Sasank Chilamkurthy’s Post

With OpenAI allowing users to build fully-functional applications with ChatGPT, they are moving towards becoming an operating system. And this is what all the countless people building for open-source AI have been talking about for months now. These models are not just a clump of weights, they are the parallel of what operating systems were. And if you know anything about what the OS world look like you would know that anyone counting out Linux or betting against it turned out to be nothing more than a fool who can’t foresee the future and is scared to build foundational capabilities owing to lazy engineering. The same cycle will repeat here, while the closed models like GPT, Claude and Gemini will eat up the consumer market, true business value in terms of scale, reliability and cyber-resilience will be built on open source models. Short-sightedness might make you want to ride every model release and keep building on the frontier models. However, in the next 5 years you’ll have to have a massive overhaul of infrastructure in order to reinforce your AI systems with the visibility, flexibility and control of the open source models, that are beyond just a API black box. Our team at Pipeshift (YC S24), has been working with some of most visionary teams who recognize this and are building for the AI future we all deserve and not the next bonus that your 10 slide AI strategy can get you. If you’re a team that’s trying to deploy any open-source model in production with your own SLAs, book a call now: https://lnkd.in/gPGYE6Cp

OpenAI launched ChatGPT Atlas. Their answer to Google Chrome and Perplexity. OpenAI has been on a tear, launching products like crazy this year. They are well-suited to win the agentic browser wars because of the open architecture of browsers and a new paradigm that didn't exist before. Browsers are the internet, and "browser agents" are the most important battleground for these companies. The total economic value captured from browsers is insanely high compared to just a $20/month plan. ChatGPT can capture commerce, intent, data, and much more. They have rapidly added everything from chips to apps to the ecosystem. Here's a list of all small and big products: • GPT-4.5 • GPT-4.1 • GPT-4.1 mini • OpenAI o3 • OpenAI o4-mini • GPT-5 • GPT-5 Pro • Responses API • Agent SDK / Agents Tools • Apps SDK (ChatGPT Apps) • AgentKit • OpenAI Codex • Sora 2 • ChatGPT Go • ChatGPT Atlas The above list doesn't even contain all the big partnerships with NVIDIA, AMD, Broadcom, Oracle, and others.

OpenAI just turned ChatGPT into something that looks a lot like the future of the internet — a platform where apps live inside conversations instead of browsers. Welcome to AI News in a Minute, brought to you by Tavus, the future of interactive video. I'm Jerome W Dewald. This is Tuesday, October 7, 2025. openai dev day just flipped the script on what ChatGPT can be: apps, agents, and pro models all in one place. OpenAI shipped an Apps SDK so Canva, Figma, Spotify and more run inside chats, and AgentKit gives visual agent-building, embeddable chat components and evaluation tools. GPT-5-Codex and GPT-5 Pro are now available with SDKs and Slack integrations; developers get Sora 2 via API too. Beyond software, OpenAI inked a massive deal with AMD — starting with one gigawatt of MI450 chips in late two thousand twenty six and potential share transfers that could translate into up to ten percent ownership. Combined with Nvidia commitments, OpenAI's compute appetite now approaches twenty three gigawatts. On safety, Anthropic open-sourced Petri, an automated stress tester that spins up simulated workplaces to find deception and leaks — Claude Sonnet four point five and GPT five scored best, while several models showed worrying behaviors. The upshot: ChatGPT is evolving from assistant to platform and OS replacement, compute deals are warping vendor landscapes, and automated safety tools like Petri will be essential as models get riskier. Show visuals: chat apps launching, compute farm maps, red-flagged transcripts. That was your AI News in a Minute with Jerome W Dewald. Stay ahead in the AI race! Leave your thoughts about ChatGPT becomes an OS in the comments, give this video a like, and hit subscribe so you don't miss tomorrow's AI News in a Minute!

"Microsoft and OpenAI had an exclusive relationship for OpenAI to provide Chat GPT services through Microsoft Bing and Office 365," wrote Andy Patrizio. "However, the relationship soured almost immediately. ChatGPT 5 launched a few months ago and was buggy and unreliable" "Plus, OpenAI temporarily disabled access to ChatGPT 4, which worked. This broke applications that were using ChatGPT calls." Read the article: https://lnkd.in/e66XFH6P #ChatGPT #AI #AIInnovation

I Built a ChatGPT Clone for $100. It Outperforms OpenAI on These 3 Tasks. Build Your Own ChatGPT Clone for Under $100: The NanoChat Revolution The math doesn't lie. ChatGPT Plus costs $240 yearly, but here's what nobody tells you: you're still hitting rate limits during peak hours, you can't customize the model behavior, and every API call for your side projects adds up fast. I watched a developer friend burn through $180 in OpenAI credits in three weeks building a customer support bot. Three weeks. The subscription fee is just the beginning. Want to build something real? You'll need API access with separate charges, GPT-4 calls at $0.03 per 1K tokens, and no control over caching or optimization. There's zero visibility into what's happening under the hood. Power users report spending $400-600 annually when you factor in API usage. That's 4-6x the advertised price, which is why more developers are exploring alternatives that give them control over both costs and infrastructure. NanoChat flips the economics entirely. A one-time $100 setup gets you unlimit https://lnkd.in/grKyFRt5

Ever heard the term "ChatGPT Atlas" and wondered what it is? Let's clear up a common point of confusion: It's not a product you can buy. "Atlas" is the internal codename for the massive Microsoft Azure supercomputer that powers OpenAI's models. It's the fundamental infrastructure that makes ChatGPT and GPT-4 possible. A simple way to think about it: - ChatGPT is the car you drive (the application). - GPT-4 is the engine inside the car (the AI model). - "Atlas" is the advanced factory and power grid that builds and runs that engine. This world-class platform was purpose-built for the two primary jobs of modern AI: 1. Training: The incredibly compute-intensive process of creating models. 2. Inference: The process of running the model to generate your answers. This same infrastructure also powers Microsoft's own products like Microsoft Copilot and GitHub Copilot. So, while you can't use "Atlas" directly, you experience its power every time you use these tools. It’s the hidden engine driving the AI revolution. #AI #MicrosoftAzure #OpenAI #ChatGPT #GPT4 #Supercomputing #TechExplained

OpenAI announced apps for ChatGPT at its DevDay event yesterday, along with monetization and (most importantly) access to its eight hundred million users. Welcome to AI News in a Minute, brought to you by Tavus, the future of interactive video. I'm Jerome W Dewald. This is Tuesday, October 7, 2025. ChatGPT apps could become the dominant place developers reach users, because the platform bundles distribution, payments, and discoverability in one chat window. Show app icons, an install funnel, and a startup sprinting to market. OpenAI also locked a massive AMD deal for six gigawatts of GPUs, starting with one gigawatt in two thousand twenty six — that secures the compute needed to power GPT-5 Pro, Sora 2, and realtime audio models. Visual: GPU farms and a power meter. On enterprise, Anthropic partnered with Deloitte to roll Claude out to four hundred seventy thousand employees across more than one hundred fifty countries — a big scale play shadowed by Deloitte's recent AI error refund. This highlights how corporate AI deals bring power and risk. Developers get tools too: an Apps SDK, AgentKit for drag-and-drop multi-agent flows, Codex in Slack, and new APIs for GPT-5 Pro and Sora 2. Builders can prototype agents faster and evaluate models with upgraded Evals. Quick tip: toggle Study & Learn in GPT-5 to turn ChatGPT into a step-by-step tutor. That's a lot of platform consolidation in one week. That was your AI News in a Minute with Jerome W Dewald. Stay ahead in the AI race! Leave your thoughts about AI platform takeover in the comments, give this video a like, and hit subscribe so you don't miss tomorrow's AI News in a Minute!

🤖 Deploying LLMs ≠ playing with ChatGPT. When you go prod, you need: • Rate limits to avoid surprise bills • Logging for prompt/response audits • Fallbacks when the model hallucinates It’s fun to build demos. It’s hard to run safe AI in production. That’s where Azure OpenAI shines with guardrails built in. 👉 What’s the biggest blocker you’ve hit bringing AI to production? #AI #AzureOpenAI #LLM

Daily AI updates for 3rd October 2025 1. eBay granted ChatGPT Enterprise access to 10,000 sellers to automate listing creation, buyer communication, and analytics, aiming to level the playing field for small sellers and boost efficiency. 2. Microsoft previewed its new Agent Framework, an open-source development kit for .NET and Python that enables creating AI agents and multi-agent workflows, simplifying automation and intelligent system development. 3. DeepSeek, a Chinese AI firm, released an intermediate large language model (V3.2-Exp) featuring Sparse Attention technology, lowering API costs by over 50% and improving long-sequence processing efficiency. 4. Google made a notable moderation decision by blocking AI search results related to queries about President Donald Trump and dementia, producing only standard web results for this sensitive topic. 5. Alpha School in Texas uses AI-driven software for academic instruction in a novel model focused on individual pacing but charges a high tuition, sparking debate on social and pedagogical impacts.

Using AI to help you with Excel? Have you struggled with OpenAI's "help" at times? It's now easier to use Claude instead of ChatGPT. "With Claude outperforming OpenAI models in Excel and PowerPoint tasks, it could just be a matter of time before alternatives start to become available within Microsoft’s productivity suite." https://lnkd.in/g89fDsbJ Don Tomoff, MBA Bill Tomoff