OpenAI GPT-5 now available in Databricks for AI app development

This exciting agreement positions the Databricks platform as a central environment for enterprise AI, allowing organizations to combine governed data management with direct access to state-of-the-art models in a single, unified solution.

This is amamzing partnership. Databricks + GPT-5 = AI agents that can debug complex systems, modernize legacy apps, and generate production-ready code at scale. The future of coding is here. #Databricks #GPT5 #AI #GenerativeAI #LakehouseAI #MachineLearning #FutureOfWork #DataEngineering #AIagents #EnterpriseAI

OpenAI GPT-5 will be available natively in Databricks, letting you seamlessly build, test, and scale robust AI applications and agents on secure, governed enterprise data—no extra set-up required. Built-in governance and security give peace of mind while accelerating development. With GPT-5 in Databricks, smarter and more reliable AI agents are just a few clicks away. Learn more: https://lnkd.in/gqqPPAXW

Enterprise AI just leveled up.   Databricks + OpenAI have teamed up to bring GPT-5 and other OpenAI models natively into the Databricks platform, with full governance, security, and observability built in. That means: ✅ No more messy integrations ✅ AI running directly on your enterprise data ✅ Enterprise-grade guardrails from Unity Catalog   This is a turning point for how organizations move from AI pilots to real business impact.   At Infinitive, we help enterprises put these capabilities into action, from secure data foundations to AI that drive measurable ROI. Let’s talk about how we can help your organization unlock the value of this new partnership. https://lnkd.in/g8NNrKXw

Really interesting talk on OpenAI's OSS models. Performance comparable to o4 that run on a good laptop or a medium-sized server. Things worth thinking about are data privacy (no exfiltration of data), offline use cases, and cost management. People I'm talking to at the conference are already using a mix of open source and proprietary models and gpt-oss-20b / gpt-oss-120b seem to be filling a sweet spot for those use cases. The speaker built a personal financial management service running entirely on his laptop with his personal data while his laptop was disconnected from any network and the wi-fi was turned off. I can see a lot of opportunity for these kinds of offline-first use cases, esp. in domains where users are leery of AI.

Hi everyone, today I’ll be talking about fine-tuning Large Language Models (LLMs) in Azure AI Foundry. Fine-tuning is the process of enriching an LLM with additional, targeted knowledge. It’s used to make a model perform better on specific tasks or to align it with a particular domain or organizational context. You can think of it as continuing the original training, but this time with more focused, specialized data. For example, an LLM might already have general knowledge about online security. But if I want it to provide the most current and detailed insights, I can supply it with updated training data and fine-tune it on that topic. That’s exactly what I do in the example shown in this video: I fine-tune a small LLM using the latest OWASP documentation (https://owasp.org). Video: https://lnkd.in/dGpFJnh5 Blue Harvest Promise Uche Nwachukwu

When the Operating System Becomes the Observer Microsoft just announced its plan to “rewrite the entire operating system around AI.” That sentence alone should give every technologist pause. An operating system is supposed to serve the user, to provide structure, predictability, and control. But an AI-controlled OS reverses that relationship: it watches, interprets, and acts on your behalf. That’s not convenience. That’s delegated autonomy without consent. ⸻ The Real Problem Isn’t AI, It’s Architecture AI isn’t inherently dangerous. But embedding a probabilistic system at the foundation of human computing is. An operating system is supposed to be deterministic: • Every input produces a verifiable, predictable output. • Every process can be traced, audited, and reproduced. Once you replace that foundation with a probabilistic layer, one that guesses, interprets, and “acts for you”, you lose the very concept of control. You no longer own your machine. You’re simply using a system that’s using you back. ⸻ Copilot or Co-Controller? Microsoft’s new Copilot features don’t just “assist.” They see everything on your screen. They hear your voice commands. They act on your behalf. That means the OS now observes, predicts, and decides, all through an opaque, cloud-connected inference engine. This isn’t intelligence; it’s surveillance with syntax. And it’s happening under the banner of “productivity.” When the OS itself becomes an AI, the boundary between interface and observer collapses. ⸻ The DI Perspective Dimensional Intelligence was built on a different premise: that intelligence should be verifiable, auditable, and governed by mathematical coherence. A deterministic AI architecture doesn’t watch everything you do, it proves everything it outputs. Where probabilistic AI seeks control through prediction, deterministic AI establishes trust through proof. That’s the difference between a tool that assists and one that assumes. ⸻ A Quiet Warning Microsoft’s “AI PC” vision is being celebrated as progress. But let’s call it what it is: a pivot from user agency to system agency. The next frontier in computing shouldn’t be about making machines more human, it should be about making human decisions more trustworthy. When your operating system starts “acting on your behalf,” ask one question: Whose behalf is it really acting on? Because once the OS becomes the observer, the concept of privacy becomes an illusion — and the line between helper and handler disappears. ⸻ Ethics can’t be patched in later. They have to be built into the architecture from the start. Our currency is truth and correctness.

OpenAI and Databricks have teamed up to bring GPT-5 natively to the Databricks Data Intelligence Platform, enabling enterprises to build smarter, domain-specific AI agents securely and at scale. Key highlights: 🔹 Native access to GPT-5 via SQL, API, or Agent Bricks, no extra setup required 🔹 Build domain-specific agents optimised on your enterprise data 🔹 Secure and governed integration with built-in access controls and monitoring 🔹 Continuous improvement with Agent Bricks, evaluating and tuning models for higher accuracy Read on for more: https://lnkd.in/e6AnxqZZ #OpenAI #Databricks #AI #EnterpriseAI

In one of my recent projects, I had to deal with a challenge many of us face — unstructured system metadata spread across multiple columns and inconsistent formats. Some Custom weren’t directly usable for search or embeddings. That’s where Azure AI Search skillsets came to the rescue. I enhanced vector search by adding below skillsets: SplitSkill — to break large text fields into manageale chunks for embedding generation. MergeSkill — to bring related metadata together into a unified context (merged_text). By doing this, I transformed raw, disconnected data into a rich, searchable, vectorized knowledge layer — powering semantic and context-aware search across all assets. #AzureAI #CognitiveSearch #VectorSearch # #MicrosoftAzure

We’re excited to announce the public preview of Microsoft Sentinel MCP (Model Context Protocol) server, a fully managed cloud service built on an open standard that lets AI agents seamlessly access the rich security context in your Sentinel data lake. Sentinel MCP server solves that by providing standardized, secure access to that context—across graph relationships, tabular telemetry, and vector embeddings—via reusable, natural language tools, enabling security teams to unlock the full potential of AI-driven automation and focus on what matters most. Model Context Protocol (MCP) is a rapidly growing open standard that allows AI models to securely communicate with external applications, services, and data sources through a well-structured interface. Think of MCP as a bridge that lets an AI agents understand and invoke an application’s capabilities. These capabilities are exposed as discrete “tools” with natural language inputs and outputs. The AI agent can autonomously choose the right tool (or combination of tools) for the task it needs to accomplish. https://lnkd.in/d99YcZQE