đ§ Neural Dispatch: AIâs had a shabby week, featuring Meta and OpenAI
The biggest AI developments, decoded. October 1, 2025.
Hello!
ALGORITHM
This week, we chat about how misspending a few hundred billion dollars is not something the AI billionaires would spend sleepless nights over, why AI in medicine is still largely a bumbling sham based on shortcuts, and OpenAI may find itself in a slightly inconvenient position now that Microsoft may have begun the process of decoupling.
Oops, just misspent a few hundred billion dollars
Sometime ago, Meta CEO Mark Zuckerberg remarked that Meta could âend up misspending a couple of hundred billion dollarsâ in this AI race. Superintelligence is the dream. Right on cue, the Vibes layer in the Meta AI app, which the company calls âa new way to discover, create, and share AI videosâ. And youâll find a how-to in the âPROMPTâ section later, since many of you love to waste time and subscription money to trick your father with fake videos. How many new ways to create AI videos are launched literally everyday, Iâve lost count. Nevertheless, all Iâll say is this â AI was supposed to be about intelligence (PhD-level too, something AI bros love to drop into almost every conversation), solve real world problems and assist humans to be better at what they do. Yet, here we are, spending billions of dollars to make (yet another) app that churns out cat videos, and what looks like Bigfoot walking in the rain with an umbrella. When Meta says âweâre working on even more powerful creation tools and models with a number of talented visual artists and creators,â it drives me to uncontrollable anger. There is nothing creative or powerful about this. Simply misspending a hundred billion dollars in a world thatâs still struggling to fix hunger, homelessness and disease, just shows how billionaires hoard money for their whims. Those whims are NOT real world problems.
AI, medicine and an all-time low trust level
What was OpenAIâs Sam Altman self-aggrandising about just the other day, that he needs a few more billion dollars and a few more GPUs, if AI has to cure cancer? Turns out, a Microsoft Research, Health and Life Sciences paper titled âThe Illusion of Readiness: Stress Testing Large Frontier Models on Multimodal Medical Benchmarksâ even though current medicine focused AI models may be made to look good in certain benchmarks, there is little to suggest these models can handle real world medical reasoning. âHealth AI has a credibility problem. Models like GPT-5 can ace medical exams and top multimodal benchmarks yet still falter on simpler tasks, such as preserving their answers when answer choices are shuffled, or justifying their predictions with medically sound reasoning,â the paper doesnât mince words. One of the paperâs key findings among many, is that models retain most of their original accuracy even when images were removed, even on questions that require vision, which signals theyâre using shortcuts over real understanding. Thereâs fabricated reasoning too â âModels trained to âthink step by step often paired confident rationales with incorrect logic producing medically sound explanations for wrong answers, or correct answers supported by fabricated reasoning.â Case in point, OpenAIâs GPT-5. Weâre really going nowhere with the hype.
Microsoft, Anthropic and an OpenAI untangling
This ought to worry Sam Altman and OpenAI, more so because their $100 billion financing from Nvidia was found out to be exactly what it is, a desperate hype pivot. A few days later, Microsoft announced a partnership with Anthropic to make Claudeâs models available to Microsoft 365 Copilot users. Specifically, enterprise subscribers, starting with Microsoftâs Researcher agent and Copilot Studio, with Claude Sonnet 4 and Opus 4.1 sitting alongside existing model options. This is the first big move away from an exclusivity that had defined OpenAI and Microsoftâs relationship till date. OpenAI is no longer in prime position as the sole AI foundation for Microsoftâs products. Could this be the first sign of realisation on Microsoftâs part? Suddenly, another $100 billion, all the compute and all of the worldâs fresh water and electricity, may provide some solace.
PROMPT
Not that you needed yet one more app to âcreateâ AI debris thatâll litter social media timelines, but hereâs still one anyway. Meta AIâs Vibes feed.
Meta has unveiled âVibesâ as part of its Meta AI app, likely pouring in what must have been a rather thick stack of dollars, into solving humanityâs most pressing problem â not having enough AI-generated cat videos on social media feeds. This new feature lets users generate AI videos from text promptsâbecause apparently, there werenât enough generative AI apps already. Where will you find this? Update and open the Meta AI app on your phone, and generate videos of a human with a pet lion, or a diner above the clouds, and a dog driving a car in sunny Miami, will hit you in the face. Youâve little choice in the matter.
A rather predictable pitch too, a âcan bring your ideas to life with new AI visual creation tools or remix an existing video by adding music or changing the style to make it your ownâ. In practice, itâs a sophisticated and most likely unnecessary tool thatâll flood social feeds with even more ephemeral, irrelevant content. If it is indeed somehow impressive, it is impressive to what end? Weâre essentially celebrating our ability to generate infinite variations of âmy coffee cup but make it animeâ while the money poured into Metaâs computational resources could be solving actual problems. The real revelation isnât the technologyâitâs what it says about Metaâs vision for social mediaâs future.
Mark my words, Vibes doesnât enhance human creativity â it also replaces the creative process with prompt engineering. If the next generation has no painting, videography or visual storytelling skills, donât blame them (and certainly donât blame the parents). The AI bros have taken it upon themselves, to make the world a dumb(er) place.
THINKING
âI canât tell you whatâs going on at AT&T and Verizon or any place else. I can tell you T-Mobileâs iPhone sales are at all time record highs. We just had the biggest iPhone weekend. Weâre up double digit from that launchâ - Mike Sievert, President and CEO of T-Mobile, after the iPhone 17 launch weekend.
The context: AI doesnât sell phones, no matter what the Android enthusiasts may want you to believe. The latest iPhones prove that. During the iPhone 17 keynote, from which emerged the iPhone 17, iPhone Air and the iPhone 17 Pro series, Apple barely mentioned AI or Apple Intelligence. Whatâs selling the new iPhones (in India too, thereâs a fair backlog of pre-orders and early adopters) centers around generational upgrades, camera improvements, the Airâs ultra-slim design, and the new colours.
Iâll give you an example. An attempt to preorder an iPhone 17 on the morning of September 29, generates an estimated delivery window between October 13 to October 18. Thatâs indicative of the backlogs in place, due to demand. You could argue that operator deals and bundles may be the reason for this bullish outlook by T-Mobile, Sievertâs quote about T-Mobileâs record-breaking iPhone 17 sales is nevertheless compelling evidence that utility still drives smartphone sales for Apple, even in the age of AI. The perspective, the iPhone 17 shopping cart observations, in India where there are no operator bundles to bring down cost.
A reality check: The iPhoneâs original success, something Apple has built on year after year, came from solving obvious problems â it put the internet in your pocket and made smartphones a thing. In contrast, todayâs AI features solve problems most people donât know they have. AI can summarise your notifications and rewrite your texts, but when did anyoneâs life meaningfully improve from having their texts rewritten by AI?
This disconnect reveals an uncomfortable truth for Silicon Valley. Consumers are remarkably clear-eyed about AIâs actual value in their daily lives. While AI executives breathlessly describe their models as revolutionary, ones that could generate a better email response or edit an image beyond recognition, most consumers hardly find it worth a $1,200 upgrade. The lesson here is brutal but important â AI doesnât sell smartphones, genuine utility and a sense of upgrades does. Better cameras, faster chips, and improved software, still trump fiction. In this case, polished experience has always been Appleâs forte.
Neural Dispatch is your weekly guide to the rapidly evolving landscape of artificial intelligence. Each edition delivers curated insights on breakthrough technologies, practical applications, and strategic implications shaping our digital future.