Apple uses AI to build its foldable iPhone

So imagine you are standing on a massive factory floor a machine is scanning this microscopic piece of titanium and it finds a physical flaw something totally invisible to the human eye It then just instantly 3D prints a custom correction to smooth out that specific flaw Now imagine that machine doing this millions of times a day on a high speed assembly line That is honestly hard to even picture It really is Today we are looking at how Apple just quietly turned artificial intelligence into a blue collar factory worker I'm Aya I'm Theo Welcome to Superpower Daily This is your daily AI news conversation And we have

a massive hardware story leading things off today You really do Apple is taking a really surprising new approach to manufacturing its very first foldable phone because they are deploying AI directly on the factory floor instead of just putting it in the software Yeah Apple officially introduced the iPhone duo This is obviously the company's first foldable device Right Pre orders open on October 16th Regular sales will begin on October 23 And the headlines are obviously focusing on the physical design because it's a foldable iPhone But the core story here is actually about how they are building this specific hardware Apple hardware chief Johnny Sruji detailed this

entirely new production method They are using an AI guided manufacturing process Right And this system uses these advanced algorithms basically to pair each specific hinge with the exact housing that fits it best Wait So it pairs them up individually Yes This incredibly precise pairing happens before the device is even fully assembled That is fascinating because mass manufacturing always involves tiny physical variations Right Exactly That is the fundamental hardware problem they are solving here A standard assembly line just grabs a random hinge from a bin and then it attaches that hinge to a random housing from another bin And that traditional method creates these microscopic gaps

Which creates hidden tension points You got it So Apple is using AI to analyze the exact physical dimensions of every single manufactured part The algorithm then finds the mathematically perfect match between a specific hinge and a specific housing I mean that requires just an extreme level of precision Especially since the hinge on the iPhone Duo is so complex It really is Apple says the hinge contains over 100 individual components Which is wild Yeah And those components have multiple distinct jobs They control the smooth opening and closing of the device Right They also provide rigid physical support to the dead center of that flexible display Because

it has to hold the screen perfectly flat when the phone is fully open Exactly And that moving joint is unusually consequential for a consumer device Because a standard smartphone is basically just a solid brick Right It has no moving exterior parts at all But a foldable phone relies entirely on its mechanical hinge It is not just a peripheral design detail No Not at all The hinge manages the central physical transformation of the device over its entire lifespan Every single millimeter matters immensely Because any internal physical stress will eventually damage that really delicate screen So Apple is tackling those physical variations with a pretty fascinating inspection

process They're actively scanning every single completed hinge unit on the line Yeah They use a tool called a confocal laser for this specific step A confocal laser Right The laser maps the precise surface topology of every single unit Okay So how does that actually work Well a confocal laser allows for incredible microscopic depth profiling It uses a tiny pinhole to block out of focus light Oh Yeah This lets the sensor measure the tiny peaks and valleys on the surface of the hinge components The residual waviness Right The waviness left behind by the industrial milling and shaping processes Because even the most advanced titanium or steel

components will have these microscopic imperfections Of course So the laser creates a flawless digital map of those highly specific physical flaws And the next step is where the process becomes truly futuristic Yes Apple takes that unique laser scan data and then they use 3D printing directly on the factory line Which is just crazy to think about They print up to 25 micro layers of a custom photopolymer right onto the hinge metal This smooths out any of that residual waviness that the laser detected Right And a photopolymer is a specialized liquid material It hardens instantly into a solid plastic when exposed to specific frequencies of ultraviolet

light So printing 25 micro layers allows Apple to level the surface of the hinge with extreme volumetric accuracy They're essentially just filling in the microscopic metal valleys Exactly They create a perfectly smooth structural foundation And this foundation is based entirely on the unique topological profile of that specific hinge You know this makes me think of a master tailor How so Well imagine a bespoke tailor adjusting a custom suit to the exact millimeter for a specific client Right They measure every contour of the client's body They adjust the fabric to drape perfectly without any tension anywhere Yeah Now imagine that tailor doing this millions of times

a day on a high speed assembly line That is a great analogy It feels like a major paradigm shift for how physical hardware is made I mean we are so used to AI being a software feature you talk to Like Siri or chat GPT Exactly You have to wonder if this is a totally new era where AI is an invisible factory worker actively calibrating the physical device That is exactly the industrial transition we are seeing here This AI is not running a helpful feature for the end user It is governing the physical fit during mass manufacturing which completely changes the narrative around foldable device durability

Because early foldable phones relied entirely on the flexibility of the screen material itself Right And Apple is shifting the focus to precision production mechanics The AI calibration and those printed photopolymer layers are meant to correct physical geometry They're solving the stress problem before the phone ever reaches a buyer Exactly But the hinge is really just one part of their durability strategy The folding display itself involves a highly complex multi layer lamination process Yeah They combine high strength glass layers above and below the actual folding organic light emitting diode panel And they also use highly custom adhesives between these structural layers Right Because those adhesives are

chemically designed to let the structural layers slide past one another slightly during a fold Which makes sense When you fold a thick stack of rigid materials the inner layer compresses And the outer layer stretches Right This creates immense shear stress Exactly The custom adhesive relieves that dangerous bending stress It basically prevents the flexible display from delaminating Or cracking over thousands of folding cycles The exterior cover layer also features a really unique nano textured finish Apple says this finish is designed to cut down on glare and surface reflections Yeah And the cover is also made from a custom polymer formulation Apple claims this specific polymer is

up to 40 stiffer than comparable materials used elsewhere in the consumer tech industry And the exterior cover over the hinge mechanism is also entirely 3D printed So you can see a very deliberate material strategy forming here Apple is not relying on a single miracle material to solve the foldable durability problem No This product launch is built around several distinct material and manufacturing interventions They are all working together simultaneously But we do have to acknowledge a very important limitation here Absolutely This is a highly notable production use of artificial intelligence It is a brilliant manufacturing achievement For sure However it is not definitive proof that the

phone will survive long term real world use Because the real world is incredibly harsh on moving consumer electronics Yes Foldable phones face repeated unpredictable mechanical stress They endure strange torsion from being opened rapidly with one hand Oh yeah I do that all the time Right And they face constant particle ingress from fine dust and pocket lint Which is awful for moving parts Exactly Conventional solid phones completely avoid these daily structural stresses Apple's new manufacturing process explains how they are trying to limit these initial weaknesses in the factory Right But it does not establish how the iPhone duo will actually hold up after three years of

heavy careless use The long term physical reality is the absolute key metric you should watch next We need to see if Apple's extreme precision manufacturing actually translates into resistance to everyday wear and tear Because a perfectly calibrated hinge might still degrade rapidly if abrasive dust gets inside the delicate mechanism Exactly We will only know the real answer after millions of people spend years folding and unfolding these devices in the real world Okay Moving away from AI building hardware in the factory we are now looking at AI reshaping personal software on your wrist Yes Apple just unveiled the Apple Watch Series 12 and the Apple Watch

Ultra 4 And these new devices are actively turning everyday conversations into searchable text Which represents a major functional shift in wearable computing It really does Apple introduced WatchOS 27 alongside the new hardware The update brings extensive Apple intelligence features directly to the wrist Right Because the new watches run on the upgraded S11 silicon chip Exactly And the most significant architectural change is the evolution of Siri Siri is definitely changing Yeah Siri is moving away from just being a reactive assistant you ask deliberate questions to It is turning into a passive tool Right It automatically recovers missed spoken details from your immediate environment Which is kind

of wild It is There are two main ways to interact with this new memory system The first feature is called Live Rewind Okay And it is designed for the immediate retrieval of very recent audio You simply double press the digital crown on the side of the watch Right The screen then instantly displays up to 15 seconds of the preceding conversation as a raw text snippet So you can read that text snippet directly on the watch face You can also save the snippet for later review in the newly redesigned Siri application Right You even have the option to ask Siri a contextual follow up question about

that specific snippet That is so useful It really is The second feature is called Siri Recap This tool takes a much broader view of your daily social interactions Yeah After a long conversation naturally ends Siri Recap uses ambient listening data to generate a structured summary It produces a clear title and a bulleted list of key points from the entire exchange And Apple demonstrated the power of this system with a real time translation and transcription example I saw that It was pretty impressive Right They showed a scenario involving a loud busy restaurant A user wearing the new watch was able to scroll backward through a block

of generated text And the text contained the exact details of the daily specials that the server had just described verbally Exactly This concept is genuinely wild for everyday social dynamics It really is I mean imagine being at a loud dinner party You zone out for a few seconds because you are tired We have all been there Exactly And your partner says something important across the table Instead of asking them to repeat themselves you just discreetly double press your watch crown And you immediately read what they just said Right It solves a very real human problem But it feels slightly unsettling It does Because it fundamentally

alters the basic social contract of human attention People generally assume their casual words vanish into the air Right This technology captures those passing words permanently And the specific hardware upgrades make this constant environmental monitoring possible The battery life must be a concern right Actually no The Ultra 4 maintains up to 50 hours of battery life while running these models Wow It also continues pulling background heart rate readings every five seconds That is a lot of processing power It is And the Series 12 maintains its standard all day battery life despite the heavy new AI features Though Apple is completely removing the classic walkie talkie feature

in this specific software update Right But the broader implication here is immense We are moving rapidly into an era of ambient conversation memory Our personal devices will soon remember absolutely everything we hear Which brings us to a massive limitation regarding fundamental consumer trust Yes The privacy boundary is the central issue here Apple explicitly claims that its audio intelligence features do not create or store any raw audio recordings Which honestly feels like a direct contradiction It kind of does I mean Siri Recap clearly relies on ambient listening to gather the necessary conversational context Yeah It physically has to listen to generate the summaries Right The critical

distinction lies in how the audio data is handled locally on the S11 chip Okay Break that down for us The watch is constantly filling a very short temporary audio memory buffer The on device AI model transcribes that buffered audio into text in real time The raw audio data is then instantly overwritten and permanently destroyed by the next incoming sound So it never saves the actual audio file Exactly The text transcript is the only thing that gets saved or summarized Apple is arguing that transcribing audio locally in volatile memory is legally and practically different from recording permanent audio files This subtle technical distinction is exactly what

you should watch next We need to see whether Apple's strict no recording boundary is actually clear and robust enough because the general public has to actually trust ambient conversation memory living constantly on their bodies Exactly A deeply technical explanation about volatile audio buffers might not convince someone They might just feel like they're being secretly recorded at dinner Which is a very fair feeling Meanwhile if personal AI is changing our individual social contracts enterprise AI is about to rewrite the entire economic contract It really is Anthropic has released a highly detailed interactive 2030 US economy model And this mathematical tool reveals a stark divide in who

actually gains from the ongoing AI boom So Anthropic previously published a deeply technical research report It was called Economic Scenarios for Transformative AI Right They have now turned that dense academic research into an accessible interactive tool It is formally called the AI Economy Explorer Which is super interesting to play around with It is Users can tweak core macroeconomic variables regarding AI deployment You can adjust the expected AI capability the corporate adoption speed and the overall system autonomy And then the model recalculates everything The model then mathematically calculates the downstream effects on GDP employment wages and labor income distribution in the year 2030 The model operates

by breaking down the entire modern economy at the microscopic task level Which I found fascinating Yeah It does not look at whole jobs like a lawyer or an accountant Right It represents occupations as massive bundles of individual tasks Exactly And an AI system can interact with a specific human task in four different ways It can leave the task completely unchanged It can augment a human worker who is performing the task Right It can automate the task completely Or it can create an entirely new kind of task for a human to do And the economic output varies wildly based on how you weight those four specific

interactions It really does The model produces several conditional scenarios rather than one fixed prediction The first major scenario is called the Substantial Case Right This specific path assumes AI can successfully perform about half of all current knowledge work It assumes the AI operates mostly autonomously Yeah However it assumes corporate adoption remains highly incomplete and uneven across the broader economy And the final results of the Substantial Case are deeply mixed Total U S GDP ends up 8 3 higher than a model baseline with no AI intervention Which is a huge jump It is And labor receives 56 1 of that total economic output Right Interestingly despite

the massive GDP growth average knowledge worker wages remain essentially flat The financial gains simply do not flow to the workers operating the systems Which is slightly depressing But the model then presents the far more disruptive Extreme Case The Extreme Scenario is wild It really is It assumes AI becomes significantly more productive than human workers at the vast majority of It assumes nearly all of these complex tasks are performed autonomously by AI agents It also assumes the new AI economy creates almost no new knowledge tasks for human beings to perform And Anthropic openly states that reaching this extreme path would require highly advanced technology Well absolutely

It would likely require recursively self improving AI models These are models that write better code to make themselves smarter without human help Right It would also require incredibly rapid adoption across the entire corporate landscape The economic projections in this Extreme Case are frankly staggering They really are The model reaches a 44 4 trillion U S economy by the year 2030 Wow That specific figure is 32 4 above the modeled no AI baseline Annual economic growth reaches roughly 15 That rate of growth is fast enough to double the size of the entire U S economy every four and a half years Yes It represents an unprecedented

historical explosion of raw wealth generation The total economic pie gets massively bigger but I am completely shocked by the breakdown of who actually gets to eat that pie It is the most alarming part It really is Labor's overall share of the economic output drops sharply to 45 2 Capital owners end up taking a massive 54 8 of the generated wealth And the situation for individual workers is even more severe in this Extreme Case Knowledge worker wages actually fall by more than 10 across the board Overall unemployment rises significantly above typical recessionary levels Why does it spike so much This happens largely because people displaced from

automated occupations take a very long time to find new work This is a profound structural paradox It is The knowledge workers who largely built trained and adopted these tools end up with a much smaller slice of the economy If you work in software or marketing you might be building the very engine that shrinks your income The wealth concentrates heavily at the absolute top The owners of the corporate capital and the underlying AI infrastructure capture almost all the newly generated value Yeah The model highlights the intense economic friction of occupational switching In highly transformative scenarios knowledge workers are forced to move into jobs less exposed to

AI automation This often means transitioning into physical labor or in person service roles Which is a huge shift It is Wages might actually rise significantly for workers outside the knowledge sector Anthropic points out that faster AI driven design and civic permitting could spark a massive demand for physical construction Oh that makes sense Right However retraining a senior software engineer to manage a concrete construction site involves immense social and educational friction That severe friction drives up the national unemployment rate during the transition years Anthropic actually surveyed 10 980 Americans regarding their personal expectations for the near future And what did they find Well the typical respondent

gave answers that implied an outcome closely resembling this substantial scenario They expected a moderate GDP bump and roughly flat unemployment Okay However about 10 of the respondents gave answers perfectly aligned with the devastating extreme case That is telling The most important limitation of this interactive tool is its fundamental nature though It is a highly conditional map It is definitely not a settled prediction of the future It measures mathematical possibilities based strictly on the user's personal assumptions Of course Anthropic released this tool to make a very specific economic argument They want to prove mathematically that massive GDP growth does not automatically equal an improved economic position

for everyday workers Because a soaring stock market and a booming GDP can easily hide a collapsing middle class The concrete signal you should watch next is enterprise behavior We need to observe whether workplace AI use shifts toward fully autonomous task completion rather than just human assistance Right The transition from a helpful copilot to an independent autonomous agent is the central trigger for the extreme economic scenario This model forces a very hard look at the dangerous divergence between national growth and individual prosperity So pivoting slightly autonomous agents are not just an economic theory for the future No they are not They are being built right now

to defend our most critical infrastructure Open AI has officially built an internal system they are calling the defense factory Which sounds very intense It does They are using highly autonomous AI to continuously find test and fix software vulnerabilities across their networks Yeah Open AI recently formalized this complex automated security loop The system relies heavily on their advanced codex model Codex is specifically trained on billions of lines of code They combine codex with isolated internal development environments And what is the main goal here The main goal is to investigate suspected security flaws And actually prepare tested software patches without any human intervention This new factory model

grew out of a massive internal security sprint at Open AI Right They aggressively mobilized over 250 people across more than 100 different service areas within the company Wow The sprint produced some highly structured operational results They achieved a 90 6 accepted ownership routing rate Okay What does that mean in practice This means the AI successfully identified a bug and routed it to the correct human engineering team who formally accepted responsibility for fixing it That is a massive time saver The automated triage phase also proved highly effective for the teams Right The system found that 37 of the initial security findings were actually just duplicate reports

37 Yes And AI accurately filtering out duplicates saves human engineers thousands of hours of wasted investigation time The core technical premise of the defense factory relies on reproducible environments Right These are temporary completely isolated copies of live code system dependencies and active services The AI agents work exclusively inside these safe digital copies Which is super important It is They can test potential exploits and evaluate proposed patches without ever touching the highly sensitive live production systems And the code fixes generated during the internal sprint were entirely based on the Codex model Wow The AI agents generated the software patches and autonomously deployed them to the isolated

test environments And how did they perform The final rollback rate for these AI generated fixes was an incredibly tiny 0 53 That is almost nothing Exactly This metric indicates the AI patches were largely successful and did not inadvertently break existing system functionality You know this operates exactly like a biological immune system That is a great way to think about it A digital immune system identifies an invading virus or software vulnerability It synthesizes a complex antibody or a code patch It then tests that patch in a safe Petri dish before injecting it into the actual patient It is a continuous rapid loop of detection and healing

It is You have to wonder about autoimmune diseases though What happens when the AI hallucinates a patch that actually introduces a brand new vulnerability That is a huge concern What if the AI accidentally creates a hidden back door while trying to close an exposed front door That exact security risk is why OpenAI encountered significant operational bottlenecks Only 19 5 of the security findings were successfully reproduced at runtime in the isolated environments Why so low Because setting up perfectly consistent isolated environments is incredibly difficult in a massive tech company Production systems have thousands of hidden software dependencies Of course You cannot fully test an AI patch

if the test environment does not perfectly match the messy reality of the live server And the automated system also struggled heavily with the final deployment phase right Yes OpenAI quickly realized that just because a patch was successfully merged into the main code base it did not mean it was actually deployed to the live system Because code deployment gets delayed frequently Exactly Deployment pipelines fail under load Live servers require carefully timed rolling restarts The AI cannot simply declare victory because a human engineer accepted the proposed code merge This severe disconnect required OpenAI to expand their independent post deployment verification checks The loop has to include automated

confirmation that the vulnerability is actually closed on the live server The most important limitation of the defense factory is that system autonomy has a very strict boundary Human oversight remains entirely in place for any consequential changes to the main code base As it should The workflow relies heavily on processing small human reviewed batches of patches The AI agents only gain increased autonomy after their results consistently earn human trust over a long period of time This is a highly cautious but absolutely necessary approach to automated cybersecurity Right We should watch carefully how the specific factory model scales across the broader tech industry OpenAI noted that CloudFlare

RAMP and Google are currently exploring very similar automated security loops The key test going forward is the operational reliability of the test environments We need to watch whether isolated testing and automated deployment verification can remain reliable as these tech giants hand more control to AI agents Definitely The continuous defense loop is incredibly promising However the operational reality of managing complex global networks makes it very difficult to perfect Okay We are moving to the Quick Read section Google is actively adding official election information directly into Gemini and AI Search ahead of the 2026 U S midterms Which is a big deal Yes Users will be able

to easily find polling locations registration deadlines and other verified election details through Google's AI interfaces Google is sourcing this critical civic data from state and local governments They are also pulling verified data directly from Democracy Works And the Associated Press will provide all real time voting results Right Google is aggressively pairing this official data with sophisticated tools designed to help users identify altered campaign media One of those specific tools is called SynthiCovel It places imperceptible cryptographic watermarks inside AI generated images video and audio Which is so needed right now It is These invisible watermarks can be checked directly through Gemini Google Search and the Chrome

browser Google is also supporting C2PA content credentials as another powerful providence signal Right C2PA acts like a digital trail of custody for media files They are also offering a tool called Backstory to let users examine the exact origin and editing history of a specific file And the platform rules for political advertisers are incredibly strict Right Advertisers must legally verify their identities They must clearly disclose the use of synthetic or digitally altered content They also must include prominent paid for by notices on all their materials And Google and YouTube establish strict enforcement policies They will restrict features for minor violations They will completely terminate accounts for

major breaches Right They are also offering free enterprise grade cybersecurity protection to eligible campaigns and journalists Through Project Shield they have committed to maintaining strict political impartiality across all these efforts But there is an important caveat here regarding technical transparency Google clearly states it is rolling out added protections to limit misleading election related images Right However the company has not actually explained how those specific technical protections will work under the hood Which is a pretty big detail to leave out Next up Apple has officially set daily limits on its server backed AI features Yeah They published new usage terms outlining strict daily caps for complex

cloud tools This specifically affects Siri AI intelligent photo editing image playground and AFM3 cloud models used inside the Shortcuts app But these caps only apply to heavy features that rely on Apple's external cloud servers Features running locally on the device processor will continue working perfectly even if you hit a cloud cap Right Reaching a server limit might simply make that specific cloud tool unavailable until the daily limit resets Apple also noted they can unilaterally restrict throttle or suspend access for excessive automated fraudulent or otherwise unreasonable use The main caveat involves pricing and specific volume thresholds Apple has not published what the actual daily allowance limit

is for any of these features Exactly They also have not announced the exact price or the launch date for future paid expanded access Expanded server access will likely tie into higher iCloud Plus subscription tiers Those specific financial details remain totally unpublished We will have to wait and see on that one In other news Neon has officially released the first teaser for the film Artificial Oh wow The movie stars Andrew Garfield playing OpenAI CEO Sam Altman And the highly anticipated film is directed by Luca Guadagnino The teaser frames Altman making a highly ominous entrance He descends slowly into a dark basement filled with machine guns That

sounds intense It is He predicts the collapse of major industries and the fall of sovereign countries The historical anchor for the film is the abrupt five day ouster and rapid reinstatement of Altman during the November 2023 leadership crisis The historical scope of the film is actually much wider than that five day window The preview indicates the movie will also extensively cover the initial creation of OpenAI and the massive global launch of ChatGPT The project had a very troubled route to the screen Amazon MGM originally held the distribution rights but they abruptly dropped the film in June Interesting timing Yes This happened right after Amazon announced

a major corporate partnership with OpenAI Neon then quickly acquired the film It is scheduled to premiere at the New York Film Festival in October before a U S theatrical release on Christmas Day I will definitely be watching that We are moving to three takeaways from today Sounds good First Apple is redefining AI integration not just as conversational software you talk to but as an invisible calibration tool on the factory floor and an ambient memory engine on your wrist Second the macroeconomic modeling of AI is forcing a very hard look at the dangerous divergence between massive GDP growth and the rapidly shrinking share of income that

may go to knowledge workers Third the push for highly automated cybersecurity is showing incredible promise Its true bottleneck is the messy reality of configuring perfect test environments and ensuring code patches actually deploy One development to watch tomorrow is how global hardware supply chains react and adapt to Apple's extreme new AI driven manufacturing standards for foldables You can find more details at superpowerdaily com Thank you for listening We'll see you tomorrow