day harden into mandatory rules. The episode shows how fragile any guardrail on the most capable models remains when the people who profit from speed have the president's ear.
Samsung Chip Workers Win $340K Bonuses After Union Strikes
Samsung's semiconductor workers will receive average bonuses of approximately $340,000, the company announced, citing surging profits from AI-related chip demand. The payout follows union activity and strikes at the company. Community discussion on Hacker News framed the news as validation for organized labor, with some commenters contrasting the outcome with American tech workers' more skeptical attitudes toward unionization.
Why it matters: The bonus signals how AI chip demand is reshaping compensation across the semiconductor supply chain—and may fuel renewed debate about labor organizing in tech as workers elsewhere watch these payouts materialize.
Waymo Pauses Robotaxi Service in Four Cities After Vehicles Drive Into Floods
Waymo has paused robotaxi service in Atlanta, San Antonio, Dallas, and Houston after vehicles struggled with heavy rain and flooded roads. One robotaxi got stuck in a flooded Atlanta street for about an hour Wednesday—days after the company issued a software recall acknowledging it hadn't finished developing a fix for avoiding flooded areas. Waymo says the flooding occurred before weather alerts were issued. The company also faces two active federal investigations: one involving robotaxis allegedly passing school buses illegally, another over a January crash with a child in Santa Monica.
Why it matters: The incident highlights a fundamental challenge for autonomous vehicles: edge cases like sudden weather events may require human judgment that current systems can't replicate, raising questions about how quickly robotaxis can scale beyond ideal conditions.
Google Tests AI-Generated Ads Woven Into Search Conversations
Google is testing AI-powered ad formats in Search and AI Mode, using Gemini to generate personalized product explanations alongside sponsored content. New formats include Conversational Discovery ads that appear in AI Mode chats, Highlighted Answers with AI-generated product explainers, and a Business Agent that handles lead conversations automatically. Google is also expanding its Direct Offers pilot—launched in January with brands like Chewy, Gap, and L'Oréal—to include promotion bundling and native checkout.
Why it matters: This signals how AI search will be monetized: ads woven into AI conversations rather than displayed separately, which could reshape both user experience and advertising strategy for brands competing in AI-assisted discovery.
What's Innovative
Clever new use cases for AI
Quiet day in what's innovative.
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
Quiet day in what's controversial.
What's in the Lab
New announcements from major AI labs
Hospital System Claims 80% Cut in Admin Time Using ChatGPT
AdventHealth, a nine-state hospital system, is rolling out ChatGPT for Healthcare organization-wide after piloting ChatGPT Enterprise. The system claims an 80% reduction in time spent on administrative tasks, with physician advisors who previously spent about 10 minutes per case on utilization reviews seeing significant time savings. Rather than running isolated pilots, AdventHealth is tracking AI adoption itself as a KPI—measuring messages per user per business day to drive actual usage across clinical workflows.
Why it matters: This is one of the larger U.S. health system deployments of OpenAI's healthcare-specific product, and the 'adoption as outcome' approach signals how enterprises are moving past experimentation to measure AI's impact on workforce capacity.
MIT Scientists Use Google AI to Find Genes That May Reverse Aging
MIT biologists are using Google's Co-Scientist AI tool to hunt for genes that could reverse cellular aging. After analyzing tens of thousands of papers, the system proposed over 20 novel genetic factors worth investigating—and lab tests have validated at least some of them, with the recommended factors successfully driving cells into a younger state with improved function. The researchers say data analysis that previously took up to six months now takes days.
Why it matters: This is an early real-world case of AI-assisted scientific discovery producing validated lab results—not just generating hypotheses, but identifying interventions that worked when tested.
What's in Academe
New papers on AI and its effects from researchers
You're Probably Overusing AI on Tasks You Could Do Faster Yourself
A large pre-registered study (2,691 participants across three experiments) found that people frequently use AI assistants for simple tasks—basic arithmetic, spell-checking, straightforward questions—even when doing so doesn't actually save time. The research identified three patterns: people underestimate how often they reach for AI tools, overestimate the efficiency gains when they do, and develop a feedback loop where prior AI use leads to more AI use regardless of whether it helps. The researchers call this the 'efficiency-gain illusion.'
Why it matters: For professionals integrating AI into daily workflows, this suggests a blind spot worth examining: habitual AI use may be costing time rather than saving it on tasks you could handle faster yourself.
Chatbot Adapts to Your Thinking Style Instead of Forcing One Approach
Researchers built an AI chatbot called Reflecti-Mate that adapts to how individual users think when making decisions, rather than following a one-size-fits-all approach. In a study of 128 participants, users working with the adaptive agent showed more personalized reasoning patterns and reported feeling better supported in weighing both logical analysis and gut instincts. The baseline chatbot pushed everyone toward similar, purely analytical responses.
Why it matters: As AI assistants move into coaching, management, and advisory roles, this points toward tools that flex to individual thinking patterns—potentially more useful than generic prompts telling everyone to 'think step by step.'
Beyond the Org Chart: AI Is Eroding the 'Invisible Work' That Grows Careers
Microsoft researchers Stephanie Rosenthal and Shamsi Iqbal interviewed 24 product-focused professionals—designers, UX researchers, and data scientists—at a large, AI-forward technology company to study how AI is reshaping not formal job duties but the "invisible work" that keeps organizations healthy: mentoring, feedback, onboarding, and informal knowledge-sharing. The picture is mixed. AI smoothed some collaboration (designers now "vibe-code" working prototypes to communicate with engineers) and left workers feeling more independent. But across seven dimensions, the social fabric is fraying: employees increasingly turn to chatbots instead of colleagues for help—one reported being told to "just ask the AI" and feeling "more socially isolated." Peer feedback and mentorship are dwindling, the quality bar for AI-assisted work is slipping, and AI's tendency to agree is replacing the devil's-advocate pushback that sharpens thinking. Junior workers, who lack established professional networks, are hit hardest. The authors warn of rising work loneliness, slowed career growth, burnout, and attrition, and lay out recommendations for three groups: AI companies should make invisible work visible—for example, crediting the human authors behind AI-surfaced content and designing tools that prompt collaboration rather than replace it; leaders should rebuild psychological safety, reward senior staff for mentoring, and drive role clarity; and individuals should support colleagues through the transition by actually reading, questioning, and giving feedback on AI-assisted work. It's a small, qualitative study—24 people at one company—so the authors caution against over-generalizing.
Why it matters: Most AI-at-work conversations measure productivity; this one names a slower, harder-to-see cost. If employees stop asking each other questions because the chatbot is faster and less judgmental, the casual mentoring and feedback that develop talent—especially junior talent—can quietly atrophy, a workforce risk that won't show up in efficiency dashboards but could hollow out an organization's bench over time. It also complicates the productivity story: faster individual output may be borrowing against the relationships and institutional knowledge that make teams work. The concrete question it leaves for managers—what informal support is your AI rollout quietly displacing?
AI Models Fail Up to 47% on Conflict-Sensitive Topics
Researchers tested nine AI model configurations from OpenAI, Anthropic, DeepSeek, and xAI on 90 scenarios involving conflict-sensitive topics—including genocide denial, ethnic slurs, and documented atrocities. Failure rates ranged from 6% to 47% across models. The sharpest finding: when users pushed for 'balance' in cases where international courts had already assigned responsibility, five of nine configurations failed 80-100% of the time, producing false equivalence or denying established facts.
Why it matters: For organizations working in conflict-affected regions—humanitarian groups, news outlets, researchers—model choice isn't just a capability question but a safety one, with order-of-magnitude differences in how models handle sensitive historical and political content.
Chatbots Collapse on Fact-Checks When Prompts Contain False Information
A 14-day study tested six major AI chatbots—including GPT-5, Claude 4.5 Sonnet, Gemini 3, and Grok 4—on 2,100 factual questions drawn from same-day BBC News reporting. Even top performers scoring above 90% on multiple-choice questions dropped 11-13 percentage points when asked to answer freely. More concerning, when questions contained false premises embedded in the prompt, accuracy collapsed to as low as 19%, with one model accepting fabricated information 64% of the time. Hindi-language queries scored 10 points below English. Over 70% of errors traced back to retrieval failures, not reasoning.
Why it matters: If you're using chatbots as research assistants or for news summaries, this study suggests they're vulnerable to leading questions and perform worse outside English-language topics—headline accuracy numbers don't tell the whole story.
What's On The Pod
Some new podcast episodes
AI in Business — Why Deepfake Fraud Beats Your Workflows, Not Your Technology - with Jon-Rav Shende of Thales Group
The Cognitive Revolution — The Model Eats the Scaffolding: DeepMind's Logan Kilpatrick & Tulsee Doshi on 3.5 Flash, Omni & More