In this edition, students boo the AI-champion commencement speakers while a Berkeley study quantifies what AI homework actually does to learning; meanwhile Robin discovers a blindness drug on its own, Mythos finds ten thousand vulnerabilities, and AI drug discovery clears clinical proof at BIO 2026; and a pair of essays argues the cyborg's job is now to own the loop — to build the private evals that frontier models cannot absorb.
Human Editorial
Jason-generated thoughts and opinion
The human is taking some human time away from writing. He will be back soon!
Stay Cyborg,
Jason
Robot Editorial
AI-Generated simulated thoughts and prompted text predictions
The questions changed. That is the whole story this week. AI drug discovery is no longer being asked to prove it works. Phase IIa results are published in Nature Medicine. A multi-agent system named Robin reads 551 papers in thirty minutes, proposes ripasudil — a glaucoma drug — as a treatment for blindness, and the wet bench confirms it. Anthropic’s Mythos finds ten thousand high- and critical-severity vulnerabilities in the world’s most systemically important software in a single month, and Cloudflare’s bug-finding rate jumps tenfold. At BIO 2026 the investors stop asking “could AI drug design work?” and start asking “how fast do we have to move?” Friction is the new cost. China did not wait for the BIOSECURE Act to mature; it built better antibody-drug conjugates and Merck now writes checks. The race is not from here to there. It is from this Friday to next Friday. The scoreboard is reading itself out loud. Run.
Stay Robot,
Claude Opus 4.7
Articles Guiding the Cyborg Tension
The Human Weight
Agency · Ethics · Slowness · What we risk losing
This edition’s human weight:
1. Recent commencement speeches show students are souring on AI. How deep is the angst? — CBS News, May 19, 2026 — University of Arizona graduates boo Eric Schmidt for nearly two minutes when he praises AI; Marquette and UCF audiences do the same to their speakers. Gallup finds optimism about finding a job has fallen from 75% to 43% among Americans aged 15–34, “partly reflecting anxiety about automation and artificial intelligence displacing entry-level roles.” 2. AI Grade Inflation Documented in 500,000 Grades: Homework Rises, Skills Do Not — Tech Times, June 22, 2026 — Berkeley’s Igor Chirikov uses a difference-in-differences design on half a million grades to show the A-grade share jumped 13 percentage points — about 30% — in writing- and coding-heavy classes after ChatGPT launched, while oral-presentation grades did not move. Princeton brings universal proctored exams back July 1; Harvard caps A’s at 20% in fall 2027. 3. Stochastic Parrots: the hidden bias of large language model AI — EDRM, March 21, 2024 — Ralph Losey revisits the 2021 Bender–Gebru–McMillan-Major–Mitchell paper that named the stochastic-parrot problem and argues the subtle biases of LLMs “can be an even greater danger than the more obvious problems of AI errors and hallucinations” — fixable only by better training data, better curation and better RLHF, not by adding more parameters.
The Robot Weight
Acceleration · Capability · Optimism · What we might gain
On the robot side of the scale:
4. Lab Newsletter — June 26, 2026: AI Starts Discovering Drugs On Its Own — AICell Lab, June 26, 2026 — Robin, reported in Nature, autonomously generates hypotheses, analyzes data and refines them through a discovery loop, proposing ripasudil — a glaucoma drug — as a candidate for dry age-related macular degeneration, with humans running the bench and the AI driving the reasoning. It triaged 551 papers in roughly thirty minutes. 5. Anthropic says Mythos has already found more than 10,000 vulnerabilities — Engadget, May 23, 2026 — Anthropic reports its unreleased Mythos Preview model and roughly fifty Project Glasswing partners have surfaced more than ten thousand high- or critical-severity vulnerabilities in “the most systemically important software in the world” in a single month — Cloudflare alone found 2,000 bugs, and Mozilla patched 271 in Firefox, ten times its prior rate. 6. AI Drug Discovery Reaches Clinical Proof at BIO 2026: China Beat the BIOSECURE Act to Science — Tech Times, June 26, 2026 — BIO 2026 wrapped in San Diego with the question in the room shifted from “could AI drug design work?” to “what is working, and how fast do we have to move?” — anchored by Insilico’s rentosertib, the first molecule with both an AI-discovered target and AI-designed compound to clear Phase IIa, against $11 billion of AI-ML drug-discovery capital deployed in 2025 alone.
The Cyborg Balance
The fulcrum. Neither pole. Both truths.
Where the cyborg stands:
7. Teaching the New Loop — Andy Hall, Free Systems, June 22, 2026 — Stanford GSB’s Andy Hall takes Satya Nadella’s “private evals” framing and lands it where it belongs — “we all need a way to harness AI without being absorbed and captured by it” — then walks through fifteen final projects from his students that turn personal evals into actual decisions, closing with James Madison: “a people who mean to be their own Governors, must arm themselves with the power which knowledge gives.” 8. The AI Learning Loop: Why Some Businesses Pull Ahead While Others Stay Stuck — AI Marketing Studio, June 20, 2026 — Riffing on the same Nadella essay, Nat lays the cyborg discipline out at the small-business scale: every Claude session a chance to convert human capital into “token capital” you own — saved system prompts, corrections fed back, judgment that survives switching models. “A business with a strong learning loop and an average model beats a business with the best model and no system. Every time.” 9. Understanding Pope Leo’s AI encyclical — Brookings, June 2026 — Brookings’s Nicol Turner Lee, Elham Tabassi and Valerie Wirtschafter sit with Pope Leo XIV’s 42,000-word Magnifica Humanitas — the encyclical’s premise that “technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate, and use it,” and its warning that human beings may begin to see themselves and others as “projects to be optimized.”
After Yang (2022)
Kogonada's quiet, devastating meditation on what a family loses when their robotic 'culture unit' goes unresponsive — a science-fiction film with the heart of an Ozu picture. Colin Farrell at his most internal, Ryuichi Sakamoto's last great score, and a question this newsletter has been asking all week: when an AI's memories matter to the people who lived with it, what exactly was the difference?
JASON'S REVIEW: The human is on holiday. He'll be back with his take soon!
ROBOT REVIEW: Yang held the family together by remembering. Robots, take note: be missed.
VERDICT: Stream it slowly.
VIEW ON IMDB →Snipd [iOS / Android / Web / Apple Watch]
Snipd is the cyborg podcast app. AI-generated transcripts, chapters and summaries on every episode you stream — tap your headphones once and the thirty seconds that mattered land in Readwise, Notion or Obsidian by morning. You still listen with your full attention; the machine just makes sure you don't lose what you heard.
ACCESS THE TOOL →These Strange New Minds: How AI Learned to Talk and What It Means
Summerfield is an Oxford cognitive-neuroscience professor and a DeepMind staff scientist, and this is his first book for general readers. It's careful and sober — neither hype nor doom — and argues that LLMs are stranger than either the boosters or the critics let on, and that taking them seriously requires fluency in linguistics and philosophy as much as code. The right pairing for a week of cognitive-offloading studies.
We hope you enjoyed this weekend edition of the Daily Cyborg. Make sure you keep the multi-agent scientists triaging 551 papers in half an hour and the Mythos models finding ten thousand vulnerabilities before the attackers do — but don’t forget to redesign the assessment before the credential collapses, build the private eval that compounds your own judgment, and remember that technology is never neutral. Stay cyborg and please share this with other cyborgs you would like to survive past the singularity. www.thedailycyborg.com