Like Swaine’s World? Scroll to the bottom of this post to tip the author or to subscribe.
THE WEEK IN ARTIFICIAL INTELLIGENCE
Quote of the Week
“[T]he true intention for these companies is to monetize AI assistants so that they fuel hype and criticism, which in turn create momentum and capital to fund research.
“ChatGPT’s and Claude’s basic purpose is as a societal klaxon, a noise maker, a vehicle to drive interest to fund research into the most speculative science fiction fantasy in human history.”
— Marc Watkins from Rhetorica [link]
I haven’t written about AI for a couple of weeks, so this week I did.
The Egbert Manoeuvre, or Elon vs. OpenAI
What do you think of the idea that “some technologies are so dangerous that they must be put beyond democratic pressure and controlled by a select few?” (And yes, we’re talking about AI.) That’s what Observer columnist Kenan Malik calls the “Egbert manoeuvre” and he sees it at play in the recent news about OpenAI.
The news of the week that triggered Malik’s thoughts: Doomsday prepper billionaires Elon Musk and Sam Altman are no longer BFFs.
Briefly, Elon sued OpenAI and OpenAI responded, “I’m rubber, you’re glue; whatever you say bounces off me and stick to you,” or words to that effect, which was so exciting that it led former software developer Marc Andreessen (Mosaic browser) and former business plan writer Vinod Kholsa (Daisy Systems, Sun Microsystems) to get into a big fight on a formerly widely-used social media platform (deadname: Twitter).
Less briefly: The nub of the issue seems to be that OpenAI was founded with certain explicit altruistic and open intentions and a non-profit status, and it quickly created a for-profit subsidiary, more recently partnering with very-for-profit Microsoft. This is surely a change of direction, but Musk charges that it is a breach of contract and has filed a lawsuit, which is something that anyone can do. OpenAI retaliated by revealing messages from Musk that he probably considered confidential and that show him promoting this same for-profit path for the company. Well, minus the Microsoft part. Gotcha!
Meanwhile, Altman is back on the OpenAI board, and Musk has shifted his focus to insulting a woman for her philanthropy, which may or may not have anything to do with her ex-husband bumping Musk out of his position on the rich person leaderboard. Meanwhile, quoting Kenan Malik, “our concern should not be that machines may one day exercise power over humans but that they already work in ways that reinforce inequalities and injustices, providing tools by which those in power can consolidate their authority.”
In other AI news…
Opus Won
Benchmarks for software used to look a lot different. A lot of concern over FLOPS. These days the benchmarks tests for AI software are things like undergraduate level knowledge, graduate level reasoning, grade school math, math problem solving, multilingual math, code, reasoning over text, and common knowledge. Those are the measures on which Anthropic’s new Claude 3 models are clobbering the competition (OpenAI, Mistral, Google DeepMind, Chinese models). (Although the latest GPT-4 model looks a bit stronger.) There are three different Claude 3 models (one named Opus), and you can choose among them based on how much intelligence you are willing to pay for.
Don’t Learn to Code
Nvidia’s CEO tells kids not to learn to code. Because, he says, AI will write your code for you. Well then what should you learn instead? How about prompt engineering?
Blogger Ethan Mollick has been studying the craft of writing effective AI prompts and has come up with some odd results. Like, you’ll get the best results if in writing your prompt, you pretend to be in a Star Trek episode or a political thriller. Like:
“You have been hired by important higher-ups to solve this math problem. The life of a president’s advisor hangs in the balance. You must now concentrate your brain at all costs and use all of your mathematical genius to solve this problem…”
He also has some more obvious tips, like adding content to your prompts. In the long run, he thinks, good prompting won’t matter, because AI will be just so good at discerning your intentions. But for now, it’s a skill worth developing. Ethan is even building a database of good prompts. And he’s not alone. Want some Midjourney prompts for generating alien images?
If you already know how to code and have been making a career of it, I wouldn’t be too worried. I suspect that there will continue to be jobs for programmers for some time. One job that the AI boom is going to generate is UX work. AI is technology, and the user interfaces to that technology leave a lot to be desired.
But I guess the one sure thing we can say about AI is…
AI Changes Everything
Can AI solve science? That’s the provocative question Stephen Wolfram is asking. It’s the entry point for a deep dive into just what AI might be able to do and might not — regarding scientific research. As always with Wolfram’s writing, it’s deep, clear, and interesting.
Blogger Hiren Dhaduk of The AI Edge reports that Google’s ScreenAI can ‘see’ graphics like humans do, and describes how AI ‘worms’ can navigate networks of linked AI agents.
AI Supremacy blogger Michael Spencer and Humanity Redefined blogger Conrad Gray describe progress in connecting brains to computers.
And Andrew Maynard writes about the significance of the absence of AI in Villeneuve’s Dune.
Image of the Week
It’s a tree.
BEFORE YOU GO…
The Pragmatic Bookshelf
And now a plug for my day job.Got an idea for a technical book? I work with the Pragmatic Bookshelf, and I suggest you consider sending them a proposal. The benefits of publishing with the Prags include:
A global platform to increase your audience and extend your message;
an experienced publishing partner who makes even the hardest parts easier;
your own personal team of editorial and publishing experts;
your work available everywhere — from local retailers to Amazon and beyond;
input and advice from leaders in the tech industry;
membership in an instantly recognizable and respected community;
and a royalty of up to 50% across the board — books, translations, licensed content, and everything else.
Blogroll
AI Supremacy
Ahead of AI
Mark Watson’s AI Books and Blog
Kent Beck’s advice for geeks
When We Were Trekkies
Tales from the Jar Side
Geena Davis Institute on Gender in Media
Bookshop.org
New York Review of Books
Pragmatic Bookshelf
ICYMI
Thanks for reading. You can read all the back issues of Swaine’s World at my blog home.
Coming Attractions
Just more of the same: explorations in generative artificial intelligence, tech news, writing advice, book recommendations, and Dirt Road Diaries.