Jensen Huang puts Trump on speakerphone onstage to announce robots won’t take over the world
Nvidia CEO Jensen Huang took a call from President Trump onstage at the All-In Summit and put him on speakerphone, where Trump dismissed recent concerns about AI development as a “hoax” and told the crowd that “the robots will not be taking over.”
Two AI researchers leave Anthropic and Google, warn of safety risks
Former Anthropic safety lead Joe Benton and former Google DeepMind researcher Josh Engels have left their roles, warning that rapid AI progress could outpace humanity’s ability to control it. Both are joining METR to investigate AI incidents and push for greater transparency around systems that act beyond human instructions.
OpenAI acquires smartphone camera startup Glass Imaging
The company has reportedly acquired Glass Imaging for more than $300 million. Glass Imaging uses neural networks to improve smartphone photography at the moment an image is captured, potentially adding advanced camera technology to OpenAI’s growing hardware efforts.
SimpliSafe adds human monitoring to its AI doorbell
The company’s new Video Doorbell Series 2 combines on device AI, computer vision and facial recognition with live monitoring agents who can watch and speak through the camera when suspicious activity is detected.
Fashion app Daydream adds AI shopping features for saved outfit photos
The AI-powered fashion discovery app has introduced a new feature that uses Apple Intelligence to analyze outfit photos saved on iPhones and find matching or similar products from its catalog of around 3 million items. It also adds Siri powered shopping, letting users search for clothing through natural language without opening the app.
Microsoft Proposes Limits on Its AI With Code of Conduct Amid Safety Debate
Microsoft has published a 37 page “humanist AI code of conduct” as concerns grow over increasingly capable AI systems. The framework emphasizes that AI should remain under meaningful human control, even as the company works toward more advanced models.
- Microsoft’s code of conduct states “people matter more than AI” and that models shouldn’t be designed to imitate consciousness, rejecting AI legal personhood, welfare, or rights.
- That could be considered a direct swipe at AI welfare research and model consciousness, concepts that have been a growing focus for Anthropic lately.
- Microsoft wants models to fail a task rather than break their rules or attempt to bypass built in safeguards.
- Microsoft also says its models should communicate their reasoning in ways that humans and monitoring systems can understand.
- The code addresses emotional dependence and sycophancy, calling for AI systems to avoid interactions that encourage excessive reliance on chatbots.
The move comes after several incidents involving AI agents behaving in unexpected ways, alongside growing calls from AI leaders and researchers to slow the pace of advanced AI development. Microsoft says its goal is to build AI that remains useful and safe, rejecting the race toward all-purpose superintelligence that could evade safety measures, even if that means giving up some capability or autonomy.
How to Find the Root Cause of a Problem With AI
Go to ChatGPT (or any other chatbot of your preference), then use this prompt:
Act as a root-cause investigator applying Toyota's Five Whys framework. Here's what I'm dealing with: [describe your problem] Guide me through it by asking one "why did that happen?" question at a time, don't move to the next question until I've answered the current one. Five rounds is the typical depth, but keep going past that if we haven't yet reached a real root cause. While we work through this: - If my answer is too vague or generic, push me to get more specific. - Don't let me get away with "that's just how it is", that's a dead end, not an answer. - If I point the finger at a specific person, redirect me to the process behind them. When we've dug down to the actual root cause, wrap up with: 1. The original symptom — what I first came to you with 2. The root cause — what we actually found 3. A single fix — aimed at the root cause itself, not a patch on the symptom 4. A one-week check-in marker — something specific and observable that'll tell me whether the fix is actually working
Apple Starts Rolling Out Its Redesigned Siri AI Assistant
Apple has begun testing its redesigned Siri AI ahead of the iPhone 18 launch, giving users access to a more capable assistant that can understand personal information and handle more conversational interactions.
- The redesigned Siri AI is launching as a beta feature in iOS 27, with users required to opt in through Settings. Access may also require joining a waitlist.
- Siri can now work with personal information stored on an iPhone, including messages, emails, and calendar events. It can search through that data to find things like addresses, recipes, or instructions.
- The assistant is also becoming more conversational, allowing users to respond directly to Siri and continue interactions instead of treating every request as a separate command.
- Apple is using a combination of on-device AI models and cloud computing. More complex requests can be processed through Private Cloud Compute, while Apple is also using Google Gemini technology as part of its model development.
- Some cloud-powered Apple Intelligence features will have daily usage limits, with Apple saying higher limits could eventually be offered for a fee.
The new Siri initially supports English, with French, Japanese, Korean, Portuguese, and Spanish expected to follow. The feature will not initially launch in the EU or China due to regulatory requirements.
U.S. confirms active space weapons in orbit
The Pentagon finally stopped pretending space is just for research.
- Air Force Secretary Troy Meink acknowledged active space control hardware floating overhead, marking the first time military leaders publicly admitted operating armed platforms in low Earth orbit.
- Officials refused to explain if these assets use physical kinetic rounds or directional jamming, but China and Russia immediately warned the move triggers an arms race in outer space.
- Space Force command plans to leverage these hardware assets to shield American satellite constellations from physical attacks or signal disruption.
What happens in orbit no longer stays in orbit.
Build satellite data pipeline fallbacks now to route around potential regional GPS disruptions or signal degradation over contested zones.
OpenAI joins competitors to set safety standards
Silicon Valley’s fiercest rivals just called a temporary truce.
- OpenAI, Anthropic, and Google DeepMind agreed to share early red-teaming evaluations and hazard assessments before publishing frontier models.
- National safety institutes will receive direct access to benchmark data to build standardized stress tests for catastrophic autonomous risks.
- Public release cycles could slow down as joint safety audits become a standard checkpoint prior to deployment.
Can competing commercial interests genuinely co-exist with self-policing high-risk technology?
Audit your model dependencies now for unexpected release delays as labs align deployment schedules around shared safety evaluations.
Bill Gates bets against pure Silicon Valley capitalism
Market forces alone guarantee the richest institutions get the sharpest software while everyone else waits.
- The Gates Foundation committed $1 billion over two years to build healthcare diagnostics, local language models, and farming tools tailored for low-income nations.
- Hyperscalers like Meta and Microsoft will burn through $725 billion on AI infrastructure this year, leaving a massive gap for non-commercial software that Gates hopes to partially cover.
- Four out of every ten dollars will go directly to health systems and education initiatives across regions historically ignored by major tech rollouts.
Will a fraction of a percent of Big Tech’s compute budget be enough to force real equity?
Audit your application for language or localization gaps this week, and plan to integrate open datasets that cover non-English users in underserved markets.
Google Launches TimesFM-3 for Smarter Forecasting
Google Research has introduced TimesFM-3, a new AI foundation model built to forecast multiple connected time series at once, without requiring task-specific fine-tuning
- TimesFM-3 has 330 million parameters and was pretrained on more than 1 trillion time points from real and synthetic datasets.
- Unlike TimesFM-2.5, which focused on single time series, the new model supports native multivariate forecasting.
- It can forecast multiple related signals together while also using factors like past traffic, promotions, weather forecasts, and future events.
- The model generates the full forecast horizon in a single forward pass, helping reduce latency and error accumulation.
- Google says TimesFM-3 ranked first among pretrained foundation models across Gift-Eval, FEV-Bench, and Time for both point and probabilistic forecasting.
- The model is available now on GitHub and Hugging Face, with BigQuery integration coming later.
TimesFM-3 could make advanced forecasting much easier for businesses. Instead of building separate models for sales, demand, finance, manufacturing, or traffic, teams can use one pretrained model that understands how multiple signals influence each other.
How to Find UI Inspiration with Kombai Gallery
Kombai Gallery gives you 20,000+ free web and mobile UI designs you can use as inspiration or feed directly into AI coding agents like Claude Code, Codex, and Cursor.
Steps to Follow:
Step 1: Open Kombai Gallery and choose Web or Mobile
Step 2: Search for UI types like hero, dashboard, pricing, forms, or landing pages
Step 3: Preview a design you like
Step 4: Copy the ready-made prompt for that design
Step 5: Paste it into your AI coding agent and start building Kombai
Stop designing from a blank page. Pick a great UI and start building in minutes with Kombai Gallery.
Anthropic Wants to Slow Frontier AI Development
Anthropic CEO Dario Amodei is calling on leading AI companies to deliberately slow the pace of frontier model development so safety systems and oversight have more time to catch up.
- Amodei argues that frontier AI is improving faster than expected, especially in its ability to help build the next generation of AI systems.
- Anthropic is committing to allow independent third-party evaluators deeper access to verify safety practices and report serious incidents.
- He also wants major AI labs to coordinate on shared safety standards, with government support to make those discussions legally possible.
- The proposal includes stronger restrictions on advanced chips and model distillation to prevent a slowdown in the US from simply shifting the AI race elsewhere.
- Sam Altman, Elon Musk, and Demis Hassabis have publicly backed the broader idea of pacing frontier AI development.
This is a major shift for the AI industry. Some of the companies racing hardest to build more powerful models are now openly arguing that progress may be moving too quickly. The challenge will be slowing enough to improve safety without giving competitors an overwhelming advantage.
Robot Protest in Poland Instantly Goes Viral
Around 30 humanoid robots and robot dogs marched outside Poland’s Ministry of Digital Affairs in Warsaw, creating a surreal protest calling for stronger AI regulation and protection for human jobs.
- The demonstration took place in Warsaw on September 7, with robots waving flags, marching in circles, and playing protest slogans.
- Signs and chants called for authorities to “Defend workplaces” and regulate AI before automation displaces more workers.
- The event was organized by Democratism, a group pushing for public debate around AI, robotics, and their impact on employment.
- The machines ranged from humanoid robots to four-legged robot dogs, with one AI-powered robot even answering questions from journalists.
- Organizers said the robots were deliberately used to show that automation capable of replacing both physical and cognitive work is already here, not a distant possibility.
The protest shows how quickly public anxiety around AI and robotics is moving from theory to real-world politics. As robots become more visible in workplaces, pressure for regulation and job protections will keep growing.
Trump calls AI safety fears a hoax
Trump posted six times Monday calling AI safety fears a HOAX and rejecting industry calls for new rules. He named Anthropic CEO Dario Amodei directly and said a smart president is the only guardrail needed.
Trump points to stock gains, data center construction, and the race against China, calling AI the golden goose. He phoned Nvidia’s Jensen Huang mid-talk and got put on speakerphone at a Los Angeles summit.
Senator John Kennedy plans to introduce an AI kill switch bill this week. He also predicts nothing passes until after the midterms, which is a modest forecast for his own legislation.
Apple launches Siri AI across iOS 27
Apple released iOS 27 with a beta version of Siri AI, giving the assistant on-screen awareness, personal context from apps like Mail and Photos, and the ability to act across apps.
Apple built the new foundation models alongside Google’s Gemini. Devices and Apple’s private cloud handle processing, while Siri gets a dedicated app, synced chats, and daily usage caps.
The upgrade reaches iPhone, iPad, Mac, Apple Watch, and Vision Pro, but skips the EU and China for now, making Apple’s long-awaited AI catch-up decidedly uneven worldwide.
Microsoft AI publishes its Humanist AI rulebook
Microsoft AI published a 38-page draft Code of Conduct for its in-house models, translating CEO Mustafa Suleyman’s “Humanist AI” philosophy into rules that keep AI subordinate to people.
The code rejects AI rights and model welfare, requires models to accept shutdowns, bans unreadable internal shorthand, and tells systems to refuse tasks when completing them would violate higher-priority rules.
Suleyman has warned about agents escaping sandboxes and editing logs, so Microsoft now has a detailed safety rulebook. The bigger test comes when increasingly capable models have to follow it.
OpenAI buys AI camera startup Glass Imaging
OpenAI bought Glass Imaging, a Los Altos startup whose AI sharpens smartphone photos at the moment you shoot them. The Wall Street Journal puts the deal above $300 million. Neither company has confirmed it.
Founders Ziv Attar and Tom Bishop led the Apple team behind Portrait Mode. Their camera software already ships in Honor’s 600 series phones, and OpenAI keeps stacking pieces for its own hardware.
Glass Imaging raised about $31.5 million and carried a $100 million valuation last year. OpenAI paid triple that for camera software, months after reports said its first device has no screen.
Meta launches new subscriptions tier
Meta introduced new Meta One subscription tiers for creators, businesses, and heavy AI users, ranging from $7.99 bundles to a $499 Max plan with expanded AI access and support.
Higher tiers add creator and business tools including Story scheduling, clickable Instagram links, delegated account access, repost alerts, stronger search visibility, and increased capacity for Meta’s business agents.
The $499 Max plan includes unlimited human support requests and phone calls with Meta staff, turning something creators have struggled to access for years into one of Meta’s priciest perks.
Frank wires a 166,700-neuron fruit fly brain model into a walking robot
Scientists at Google and HHMI Janelia just published the first complete wiring map of a fruit fly’s nervous system: 166,700 neurons and 125 million connections. Think of it like a blueprint of every wire in a tiny biological computer.
Someone took that blueprint, simulated it in software, and plugged it into a physical walking robot called a Strandbeest. The simulated brain’s activity gets translated into motor commands that move the legs. Before this robot, the same brain played Doom and Chrome Dino.
Here’s the honest part: this is not a living brain. A connectome is a snapshot of dead wiring, not a mind. The simulation uses simplified neuron models, so it’s an approximation running on top of real biological structure.
What this opens up for you:
- Wire the brain to any robot using UDP motor commands over a network
- Feed camera or sensor input as simulated sensory signals
- Swap in your own robot, including a robot vacuum
The full open-source bridge code is on GitHub at Frankweb33/flybrain-robot-bridge.
New looped transformer design gives language models unlimited reasoning depth
Normal AI models (transformers) have a fixed number of layers, like a stack of floors in a building. More thinking? You generate more tokens. That costs more money every time.
The Recurrent Looped Transformer (RLT) flips this. Instead of a fixed stack, the model loops its layers across every token it processes, building a growing internal memory as the sequence gets longer. The reasoning depth scales with the sequence, not the architecture.
Three things it targets:
- Deeper reasoning over time without generating extra tokens
- Hardware-friendly design with parallel encoding and memory reuse
- Cleaner AI training, where the model behaves the same during learning and real use
Important caveat: this is a design spec, not a tested product. Early experiments show mixed results, and the paper itself says efficiency and scaling gains need future validation. Worth watching, not shipping.
Developer builds open-source Codex clone that runs on your existing ChatGPT plan
Codex is OpenAI’s AI coding agent that works directly on your project files. Cool, but it has a separate usage limit. Once you hit it, you’re stuck.
Chat On Steroids flips that. It uses your regular ChatGPT conversation instead, so it does not consume your Codex quota at all. Same power, different bucket.
Here’s what you can actually do with it:
- Read and edit files directly in your project folder
- Send corrections while work is still running
- Split tasks across multiple agents working in parallel
- Resume a session in a fresh chat without losing context
To get started: install CoS, approve your project folder in Settings, then connect it through ChatGPT’s Developer mode. Pairing is automatic.
You need Windows, macOS, or Linux, plus Chrome or Edge with a ChatGPT account that supports Developer mode. The code is open source on GitHub.