AI Reading for Monday September 22

Scientists Printed Viruses Designed by AI and They're Successfully Reproducing - Futurism

We are so cooked - Semafor

Craig Venter urged “extreme caution” for any viral enhancement research, warning that if similar methods were used on pathogens like smallpox or anthrax “I would have grave concerns.”

I feel like we could get turned to grey goo by stuff that wasn’t known to be pathogenic.

MIT and Microsoft introduced PDDL-INSTRUCT, improving LLM planning accuracy using planning domain definition language with logical traces and external validation, achieving large gains on benchmark tasks. - MarkTechPost

Syngenta Vegetable Seeds and Heritable Agriculture harness AI to identify the best vegetable varieties for growers, using historical geographic, crop-trial, weather, soil and proprietary data to better predict top-performing varieties by region—at up to 10‑meter resolution. - Business Wire

What makes China confident enough to ban microchips made by US firm Nvidia? - ABC Australia

Need for speed: Nvidia wants 10Gbps HBM4 to blunt AMD’s MI450, report claims — company said to be pushing suppliers for more bandwidth - Tom's Hardware

In addition to the A19 SOC, Apple added in-house wireless and modem chips, replacing Broadcom parts, giving Apple control of all core iPhone chips. Interviews with Apple's Tim Millet, Arun Mathias, and Kaiann Drance on the new neural accelerators in A19 Pro GPU cores, C1X modem, N1 wireless chip. - CNBC

CoreWeave courts controversy: (“This is a value‑destructive entity,” says D.A. Davidson’s Gil Luria) and (“No shit, I’m borrowing money” - CEO Michael Intrator). If GPUs stay in short supply they mint money, if they borrow a ton and can't rent out the GPUs they are toast. - Forbes

Core Scientific shareholders want more money from CoreWeave merger due to prized power access. - Business Insider

LinkedIn will start using member profiles, posts, resumes and public activity to train its AI models, with the change enabled by default for users; data from under-18s is excluded. - TechRadar

Users can opt out via the “Data for Generative AI Improvement” setting which only stops collection going forward—data collected before opting out will remain in the training environment;

Data Sovereignty for Financial Institutions: External AI and SaaS services deemed harmful. - Finextra Research

AI smart glasses are starting to spread across workplaces for guidance and documentation, but raise surveillance and privacy concerns and spur calls for contractual and legislative safeguards. - Bloomberg Law

ByteDance reportedly has AI glasses prototypes ready but may wait for for market to heat up - DIGITIMES

Two Tesla shareholder-influencers attempted a coast-to-coast drive from San Diego to Jacksonville using a Model Y on Tesla FSD v13.9 and crashed before about 60 miles after hitting avoidable road debris; occupants spotted the debris in time, but the vehicle sustained major damage - Electrek

UK Nvidia-backed startup Wayve has begun testing its self‑driving technology with Nissans on Tokyo streets ahead of a planned 2027 consumer launch; implementation will be an “eyes on, hands off” (level 2) system using cameras, radar and a more expensive lidar for redundancy. - the Guardian

The car as software platform, with AI and over-the-air update for everything. - Ward's Auto

Hertz’s AI vehicle scanner triggered a billing dispute and backlash, prompting the company to add human contact options and review consumer-facing AI processes. - CX Dive

Needs a process to drive through a lane before and after, take pictures and consent and appeal and talk to a human if necessary.

Fast inference can unlock parallel agentic coding workflows but escalate compute demand and costs, with pricing models currently in constant flux. - Martin Alderson

Generative AI was sold as a time-saver for creatives, but it turns them into content mills. - It's Nice That

AI's repackaging of content without credit creates a tragedy of the commons, which in turn leads to fencing the commons. Perhaps attribution layers and compulsory licenses could enable creators to get credit and funding. - Hacker Noon

Thomas Friedman's Magical Thinking on AI - aiguide.substack.com

TL;DR Friedman is not wrong to worry about what's going to happen vis a vis the U.S., China, and AI. However, we need less magical thinking based on what he read in newspapers (or heard from his friend Mundie) and more realism. Those who are in a position of crafting AI regulations would be much better off digging into the many excellent reality-grounded proposals, ones that require reasoning from, yes, those slow, analog-era humans. Gatekeepers who dumb it down to a 3rd grade level for legislators and voters shape the narrative and wield a lot of power.

Organizational roles for AI agents - CIO

LLMs are terrible at lists, better at simple yes/no questions. - Terence Eden’s Blog

A Whitney Houston tribute tour with AI plus live musicians - TechRadar

Teen says his Facebook and Instagram accounts were wrongfully locked by AI and he can't get them back - Connecticut Inside Investigator

Australia found that Kmart breached Australians’ privacy by using facial recognition technology on every shopper in attempt to reduce refund fraud. - OAIC

Google get SOTA deep research without transformers using diffusion. - Google Blog

Feeding stray cats via drone (preferably without getting swatted out of the air) - Instagram

Follow the latest AI headlines via SkynetAndChill.com on Bluesky