AI Reading for Wednesday November 13

OpenAI, Google and Anthropic Are Struggling to Build More Advanced AI. - Bloomberg

AI hitting the wall is sort of a meme of the week. But Moore's law is similar to AI scaling laws. Decades of chip progress were earned through hard science and engineering in physics, materials science, fab processes. Same maybe for AI, progress is not a given, you have to earn it. You could spend $1T on planet scale computers but by itself that might not make commensurate progress.

---

Apple's AI tablet expected in March - Yahoo Finance

---

Apple AI notification summaries create daily weird mysteries to unravel - The Verge

---

Brockman is back at OpenAI, working with Altman to define a new role. - Business Insider

TBH if someone takes a voluntary or not-so-voluntary leave of absence, it's normally to define a new role somewhere else, but OpenAI is not normal.

---

Is AI a Frankenstein that wil kill its creators, an Icarus story of humanity flying too close to the sun?

---

AI Can Save Humanity—Or End It - Kissinger, Schmidt, Mundie - The Atlantic

---

Can We Avoid a Franken-Future with AI? - naked capitalism

---

Neurology nurse fears encroaching AI isn't just a smart assistant, but reduces nurses' autonomy, increases surveillance - Coda Story

---

Burn it down: a license for AI resistance in education - Inside Higher Ed | Higher Education News, Events and Jobs

---

Adobe introduced AI tools that allow users to create modified images based on its stock photography library while compensating original creators. - Yahoo Tech

---

YouTube is testing a feature that lets select creators remix artists' songs using text prompts. - TechCrunch

---

Particle AI partners with publishers to distribute AI-curated news, instead of just scraping their work, like some people - TechCrunch

---

The AI lab waging a guerrilla war over exploitative AI - MIT Technology Review

---

Artists express takes on AI in their art - NY Times

---

SoundHound for drive-through ordering helps restaurants save money and upsell customers, never forgets 'would you like fries with that' - PYMNTS.com

---

Secret math benchmark stumps AI models and PhDs alike. Seems hard to keep out of training sets. - Ars Technica

---

When Masayoshi Son offered Jensen Huang to help buy out Nvidia, around 2016 sounds like - Yahoo Finance

---

Qwen2.5-Coder-32B is an LLM that can code well, that runs locally on a 64GB Macbook Pro

---

Qwen2.5-Coder is free in many size variants and in many languages, will help democratize coding - VentureBeat

---

How to make a Perplexity search shortcut in Chrome (like, in the search bar you could type "perp what are pros and cons for substack, ghost, and beehiiv") or even make it the default search engine - Perplexity

---

An explanation of the paper shared last week about flaws in LLM's implicit models of the world. - MIT News

---

Waymo’s robotaxis are now generally available to everyone in Los Angeles, but no highway trips, which seems pretty limiting in LA, not really as much a factor in SF. - The Verge

---

Amazon offers $110 million worth of free compute to Berkeley and CMU in bid to establish Trainium GPUs and wean users off Nvidia - Yahoo Finance

---

Welcome to LLMdeflation - LLM inference cost is going down fast ⬇️ - Andreessen Horowitz

---

A malicious actor uploads a poisoned model to a public repository, your team deploys it in your environment, it exfiltrates your secrets. - Unit 42

---

if you've got 5h to spare, here's Dario Amodei on the Fridman podcast - YouTube

---

So Human It’s Scary: Joanna Stern spends a day only talking to bots. - WSJ

---

The viral post about chatgpt saving someone having a heart attack, was totally made up by AI. - Reddit

---

"based on every single conversation we have had since the beginning, give me some tough love you think I need to hear that will help me the most grow as a person." dang chatgpt, you didn't have to go so hard

I tried this, it told me

1. Stop Seeking Perfection in Others

2. Prioritize Ruthlessly

3. Face Discomfort Head-On

4. Learn to Let Go

5. Reassess Your Relationship with Control

---

Follow the latest AI headlines via SkynetAndChill.com on Bluesky