The Intelligence Age, Touchy Robots, & SaaS Dangers

TLDR;

Technology:

🍓 Sam Altman’s ‘The Intelligence Age’

👜 Arcade.ai’s Text-to-Product

🖐🏽 Touchy and Store Robots

Markets:

😱 Anthropic Eyeing a $40 Billion Valuation

🚨 Klarna to Replace Salesforce and Workday

Product:

🎐 Sam Altman & Jony Ive Confirm AI Hardware Company

⚜️ Electronic Arts’ New AI Features

🫡 Is Meta Running Away with the Wearable Market?

FRONTIER AI

Sam Altman’s ‘The Intelligence Age’

Sam Altman has published a blog post laying out his vision of what AI will mean for civilization over the next decades.

He discusses the importance of energy and chips produced at scale and the arrival ‘in the next few thousand days’ of superintelligence. He generally shares an extremely bullish and positive view of what’s to come, breaking with Silicon Valley's tendency to show more concern about the dangers of AI.

In the meantime, OpenAI has released the biggest multilingual dataset for LLM evaluation across 14 languages, including less common ones like Yoruba and Swahili, and they just announced the launch of the OpenAI Academy for developers, which goes to prove how OpenAI is transitioning to a product company instead of a ‘God AGI’ one. Good.

TheWhiteBox’s take:

I have to give it to him; he’s good. It is remarkable how someone can write 1,109 words about how AI will change the world and show no proof whatsoever that we are getting anywhere near that vision.

But I don’t think Sam is trying to prove anything. This blog is actually an outcry for more chips and energy, which is interesting timing for Sunday’s article, in which we talked about Big Tech’s plans to build that vision Sam is precisely talking about.

He talks about Deep Learning being ‘the technology’ and ‘the algorithm’ (neural networks, specifically the Transformer) that will ‘inevitably’ lead us to a new age for humanity. Does this narrative remind you of anything? Of course, it’s The Bitter Lesson, the mind virus/religion we discussed last Sunday!

The idea that all we need is more compute, is extremely palpable in Sam’s post, and he clearly thinks the necessary breakthroughs to unlock that compute have already been discovered.

I don’t know about you, but I don’t understand how a technology that, after seeing 15+ trillion tokens of data and consuming GWh of energy*, still can’t multiply 10-digit-numbers×10-digit-numbers consistently is taking us anywhere near superintelligence. How these people believe that simply adding more compute on top of such models is enough is not based on facts but beliefs.

* I estimated Meta’s overall energy consumption to train Llama 3.1 at 11.6 GWh, knowing it ran on a 16k GPU cluster for 54 days at an average of 40% MFU (40% of its theoretical processing peak), and with each GPU consuming 1,400 Wh. That’s enough to provide energy to 1.1k households an entire year.

And still, they still can’t multiply consistently.

TEXT-TO-PRODUCT

Arcade.ai’s Interesting Proposal

Arcade.ai is a company that offers a unique perspective: using AI to generate products that are then manufactured in a custom-made fashion. This…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.