AI's Productivity Puzzles
Working at OpenAI • Non-WebKit iOS Browsers • Grok AI Bots • Emmy Noms • Meta's Open Source AI Path • NVIDIA's China Trade Deal Win • Zelda Casting!
While linking to the report below about questions of actual productivity gains with AI coding tools (at least for a subset of seasoned developers), I was reminded of this column from a month ago in The Financial Times that I had been meaning to link to. In "AI alone cannot solve the productivity puzzle", Carl Benedikt Frey, an author and associate professor at Oxford university, makes the case that efficiency gains alone may not be enough to solve productivity issues – something the world has been grappling with despite the constant rise of newer, better technologies.
Certainly an argument can be made that AI is both a different and potentially far more profound technological breakthrough than others in the past couple of decades – perhaps even the iPhone! – but there's also a case to be made that freeing up time, long cited as one key future benefit of AI, doesn't actually directly translate into productivity gains. I think that's fine if that time is filled with other endeavors, including more leisure/friend/family time. But there's a world in which it's also akin to the situation where you build more freeway lanes to free up traffic – and what actually happens is that more traffic just appears. Induced demand.
Anyway, not a fully-fledged thought I have myself yet, but Benedikt Frey's column is worth the quick read. Though I have thought about it in the context of – what else – email. As I think one of the first productivity changes of AI may turn email from your quasi to-do list into an actual to-do list, where agents actually do the emailing on your behalf (to other agents in many cases) and then collect any actual follow-up items you have to take action on. He briefly touches on email and the notion of statistical consensus and how that could be a major problem going forward with AI as well.
• Listening to "Don't Dream It's Over" covered by The Head and the Heart
• Written on an M4 MacBook Air 💻
• Sent from London, England 🏴 Notebook
💨 A Year at OpenAI
An interesting and easy-to-digest read on what it was like to be working inside of OpenAI over the past year. Spoiler: despite their influence, media mentions, and valuation, OpenAI is still very much a startup. Mostly in ways that are very good and have helped them stay ahead of the competition, but obviously some clearly chaotic downsides (which he doesn't really go into). Interestingly, he notes just how much Meta talent and influence there is at the company now (sort of counter to the pipeline the other way you keep hearing about). He sees the race to AGI as a three horse one: OpenAI, Anthropic, and Google. And that alone is perhaps why Meta is now doing what they're doing... Watching Meta try to recreate such a culture and environment inside of a nearly $2T company will be interesting, starting with "everything, and I mean everything, runs on Slack. There is no email." [Calvin French-Owen]
📲 Non-WebKit iPhone Browsers
Almost a year and a half since the DMA ruled that Apple couldn't restrict iOS to allowing web browsers with only their engine, and... crickets. Why? Obviously because the DMA only applies to the EU and so unless you want to build and maintain two different apps with two different code bases, you're going to still default to what the US market uses. Beyond regulatory issues, I wonder if this isn't going to be a problem as we start to enter the age of the newfangled AI web browsers – all of which are currently building on top of Chromium. But I'm sure Apple will have an answer in a few years. [Verge]
🤖 Grok AI Companions
While there have been a few startups focused on this use case, I'm mainly surprised that none of the other major AI players have tried to anthropomorphize their bots yet with avatars. Microsoft has clearly been thinking about it – yes, with Clippy – but the closest we've gotten is perhaps the "Santa Mode" for ChatGPT over the holidays (which was just a special Santa voice and some new effects, but actual animated characters). Oh yes, and Meta's dumb AI avatars (RIP 2024 - 2024). Elon is clearly happy to push boundaries – "Ani" and even "Bad Rudi" are certainly doing that – and it seems to be working with perhaps their first viral moment (well, maybe aside from the other really bad ones) with error loading messages galore. So... expect more of this, I imagine. [Platformer 🔒]
📺 The Emmy Nominations
The fact that Warner Bros. got 142 of them and Netflix got 121 suggests that there are just far too many nominations – there are something like 1,000 total each year now – for such nominations to actually be all that meaningful. Still, there can be interesting trends that point to overall vibes of how the studios are doing content-wise relative to their peers. Max/HBO/HBO Max is clearly sitting pretty. As is Netflix with more nominations than last year despite finishing second overall. (Disney technically had 128 noms but that's sort of cheating since they have so many different content arms.) The other big winner was clearly Apple TV+, which not only hit a new high in overall nominations (79) but had the show with the most overall nominations (Severance) while The Studio had the most for a comedy (23). The TV division is clearly humming, and now the movie division finally has some level of footing with F1. Now just imagine if Apple acquired Warner Bros... I can count 221 reasons to do that here – how would you like to have 100 more nominations than Netflix? (Aside: Andor was robbed with a "mere" 14 noms.) [Deadline]
Spyglass
🤲 Open Source AI was the Path Forward
...until it wasn't for Meta
Loose Leaf
What if AI coding tools are actually hurting productivity? One study suggests that could be the case at least with "experienced developers working on mature projects" – which also sort of makes sense for that specific cohort? The opposite can still be true for less experienced devs and/or less mature projects, of course. [Second Thoughts]
Speaking of, Amazon's new 'Kiro' AI software development tool is less about coding and more about using agents to assist developers in doing everything else they need to do around their code – going from "vibe coding to viable code" as they put it. While it's buried in the FAQ, Kiro is built on top of Claude Sonnet, with other models coming soon. [GeekWire]
Meanwhile, Apple doing anything to make it easier for developers to work with NVIDIA's CUDA seems like a good thing for AI development, even if it's just an early project to assist with exporting... [AppleInsider]
The rights to the Fyre Festival brand were sold on eBay. The cost? $245k. Which will barely make a dent in Billy McFarland's debts. [TC]
A look inside China's government-led push to win in AI. One key: the use of open source AI from DeepSeek and others as a form of "soft power" spreading influence in the space around the world. [NYT]
Worth noting just as Meta may be backing away from their own lead in "open source" AI (still, they'll undoubtedly open source some of the older, smaller models just like everyone else is moving towards in the US).
On the government support-front, the US keeps doing these sort of piecemeal, state-focused deals (which, of course, are always political), such as in Pennsylvania (though this one seemingly makes sense given the proximity to cheap natural gas for power needs). [NYT]
Mistral's open source "Voxtral" AI audio model sounds interesting... [TC]
The BBC may be about to outsource a lot of their internal work on products and tools such as iPlayer to "Big Tech" in order to cut costs. [Guardian]
Two more OpenAI researchers seem to be jumping over to Meta's "Superintelligence" ship, Jason Wei and Hyung Won Chung sort of sound like a package deal as both came over from Google originally and worked on the same projects – reasoning, deep research, etc – at OpenAI. No word on comp, though presumably it would leak if they got the "Godfather" offers... [Wired]
Legendary would inch closer to becoming an a full-fledged studio if they were to acquire Lionsgate, with its own theatrical distribution arm (big topic these days). Undoubtedly, Dune: Part 3 as well as any Minecraft Movie sequel would still be distributed by Warner Bros though... [Bloomberg 🔒]
Internal Apple documents seem to confirm the color tweaks to the iPhone lineup this year – including, yes, a Copper/Orange iPhone 17 Pro model. Well, technically, "Papaya" per Pantone. We'll see what Apple markets it as – "Sunbeam"? [Macworld]
I Quote…
"We put that in the trade deal with the magnets."
– US Commerce Secretary Howard Lutnick, telling reporters very directly why the restrictions on NVIDIA AI chips to China were lifted. Lutnick loves to give goofy quotes, but it's all about the rare-Earth materials which China was always going to use as a counter-measure against the chip ban.
To be clear, it's only variations of the less powerful H20 chips that are now legal for Chinese import, but that alone obviously super-charged NVIDIA's stock once again. Just days after it became the first $4T company, they're now almost at $4.2T – notably a full trillion bigger than Apple, nearly a half trillion bigger than Microsoft.
I Spy...
Zelda and Link. Shigeru Miyamoto has revealed the actors chosen to play the two roles in the live-action movie version of The Legend of Zelda due in May 2027. They're both English and both young. Bo Bragason (21) will be Princess Zelda and Benjamin Evan Ainsworth (16) will be Link. Oh boy I hope Nintendo and director Wes Ball get this right...
