AI Reading for Friday April 18

Artificialanalysis.ai/ initially puts o4-mini on the efficient frontier, beasting (is that a word?) Gemini 2.5 on cost and performance curve.

edit: ah I forgot I shared that yesterday! slipped it in at last minute and it was still in the ‘new’ queue

Grok 3-mini with their new API is pretty good and cheap.

---

Vibe Check: o3 Is Here—And It’s Great - Dan Shipper

Deep research ‘light’, tools, long context, fast.

only 30% cheaper than o1, not very cheap, I suspect we might see rate limiting in the $20-month plus subscription, you are not going to be able to summarize all the 200k-token books you want. tempting to get the $200 pro subscription even, for now I have deep research in gemini and perplexity and soon claude, not going to pay $200 for multiple subs at this point LOL.

---

Despite leadership in technology and research, OpenAI's path to big profits and competing with and joining the Mag 7 is still highly uncertain - FT

---

OpenAI has published a 34-page "A practical guide to building agents" - OpenAI

---

Google rolls out Gemini 2.5 Flash, gifts college students a free year of Gemini Advanced - Ars Technica

It’s a war on all fronts, AI capabilities, product, pricing, speed

---

David Silver and Richard Sutton on the future of agents coworking with you and gathering and learning from vast long-term streams of experience - Google

Paper covers research discussed in recent Google video with David Silver and Hannah Fry

---

12-factor agents seems like an interesting concept. - GitHub

---

Is using AI to write an email poor etiquette? Or does it respect the recipient's time by using all available tools to address important stuff as effectively as possible? - Fast Company

---

Companies begin to pivot AI efforts from exploration to exploitation - WSJ

---

How 7 Goldman Sachs professionals use the company's AI tool, which David Solomon says can automate 95% of writing a prospectus. - Business Insider

---

Company apologizes after AI support agent invents single-device policy, causes user uproar - Ars Technica

---

Now when you post photos to socials, stalkers can probably figure out where they were taken in seconds. - TechCrunch

ChatGPT can now make a good guess where a photo was taken, which is slightly terrifying - BGR

---

Musk not banned from Yankee Stadium after offering to buy the team and fire Aaron Boone, AI overviews notwithstanding - Snopes

---

Google’s AI Overviews direct women to get pre-abortion ultrasounds from anti-abortion groups - Bloomberg

---

BYU CS grad student from Japan faces deportation over a Mormon fishing group that took too many fish - Deseret News

---

Discord's AI face scanning age checks 'start of a bigger shift' - BBC

---

How Elon Musk reportedly staked Tesla’s future on robotaxis against the advice of his own senior management team warning it might not ever make money. - Fortune

---

British army's drone bug-zapper energy weapon - New Atlas

---

LMArena Chatbot-Ranking Website Is Becoming a Real Company. - Bloomberg

new website in the works. looking forward to the rating of all the new stuff from OpenAI, Google, Grok.

---

Venezuelan migrants relied on clickwork to survive. Now AI is replacing them. Initially AI increased demand for labeling, is that already going away? Highly skilled educated annotators in math and science are in huge demand. - Rest of World

---

Follow the latest AI headlines via SkynetAndChill.com on Bluesky