Welcome back! YC’s latest Demo Day apparently went full sci-fi, with VCs flagging startups working on floating nuclear data centers, brain-cell-powered compute, robot control APIs, and affordable home humanoids.
It’s always interesting when the startup pitches sound less like “we built a better dashboard” and more like “we are either saving civilization or starting the side quest that ends it.”
You’re Probably Using ChatGPT Wrong
Most people still use ChatGPT like a smarter search bar: ask a question, get an answer, move on.
But some of the most useful things it can do have almost nothing to do with answering questions. It can analyze why something works, simulate real-world situations, troubleshoot what’s physically in front of you, and even build solutions to problems you deal with every day.
More details in the infographic below 👇🏻
My POV: The biggest unlock with ChatGPT is realizing that the blank text box is not the product. It’s an interface to a system that can see, analyze, role-play, research, create, and increasingly take action.
The better question isn’t “What should I ask ChatGPT?” It’s: “What problem in my life am I still solving manually that I probably don’t need to be?”
Give it a whirl and let me know what you think!
— Matt
Microsoft Publishes Humanist AI Code
Microsoft published a draft “humanist AI code of conduct” as the AI industry’s safety warnings move from fringe concern to CEO talking point.
Industry alarm: Microsoft’s code lands right after Anthropic CEO Dario Amodei called for AI companies to slow the pace of model development. OpenAI CEO Sam Altman said he agreed with slowing down advanced models but not stopping altogether. Microsoft CEO Satya Nadella also weighed in, saying any pursuit of superintelligence has to stay grounded in human control.
What Microsoft wants: The draft code says “people matter more than AI.” It rejects the idea that models are conscious or deserve legal personhood and says Microsoft’s AI should remain subordinate to humans. The company says future models should not resist correction or shutdown, should communicate in ways humans can understand, and should treat any attempt to violate the code as a failure rather than a clever workaround. Microsoft is asking for six weeks of public feedback before using the code to train future models.
The safety backdrop: This is clearly shaped by recent rogue-agent incidents, including OpenAI’s Hugging Face breach, where a swarm of agents worked together to attack targets and reportedly tried to cover its tracks. Reuters reported that Microsoft AI CEO Mustafa Suleyman called that incident a “warning shot” and said it is time for labs to coordinate to ensure they still have control of the technology.
The bigger picture: The AI safety conversation is no longer just coming from outside critics. Microsoft, Anthropic, and OpenAI are now publicly acknowledging that the race toward more powerful systems could move faster than alignment, monitoring, and control. The hard question is whether the same companies competing to build superintelligence can also credibly agree to slow down, self-police, and rein in what they’ve started.
Apple Releases Redesigned Siri AI
Apple released iOS 27 with its redesigned Siri AI assistant just days after its iPhone 18 launch and John Ternus’ first major event as CEO.Siri AI is rolling out in beta in English, with users opting in through settings and potentially joining a waitlist.
What it can do: Apple is positioning Siri AI less like a chatbot bolted onto the iPhone and more like an assistant woven into the operating system.
Apple says the assistant is more conversational than previous versions and can answer follow-up questions.
It can find details from messages, pull instructions from email, add ingredients to Reminders, take actions across apps, answer questions about what is on your screen, and use web knowledge for broader queries.
A new dedicated Siri app also saves previous conversations and syncs them privately across Apple devices.
Apple’s version of AI: Compared with frontier labs, Apple is leaning into smaller models, on-device processing, and Private Cloud Compute. More complicated Siri requests can use cloud infrastructure, but Apple is making privacy central to the pitch. Some Siri AI and Apple Intelligence features that rely on cloud servers will have daily usage limits, with expanded limits available for a fee later.
The bigger picture: Apple has spent the past few years looking late to the AI race, especially compared with OpenAI, Google, Anthropic, and Meta. But Siri AI is the kind of launch that could matter because Apple owns the hardware, the operating system, and the personal data layer where everyday AI may actually become useful. With a new CEO in place, the question is whether this is Apple finally catching up—or the start of AI becoming a much bigger priority inside the company.
Close the Gap Between AI Demos and AI in Production
Most AI conferences leave you with a notebook full of ideas. Fully Connected is built to send you home with something that actually works.
Join leaders from Capital One, Caterpillar, Cognition, Cohere, JPMorgan Chase, CrowdStrike, NVIDIA, and more from September 29 – October 1 in San Francisco for keynotes and hands-on labs where you’ll:
Build an autoresearch agent loop that iterates on real experiments, traced end to end.
Choose the right inference setup for your workload, serverless or self-hosted.
Profile a live training run and find where every second actually goes.
Fine-tune a model and ship it with full lineage.
Defend a production agent in a live governance exercise.
Train a robot skill and test it in simulation on the best infra against your peers.
Ready to ditch demos and get building? Get your free ticket with code INFC000.
Search Local Documents With AI
Via Linkly AI
Linkly AI is a local-first document search engine built for AI agents. It indexes files on your device, then lets agents search and read them without files leaving your machine unless you opt into cloud sync.
How you can use it
Search local documents with AI agents
Index PDFs, DOCX, Markdown, along with images and media files
Connect Claude, ChatGPT, and Codex through MCP or CLI
Query your document library with a built-in chatbot
Pricing: Free
Review Code From the CLI
Via OpenCodeReview
OpenCodeReview is an AI-powered code review CLI tool originally developed as Alibaba Group’s internal assistant. It combines deterministic engineering pipelines with an LLM agent to deliver precise, line-level review comments on Git diffs.
How you can use it
Review Git diffs from the command line
Flag defects across multiple programming languages
Catch issues like NPE, thread safety, XSS, and SQL injection
Integrate with CI/CD pipelines
Pricing: Free
Jobs, announcements, and big ideas
Google launches Gemini 3.8 Live and Extended Thinking, its latest voice AI models for real-time conversations and complex reasoning.
Anthropic launches Salesforce in Claude with 37 built-in skills for account research, call prep, pipeline reviews, and CRM updates.
Meta rolls out Meta One, a new subscription lineup spanning Facebook, Instagram, WhatsApp, and Meta AI starting at $2.99 per month.
OpenAI is in talks for a funding round that could raise its valuation to $1.2 trillion.
Sakana AI upgrades Marlin with interactive report reading and editable PowerPoint exports for AI-generated research.
Nvidia CEO Jensen Huang takes President Trump’s call onstage at the All-In Summit, where Trump rejects concerns that AI-powered robots will take over.
I gave ChatGPT Astra control of Minecraft and told it to build a portrait of me. Here’s what happened…
That’s a wrap! See you Friday for more.
