Developers run a fruit fly brain map in Minecraft

Welcome to The Signal from Superpower Daily with Maya and Theo Today we are looking at how a massive new mapping of a fruit fly's brain has unexpectedly become a viral sensation for developers Yeah it is a genuinely wild crossover We are watching this incredibly specialized data set just get hijacked by the open source software community Right and our mission today is to explore what happens when that hard boundary between biology and software completely collapses Exactly because if you want to understand where technology is heading next well you really need to look right at this intersection So this whole story starts with a massive data release

from Google Research and HHMI Janelia On September 3rd they published a data set called MalaCNS V1 0 Which you know it sounds like a routine software update Right it really does But it is actually a historical biological milestone They released a completely proofread wiring diagram of an adult male fruit fly's nervous system Which is just wild to think about It is I mean you have probably heard the term genome before Sure like mapping the DNA of an organism Exactly But this is a connectome A connectome maps the actual physical wiring of the nervous system Okay so it traces every single connection Yes between every single

neuron You can think of it as well the ultimate electrical schematic for a biological entity And the scale of this schematic is just almost impossible to visualize Oh absolutely This wiring diagram maps more than 166 000 neurons Right And between those neurons it maps 125 million synaptic connections Yeah the physical coverage of this map is what makes it such a huge breakthrough Because it is not just the brain right Right It covers the entire central brain It covers the complex visual regions And crucially you know it includes the ventral nerve cord Okay wait Let's stop right there What exactly is a ventral nerve cord Like

why should you care about that specific part of a fly Well if you think about the human body we have a brain and a spinal cord Right The spinal cord is basically the highway It takes the brain's commands and delivers them to the muscles So in an insect like a fruit fly the ventral nerve cord serves roughly that same purpose Oh wow Yeah It is the biological structure analogous to our spinal cord It translates thoughts into physical movement We should really put the sheer scale of this new map into perspective Yeah we definitely should Because Google released a female fruit fly connectome back in 2020

but that map only covered half of a brain Right This new male release is more than six times larger And that expansion is a massive technical achievement I mean reaching beyond the brain boundary into that ventral nerve cord it just changes everything Because it shows the whole pathway Exactly It extends the wiring diagram all the way down the pathways the brain uses to control the physical body It shows us how a decision actually becomes an action But getting to this point it was a brutal grueling process right It was You do not just snap a photo of a brain and get a map No not

at all The structures are far too small for normal light microscopes Well how do they even see them Well they have to use an electron microscope Scientists take a physical fruit fly brain and they slice it into thousands of layers Just tiny tiny slices Yes Each layer is thinner than a human hair Oh wow Then they hit those slices with a beam of electrons that create these incredibly detailed flat images And you end up with millions of these flat images Exactly I mean it is basically the world's most impossible jigsaw puzzle It really is You have to trace a single microscopic wire across thousands of

individual slices just to see where it goes That sounds miserable for a human to do Yeah No human team could do that manually in any reasonable time frame Right So artificial intelligence took over the heavy lifting The AI stitched together millions of these electron microscope images It mathematically predicted how the lines connected Yes It built three dimensional neural structures out of a massive stack of flat two dimensional images But the AI was not perfect right It made mistakes Right It definitely made mistakes So human experts at Genelia had to step in They had to manually verify and annotate the whole reconstruction Exactly They spent thousands

of hours essentially proofreading the AI's geometry And that human verification is what makes MEL CNS V1 0 so valuable Yes It is not just an AI guess It is a fully proofread validated representation of reality And once it was validated well the researchers made a very consequential decision They did They made the entire map available to anyone They released it through this open source visualization tool called NeuroGlancer Which was actually built by Google Right And it lets anyone with an Internet connection just browse this incredibly complex data set right in their web browser It is amazing You can zoom in on individual neurons Just like

you zoom in on a street in Google Maps And that open access I mean that was the catalyst for the viral sensation we are seeing now Absolutely The scientific intent was simply to provide an anatomical reference for biologists Right But software developers looked at this tool and you know they saw something entirely different They basically looked at a biological brain map and saw an empty playground Yes They recognized it as a highly complex digital substrate They started treating this wiring diagram as a new environment for software experimentation Right And that is exactly how we end up with the headlines Developers are downloading the exact structure

of these 166 700 neurons And they are using that structure to run viral software simulations Yes They are dropping this simulated neural network into video games Which is just crazy People are putting the fruit fly connectome into Minecraft Yes They are hooking it up to the monster AI in Doom They are piping its structure into Beat Saber I even read that someone tried using the structural complexity of this brain map to run Bitcoin trading software Yes I saw that too But you know the Minecraft project is probably the clearest example of how this actually works Right A report from the technology outlet 404 Media broke

down exactly what a developer did They took the full data set Yes And they mapped the simulated neural pathways directly to the game's internal logic systems Because in Minecraft there is this system called Redstone Right Which acts like digital electrical wire Exactly You can build logic gates with it So the developer essentially translated biological synapses into Redstone circuits Yes And the simulated neural activity actually controlled a virtual fly's movement within the blocky world of the game The visual of that is just striking It really is You have this rigid pixelated game environment But the entity moving around inside it is being driven by the chaotic

organic geometry of an actual insect brain And that specific Minecraft project it garnered millions of views in a single day Yes It went completely viral It is completely surreal to watch I mean a highly rigorous years long neuroscience project basically became a developer toy overnight Yes it did But we really need to take a step back here I want to push back on the viral hype for a second Please do The hype often obscures the reality with these things Right Is putting a fruit fly brain into Minecraft actually proving anything useful about biology That is a great question Or is it just you know a

very complicated very shiny tech demo That is literally the most essential question you could ask about this trend Yeah And to answer it we have to look closely at what is actually being simulated Okay Let's unpack this with an analogy Sure It feels like having the complete electrical wiring blueprint for a massive skyscraper A blueprint is a very good way to look at it Right So you can look at the blueprint and see where every single copper wire goes Yeah You can see every light switch You can see every server rack connection But you have absolutely no idea what kind of data is actually flowing

through the server cables Exactly You do not know if the lights are turned on or off You just have the structural map You do not have the electricity And that analogy perfectly illustrates the scientific limitation here We really have to establish a hard boundary This release is a connectome A connectome is purely a structural wiring diagram It is an anatomical map of the hardware Okay It is absolutely not a digital recreation of a living animal So it really is just the hardware with none of the software That is the crucial distinction When a developer hooks this map up to Minecraft the simulation entirely lacks biochemistry

Because real brains are messy Very In a real biological brain neurons do not just pass electrical signals They bathe in neurotransmitters Hormones wash over the system and change how all the switches react And a structural map captures none of that chemical reality None of it It lacks sentience It lacks any sort of conscious perception of the digital world it is placed in So generating computer code from a static structural diagram that does not equal a living mind Right An animated virtual creature in a video game is running on simulated binary logic It is not experiencing the world the way a biological carbon based organism does

That is a very serious caveat for you to keep in mind as a listener We have not uploaded a living fly to the internet No We basically just scanned its circuit board But you know if this does not create a living digital fly why does this map matter so much to actual neuroscientists Well the scientific value lies deeply in comparative data Remember Google mapped a female brain in 2020 Right Now researchers have matched maps They have a complete female connectome and a complete male connectome And both are available at synaptic resolution Exactly They can see every single connection point This allows biologists to study structural

differences directly Okay So they can overlay the two maps Yes They can look for anatomical variations that might be linked to specific courtship behaviors They can trace the exact pathways that govern aggression Oh wow They can examine vision processing taste reception and social behavior at a really granular structural level So if a male fly reacts differently to a visual stimulus than a female fly they can trace that difference down to the specific wire that routes the signal differently Yes That scientific utility is incredibly clear Yeah But the public reaction is what makes this specific moment so unique Because it just exploded online Right This release

crossed a massive cultural threshold It proved that complex esoteric anatomical data sets can be understood and repurposed by the general public Developers are essentially treating biology as open source material Yes They are taking millions of dollars of scientific research and treating it like a free software library on GitHub What is truly fascinating here is the collision of disciplines Neuroscientists built this map strictly to understand biological wiring but developers are using it to test computational boundaries Right They are asking what happens when you use biological complexity to drive digital systems Because biological brains are incredibly efficient at processing messy chaotic data Yes And our silicon computer

chips actually struggle with that So developers are trying to see if biological architecture can teach silicon new tricks They want to know if the physical layout of a brain is naturally better at solving certain types of problems than a human design algorithm Exactly So what does this all mean for the immediate future What should we watch next We really need to observe whether these viral gamified simulations end up producing any genuinely useful biological or structural insights Right That is the ultimate test It is The boundary between a static wiring map and a living organism is hard But the experimentation happening right now it is completely

unprecedented Will a developer who is basically just trying to optimize this map to run faster in a video game accidentally discover a more efficient way to process complex network data It is entirely possible The scientific community will certainly use this map for rigorous biology But the open source software community will use it for something entirely unpredictable And that unpredictability is the true story here We are officially closing out our deep dive into the fruit fly connectome In other news we are actually seeing another deep intersection between organic brains and artificial intelligence Yes we are The concept of repurposing neural data brings up a really interesting

parallel with what is happening in human education right now Google is turning its Gemini notebook into an interactive study space The company is rolling out mobile voice chats live lecture recording and generated quizzes tied directly to a user's own sources This represents a massive evolution for consumer educational tools It really does Google is basically attempting to solve the classic blank page problem in studying Right When a student sits down to review material starting from zero is always the hardest part Exactly And Google just wants to eliminate that initial friction Let us break down exactly what is happening with this software update The Gemini notebook application

on Android and iOS devices is getting a major multi tiered upgrade And the most immediate change is voice interaction Yes Users can now engage in real time fluid voice conversations directly with the application And this is not just traditional dictation right No not at all Google says this conversational feature is rolling out in nearly 100 different languages Oh wow The system is entirely designed to handle natural conversational speech patterns The interactive nature seems to be the key selling point here It is Users do not have to wait for the AI to finish a long monologue Right They can interrupt the AI's spoken responses They can

redirect the conversation entirely They can even ask for step by step guidance in real time if a concept is confusing them So it essentially transforms a static lonely reading experience into an active dialogue Yes You are basically talking to a tutor who has completely memorized your textbook But the most important detail is where that tutor actually gets its information Right This is not a general chatbot pulling random facts from the Internet Next week Google is dropping a dedicated mobile audio recorder directly into the app And that recorder changes the input dynamic entirely Because students will be able to record a professor's live lectures right into

their digital notebook Yes Those raw audio recordings immediately become part of their core source material That technical distinction is fundamental to how this whole system works The AI is strictly restricted It bases its spoken answers and all of its generated study materials only on the specific sources a student uploads It creates a walled garden for information It does The system is grounded exclusively in the user's curated collection of documents PDFs and live audio recordings Right If it is not in your notes or the lecture recording the AI is not supposed to talk about it And that grounding powers the rest of this massive update Because

the AI trusts the source material it can act on it The AI can now generate a whole suite of automated study materials from those specific sources Yeah It creates structured interactive learning overviews It automatically builds decks of digital flashcards It even takes your notes and generates 60 second short video overviews And it can produce those videos in more than 80 languages That is just wild And the testing and assessment capabilities are also expanding significantly right Yes The notebook system can generate specific graded quizzes based entirely on the uploaded lecture material The variety of those assessments is pretty notable These include traditional short answer questions Right

It also includes complex multiple select formats It even includes targeted fill in the blank questions to test vocabulary retention But the users are not just passively taking the quizzes They have editorial control Yes they do They can edit the AI's generated questions They can manually add their own questions to the test bank They can even ask the AI follow up questions if they get a quiz answer wrong Or if they just struggle with a specific flashcard Google is also using this massive feature launch to aggressively expand its market share among students They are definitely offering heavy capacity incentives to capture specific student populations Yeah Eligible

college students in the United States can get a subscription to Google AI Pro completely free for a full year And that pro tier offers four times higher notebook usage limits than the standard free version And the rollout strategy goes far beyond the United States Right Eligible college students in more than 140 other international markets can get a subscription to Google AI Plus free for a year Which provides twice the normal usage limits Exactly They want every student in the world hooked on this ecosystem But you know here is where we need to step back and look at the actual mechanics of learning Are we inadvertently

outsourcing the actual cognitive struggle of learning to an algorithm That is the core psychological dilemma with this entire technology Right Let's use an analogy here Imagine you want to get physically stronger So you go to the gym and you hire a highly advanced personal trainer A very logical step for improvement But this trainer does not just tell you what to do The trainer actually picks up the 50 pound dumbbell and lifts it for you Right They do the whole set Then they turn to you and ask well how was the workout Sure the trainer's muscles got huge But you did not break a sweat Your

muscles experienced zero friction Exactly If the AI summarizes the dense reading and the AI builds the flashcards and the AI scripts the video overview the student is just a passive consumer of content Right How does the human brain encode memory without that initial friction Neuroscientifically it is a very sharp critique The physical act of learning inherently requires cognitive resistance The messy process of struggling to summarize a complex lecture or physically writing out a flashcard by hand That is precisely how the brain encodes new information into long term memory It is active recall versus passive recognition Exactly If you just read a pre made flashcard you

recognize the answer You feel like you know it But you did not actively pull it from your own memory Right If the AI eliminates all the heavy mental lifting the student might feel incredibly productive while actually absorbing almost nothing The tool basically creates the illusion of competence We must also seriously consider the technical limitations of this closed loop design We must Grounding an AI exclusively in a user's own sources is supposed to prevent hallucinations It reduces the chance of the AI pulling irrelevant or wrong outside information from the wider web But that grounding does not inherently prevent the AI from totally misinterpreting a complex lecture

that it recorded Exactly The AI could listen to an hour long philosophy professor's lecture It could completely misunderstand the core concept And then it could generate a beautifully formatted highly convincing quiz that actively teaches the student the wrong thing If the live lecture contains subtle nuance or deep sarcasm or a highly complex theoretical framework well the AI's transcription and summarization might compress it inaccurately The AI does not understand sarcasm If the professor makes a joke about a historical event the AI might encode that joke as a hard historical fact on a flashcard And therefore the students are trapped in a new kind of work They

must still manually verify all of the AI generated study materials against reality Right The tool does not eliminate the need for critical manual review of the facts No it does not This raises a vital question about future user behavior What should you watch for next We need to see if tying every single study format strictly to a student's own material actually creates significantly better academic habits Because the alternative outcome is highly concerning this automated system could simply generate a massive overwhelming pile of AI made content Exactly A student might hit the end of the semester and find they have generated thousands of flashcards hundreds of

quizzes and dozens of video overviews that they never actually had the time to review It really highlights the vast difference between actively studying a subject and just mindlessly hoarding generated content on your phone Very true Next up a Polish developer named Paul Siedlowski has released an app called ZuckOff It uses Bluetooth signals to warn bystanders when compatible smart glasses are nearby Yes And this story perfectly highlights a rapidly growing tension in our physical environment It really does Moving from the friction of studying in private to the friction of existing in public spaces Exactly As wearable recording technology becomes cheaper and more common the social friction

between the person wearing the device and the unsuspecting bystander is increasing exponentially People just do not like being recorded without their knowledge And this application represents a fascinating grassroots response It is essentially a software based immune response from the general public against corporate hardware Let us look closely at the facts of this specific release The app quietly came out in August Right It is a free tool readily available to download on standard mobile devices And its core function is highly targeted It specifically listens for and detects the background Bluetooth broadcasts originating from Ray Ban Metaglasses Oakley Metaglasses and Snap Spectacles The underlying mechanism here is

straightforward but it is a clever use of existing network protocols It is If you own a pair of smart glasses they cannot process heavy AI tasks or upload high definition video on their own Right They need an internet connection So they constantly broadcast a Bluetooth low energy signal They are constantly searching for their paired smartphone to offload the heavy data processing They are essentially shouting into the digital void saying I am here Where is my phone Exactly And the Zuckoff app simply listens for that specific digital shout It captures those Bluetooth signatures and matches them to known manufacturer identifiers Because every Bluetooth chip has an

address that points back to who made it Yes But the app goes one step further than just detecting the presence of the device The software analyzes the signal strength of that Bluetooth broadcast It measures what engineers call the RSSI or Received Signal Strength Indicator Right By analyzing how loud that digital shout is the app estimates how physically far away the smart glasses actually are So it provides the user with a rough real time proximity alert It tells you there is a camera nearby And it gives you a sense of whether it is across the street or sitting at the table next to you And the

adoption metrics demonstrate a very clear public appetite for this kind of defense tool They really do As of Wired Magazine's publication on this story the app had quickly passed 5 000 downloads on Apple's App Store And it had also reached 1 000 downloads on the Google Play Store Now the basic environmental scanning is entirely free to use But the developer has introduced a paid option for power users A 25 ProTier unlocks much more persistent tracking features The ProTier adds continuous background alerts This means the app constantly scans the environment even when your phone screen is off and in your pocket So it literally vibrates if

a camera walks past you Yeah It also provides a historical detection log It allows you to export your encountered data via CSV files And it even adds a home screen widget for continuous passive monitoring of your surroundings We really need to connect this single app to the massive overarching picture of consumer hardware We do Why is a simple proximity app suddenly gaining so much traction right now Well the broader context is absolute market saturation Approximately 7 million pairs of MetaSmart glasses were sold globally in 2025 alone That is a staggering number It is These devices are no longer a niche tech novelty They are everywhere

And the general public's reaction has been incredibly mixed It's very mixed Educational institutions are proactively banning them from classrooms Major cinema chains are banning them from theaters to prevent piracy Right There is a rising ambient anxiety over the threat of covert recording People deeply resent the feeling of invisible surveillance hardware entering their daily private lives You just do not know if the person looking at you on the subway is just looking at you or if they're live streaming your face to thousands of people And Zuckoff is a desperate attempt to make that invisible hardware legible to the people around it But if we look at

the actual engineering here there is a massive practical flaw in the app's foundational design Yes The technical constraints dictate exactly how useful this tool actually is Let's think about this like a radar detector for your car Okay A radar detector tells you a police frequency is active in the area But it does not tell you if the cop is actually pointing the radar gun at you or if they are busy eating a donut That is the exact technical boundary of the Zuckoff software Right It is basically a radar detector for your privacy It tells you a specific brand of device is in the general area

but it absolutely does not tell you who is actually wearing it And more importantly it cannot tell you if they're actually recording you Exactly The app only matches basic Bluetooth signatures to manufacturer ID codes It recognizes a broad device category It is technologically impossible for it to identify a specific person So it might just make the user incredibly paranoid without actually making them any safer Right Your phone pings to say metaglasses are 30 feet away Suddenly you are looking suspiciously at every single person in the coffee shop trying to spot thick plastic frames That heightened paranoia is a highly probable outcome Furthermore the app cannot

access the internal functional state of the glasses themselves Because it is only reading the background pairing signal Correct The app absolutely cannot tell if the camera sensor is currently active Right It cannot tell if the microphone array is currently picking up audio It only knows that the physical device is powered on and transmitting a routine Bluetooth heartbeat This leaves unsuspecting bystanders in a very difficult awkward position The app gives them raw data but it does not give them any actionable clarity Exactly You know a camera is near but you do not know if it is a threat What should you watch next regarding this social

dynamic We really need to observe whether these proximity alerts actually change how bystanders behave in public spaces Will people physically get up and leave a room when their app pings Or will they aggressively confront random individuals wearing sunglasses indoors Or will the software just create a low level hum of ambient anxiety It provides a persistent digital warning without solving the underlying social friction of wearable cameras The detection technology exists but human social norms are desperately struggling to keep up with the hardware Yeah that is very true Meanwhile Bearstev has released its Q3 2026 Dev Barometer Survey showing a massive shift in software engineering 42 of

developers now say AI writes at least half of their code up sharply from just 12 a year ago That is a staggering year over year increase It really is A 30 percentage point jump in a single 12 month period indicates a fundamental reorganization of how commercial software is actually built But the deeper survey data reveals that the daily nature of the job is changing in highly unexpected challenging ways Yes let us dive into the actual time tracking findings The survey specifically asked developers about their perceived time savings Developers self report that AI tools are saving them about 13 coding hours every single week Which is

nearly double the seven hours of savings they reported during the previous year's survey However those massive double digit time savings are not translating into free time or spare organizational capacity No they are certainly not working 13 fewer hours a week They are still working full schedules The survey found the true shift Only 21 of developers say they spend more than half their week writing new original code from scratch The act of pure creation is basically becoming a minority task So if they are not staring at a blank text editor writing from scratch what are they actually doing with all that saved time Well the actual

labor is simply moving downstream in the software development lifecycle 67 of surveyed developers now spend the majority of their time reviewing AI generated code written by a machine And 52 spend their extra time debugging the subtle logic problems that were quietly introduced by the AI And the organizational response at the corporate level is rapidly aligning with this downstream shift Right The survey included enterprise chief technology officers 78 of those surveyed CTOs have actively increased their departmental spending on code review processes quality assurance teams and automated validation tools Individual developers are also spending heavily on their own continued education Yes They report spending an average of

9 hours a week just learning how to prompt manage and integrate new AI tools Wow And last year that educational overhead was only 4 hours a week This specific survey data systematically dismantles a very prominent very loud industry narrative Right For the last two years the standard assumption was that AI coding assistants would quickly and ruthlessly shrink engineering departments Venture capitalists and executives thought these tools would create massive amounts of spare capacity They assumed it would lead directly to widespread immediate job losses in the tech sector But this data shows the exact opposite phenomenon happening on the ground Exactly Generation the actual typing of the

code is becoming cheap commoditized and nearly instantaneous But validation proving that the generated code actually works without breaking the server is becoming the new highly expensive human centric bottleneck This operational shift sounds incredibly exhausting for the worker I mean if developers are spending all their time just hunting for and fixing AI mistakes doesn't that remove the fundamental joy of building software It radically alters the psychological profile of the job Think about it like trading writer's block for proofreader's fatigue That is a great analogy With AI you never have to stare at a blank page You never have to struggle to write the first draft But

your reward is that you get to spend all day fixing a machine's subtle typos and logic errors That is a very acute observation regarding the daily developer experience Ask any senior engineer Reading and reverse engineering someone else's messy code is inherently more cognitively demanding than writing your own logic from scratch Right Now multiply that cognitive load by the massive volume of code an AI can spit out in three seconds It is intense But we absolutely must highlight the limitations of this specific dataset Right We should The survey provides a strong signal but it has strict methodological boundaries Yes The underlying methodology requires careful scrutiny The

survey included 705 developers spread across more than 60 countries However many of those specific respondents were active job applicants currently in the BearsDev screening and hiring process They were not necessarily securely employed engineers safely sitting in stable enterprise roles And that applicant status could easily skew their self reported reliance on AI tools Furthermore the CTO sample size was statistically quite small It only included 41 individual executives We cannot treat this single survey as a definitive flawless census of the entire global software profession Even with those caveats we must look at the hard boundary of automation that the data clearly reveals Right Only 7 of survey

developers claim that an AI system can ship code entirely to production without any human input Which means 93 of the time a human being is still standing squarely between the AI's output and the final live product Human accountability remains the absolute hard boundary in commercial software BearsDev CEO Darren Shimkus specifically noted this dynamic He said that engineers increasingly have to explicitly explain and defend the AI generated code before checking it into the main repository especially for high stakes production systems Generating a block of code and legally accepting responsibility for that code are two entirely different professional tasks Exactly The AI algorithm cannot be held legally

or operationally accountable for a catastrophic database failure in a live production environment A human engineer with a salary must ultimately sign off and take the risk This framing points exactly to what we need to watch next in the industry We have to monitor whether corporate validation and QA demands keep rising proportionally with generation speed As AI models take on larger vastly more complex initial architecture drafts does the human proofreading burden eventually become physically unsustainable If the AI generates 10 times more code per minute but the machine's underlying error rate remains constant the absolute raw number of bugs a human must manually find and fix will

skyrocket The tech industry must solve this validation bottleneck before it can truly capitalize on the promise of infinite generation speed Starting our Quick Reads Google has launched an Open Access Explorer for its AI and Economy ATLAs data This public platform provides unprecedented visibility into a massive dataset regarding how people are actually utilizing AI globally The data covers 15 million distinct user interactions across the Gemini app Google's AI mode and the Enterprise Gemini API The geographic spread of the data spans more than 150 countries It tracks usage across 140 languages 800 specific occupations and 4 000 highly specific work tasks And the data highlights a very

sharp surprising geographic divide in how different cultures use the tool In the United States 30 of all work related AI usage is concentrated strictly in computer and mathematical occupations That is exactly twice the share seen in the rest of the global data Conversely in India occupations related to the arts design and media account for 19 of total usage A linked academic study involved more than 600 active scientists in the U S and the U K It found that nearly half of them now use generative AI on a daily basis They report saving about seven hours a week strictly on literature review and experimental prep work

However this data brings us back to a critical physical limitation Saving seven hours on digital preparation work does not speed up the laws of physical reality Right Validating the AI's complex outputs setting up physical lab experiments and conducting slow human clinical tests still takes exactly the same amount of time This dynamic has created a massive towering backlog of brilliant AI generated scientific hypotheses that are just sitting in a queue awaiting slow real world physical testing It is a real bottleneck Next Mozilla has published new analysis on the shrinking gap between open and closed AI models This benchmarking report severely challenges the prevailing industry assumption that

massively expensive closed frontier models will always maintain a permanent insurmountable intelligence lead over cheaper open source alternatives The Mozilla analysis finds that Chinese open weight models are now roughly only 4 4 months behind the capabilities of United States closed models And they achieve this performance while operating at a tiny fraction of the computational cost The specific benchmarking data is highly revealing Moonshot AI's open Kimi K3 model scored just three points behind Anthropic's closed Fable 5 model on a massive composite testing index But the Kimi model costs about 30 as much to run Another rigorous test put Zeta AI's GLM 5 2 model within a single

point of the massive Claude Opus model on the terminal bench assessment And it achieved that score at roughly one fifth the per task operational cost But there is a significant performance caveat here Yes While these open models are incredibly efficient for short routine tasks the massive closed models still win decisively on long running complex reasoning tasks Tests showed the best closed model could reliably handle unbroken agentic work lasting 12 hours without hallucinating The best open model degraded after roughly seven hours of continuous complex reasoning We must also heavily acknowledge the massive geopolitical concentration risk identified by Mozilla's data Eight of the top ten open models

by total token volume on the OpenRouter platform were developed by Chinese organizations Yet open models as an entire global category currently capture only a paltry 4 of total commercial AI model revenue Wow Finally WIRED has highlighted a growing privacy clash surrounding smart glasses This specific report grounds the abstract theoretical debate about surveillance hardware in a very specific deeply uncomfortable personal encounter A woman named Courtney McAnuff recently discovered that her date had secretly recorded their entire evening using a pair of Ray Ban Meta glasses And he subsequently posted the point of view footage of their private bar visit and their subway ride directly to his public

Instagram account without securing her consent They had absolutely never discussed recording the date or posting the video online In response to incidents like this Meta routinely points out that the glasses feature hardwired hardware safeguards A prominent front facing LED light blinks brightly when taking photos That same LED stays lit continuously during active video recording Meta says the internal camera circuit is physically designed to shut down entirely if that external LED light is blocked with tape disabled or physically tampered with But if we connect this incident back to the Zuckoff app story the limitation of this corporate defense is glaring Hardware safeguards like a blinking LED

light only solve the mechanical problem of hidden capture Right They do absolutely nothing to establish active informed social consent between two human beings An LED light simply indicates a mechanical state of recording It cannot negotiate verbal permission It cannot legally govern what happens to the footage after the file is saved to the phone And it certainly does not dictate whether the private video should be broadcast to thousands of strangers on the Internet The sheer technological capability to record everything is drastically outpacing the human social norms required to manage that recording respectfully We are moving to three takeaways from today Okay First automation is simply moving

the bottleneck Whether it is developers debugging AI code or scientists waiting to run physical clinical trials generating ideas is fast but validating them is becoming the new grind Second hardware safeguards cannot fix human social norms From Bluetooth warning apps to blinking LED lights technology is struggling to mediate the awkward reality of wearable cameras in public and private life Absolutely Third the open source ecosystem is driving unpredictable innovation From Chinese models rapidly closing the performance gap to developers running fruit fly brains in Minecraft open access is constantly redefining what these tools can do Well said Watch tomorrow whether the backlog of AI generated hypotheses forces the

scientific community to radically speed up how they conduct physical real world validation For more head over to superpowerdaily com Thank you for listening and we'll see you tomorrow