Signal: iPhone RAM 📧
Two quick thoughts on iPhone unveil day. First, while I still see it getting quite a bit of hate online, I actually really like iOS 26 – yes, even the liquid glassiness of it. But mainly I love how fast it is. Little animation speed ups really pay off. (I like the new OSes less on iPad and Mac, but they're all fine.) Second, while I said yesterday that the thing I'm most looking forward to are the AirPods Pro 3, I think it's actually the (long rumored) more RAM in the iPhone. Perhaps 12GB up from 8GB.
Even beyond the AI equations, I still run into issues with apps quitting on me when I seem to "overdo" it. I wish I could pay more money for more RAM, but Apple would undoubtedly think that would make the device seem too much like a computer caught in a spec race (resource issues, etc). At least until they need more margin.
I Wrote...
Thinking through the position in which OpenAI finds themselves feeling the need to burn $115B by the end of the decade. These are wild numbers. Bonkers. But they also almost make sense if you just back into what is happening... The next big race in AI may be to control the whole stack, as it were, because it may be the only way to make the model work to keep making models...
M.G. Siegler
Others Wrote...
🤔 Why Language Models Hallucinate
I found this to be an easy-to-read overview of research OpenAI has put out there about the "hallucination" phenomenon with AI. High level: the models to date have rewarded guessing over admitting uncertainty. Fixing it requires changing the incentive structure – something which human teachers learned long ago with regard to human students taking human tests. It clearly won't be so simple – and we obviously don't want to go too far – but if you penalize being wrong while giving some credit for admitting when you don't know something, you're at least heading in the right direction. AI bots, they're just like us... [OpenAI]
🇺🇸 All the President’s Tech CEOs
I get the argument that these CEOs of major public companies may have felt like they had no choice but to participate in such an event – and they probably need to be around the literal table when it comes to discussing the future of AI – but let's not conflate that with deeming the footage coming out of the around-the-horn praise-the-President-before-you-get-dinner exercise as "normal". It was just a hugely pathetic and embarrassing literal show of fealty, as Brian Barrett rightly notes in his piece. How will history look back upon all of this kissing of the ring, one wonders? Probably not too kindly, though that hardly matters for the bottom lines of these businesses right now. But avoiding tariffs does! Mark my words, at some point in the not-too-distant future as Trump glides into lame-duck status, we'll start getting some fun attempts to change the narrative around this moment in time. [Wired 🔒]
Some Notes...
- In a twist, the judge at first rejected Anthropic's proposed settlement over illegally obtaining books for AI training before later saying he was postponing approving any deal pending a new clarifying submission. [Bloomberg Law]
- He's worried that too many of the details are to be determined at a later date and that many would-be claimants in the class action will "get the shaft".
- He said he also felt "misled" by the way the settlement has been framed...
- More bumps in the flight of Microsoft Copilot as the company considers simplified business pricing and yet more branding techniques with 'Agent 365' – which sounds like an evil James Bond clone. [Verge]
- Might today's Apple event also feature the M5 iPad Pro? Normally such things happen in an October event but there's both a (apparently reputable) Xitter rumor, and Ryan Christoffel recalled that this will be 10 years to the day that the first iPad Pro was announced... [9to5Mac]
Below, members of The Inner Ring will find thoughts on:
• 'Startup Energy' at Snap
• SpaceX Buying Spectrum (CYBERPHONE?)
• Google's 'Open Web' Gaffe
• OpenAI Ongoing Melodramas
• Howard Stern Lives
• Vision Pro in Black
• and more...