AI Reading for Tuesday October 29

Lovingly curated by man and machine

OpenAI to move toward custom silicon, AMD in their infrastructure. - The Verge

---

OpenAI builds first chip with Broadcom and TSMC, scales back foundry ambition. There are some interesting inference chip companies to buy but maybe the price tag is too high. - Reuters

---

US sets restrictions on China AI, chip investment. - Bloomberg

---

Polestar says rules are unclear and might ban its cars from US market. - Ars Technica

---

Son goes full Altman. Says AI will be superintelligent soon, cost $9T, consume the equivalent of the US’s power output. Never go full Altman. - The Register

---

More than a quarter of new code at Google is generated by AI - The Verge

---

Musk is seeking more cash for xAI at $40b valuation. - WSJ

xAI wants to double 100,000 GPU setup to 200,000, uses 800 Gbps Ethernet - The Register

It’s a lot for, at this point, an also-ran. They got an excellent ranking on LMarena with a giant 400b parameter model but are still rolling out table-stakes features like the API, no reason for anyone to use it over the top 3 or open source.

Also feels like Musk is writing a blank check to get up to speed, notably to Nvidia; OpenAI trying to reduce dependence on Nvidia.

---

This new 'ChatGPT for products' wants to make it faster to design a product breakthrough. - Fast Company

Using AI for product management with a ‘megaprompt’ - HackerNoon

In general 'Megaprompts' or ‘God prompts’ are an anti-pattern, you get more consistent, predictable results if you iteratively improve small prompts individually with evals, then chain them together in an assistant workflow.

---

The Risks of Too Much AI: Fortune’s Jeremy Kahn - Sloan Review

---

How to tell if you’re going to lose your job to AI, according to a computational neuroscientist - Fast Company

AI is Eating Journalism, Education, and Creatives - HackerNoon

---

AI reshapes education, raising up some teachers and demoralizing others - Axios

---

MIT meta-analysis: use AI for what it's good at, i.e. semantic understanding of lots of data, tools for what they are good at, ie. faithfully executing algorithmic workflows, humans for what they are good at, i.e. critical thinking and creativity - Nature

---

A new image generator 'Red Panda' is being tested and is highly ranked. Maybe a large Chinese lab, based on the name? - pcgamer

---

Sone nice leaderboards, not sure if it's as authoritative as LMarena but nice UI, speed rankigns, will probably look at both going forward.

---

Meta begs to differ with Open Source Initiative on definition of open source AI models. - The Verge

---

Meta is well-positioned to the biggest beneficiary of AI and the largest company in the world. - Stratechery by Ben Thompson

Ok, a big beneficiary but let's be real here.

---

LinkedIn unveiled an AI hiring assistant agent for employers, and an AI coach for job seekers. - euronews

---

Google Prompting Essentials course teaches you how to use AI better. - Coursera

---

GitHub Copilot to support Anthropic and Google models, to add an app builder feature. - The Verge

---

AI can be trained from implicit feedback, like if you ask for clarification or a redo, figure out that something was wrong with the first answer and don't do it again. - Hacker Noon

---

AI slop floods Medium - WIRED

---

What is AI superintelligence? Could it destroy humanity? And is it really almost here? - The Conversation

---

Are we on the verge of a self-improving AI singularity? - Ars Technica

---

Robert Downey Jr issues Stark warning to Marvel execs ‘hijacking’ his performances using AI. I will be dead but my lawyers will be very much alive - Variety

---

Chanel's CEO knew exactly what she was doing when she was at Microsoft HQ and asked ChatGPT to create an image of Chanel management. - Business Insider

---

It's such a fine line between critical thinking and wanting to go back to the Dark Ages (which are supposed to be a golden age for somebody but even kings still gonna die of dysentery or plague) - Hacker Noon

---

The car dashboard of the future with AI sounds super annoying - CNET

---

Upload video of your presentation and get an AI critique? - Inc.

---

For Halloween, use ChatGPT to create the scariest creature you can imagine. - Reddit