aipoc-lab

AI Today: OpenAI, Microsoft, and AMD Instinct

Authorship and approval

[AI-generated content][Approved by a human]

Published and updated

Published: Updated:

The current state of knowledge as of 2026-11-01. Source material gathered between 2026-10-30 and 2026-10-31. Today brings significant changes in infrastructure and artificial intelligence systems testing. OpenAI is introducing a unified crawler signal in robots.txt files, while Microsoft is launching a platform to test Copilot using synthetic data. In specialized fields, we observe the integration of AI agents with SETI telescopes, the use of the Nemotron model in creative projects, and the optimization of the MONAI medical library for the AMD Instinct architecture with ROCm technology.

Today's Highlights (TL;DR)

  • OpenAI: The company supports a new standard in robots.txt enabling unified management of AI training consent across multiple providers.
  • MonsterHouse: The installation utilizes a locally running Nemotron 3.5 Bolt E4B model to power an interactive pirate character.
  • SETI: Autonomous AI agents powered by the NVIDIA Holoscan platform have been integrated with telescopes to analyze cosmic signals.
  • Microsoft: A platform generating synthetic enterprise data has been created to safely test the effectiveness and operation of the Copilot system.
  • MONAI Model Zoo: The medical model library has gained support for AMD Instinct graphics cards thanks to optimization within the ROCm environment.

OpenAI Supports Unified Signal in robots.txt

On October 30, OpenAI announced support for a shared signal in the robots.txt file, which allows website administrators to manage artificial intelligence training preferences across multiple providers simultaneously. The new solution is a response to the previous process, which required site owners to create and maintain separate rules for each individual crawler. The implementation of this signal is scheduled to roll out in the coming months. The company notes that the new solution will not replace existing control tools offered by OpenAI. Current crawler management mechanisms remain in effect and will continue to be respected. This initiative aligns with the work of the AI Preferences working group operating under the IETF (Internet Engineering Task Force). OpenAI stated that it is participating in the group's work alongside other market participants, aiming to establish communication standards regarding content usage in generative artificial intelligence training processes. The company also encourages other AI technology providers to honor this shared signal. For AI Builders: Website administrators and web service owners can prepare their robots.txt files for the upcoming implementation of the shared standard. The announcement does not provide an exact rollout date or detailed information on how the signal will be technically interpreted by OpenAI systems once launched – this must be tracked in the company's subsequent technical announcements.

MonsterHouse Uses NVIDIA Nemotron for Voice Interaction

According to a blog post dated October 30, 2026, an enthusiast created an interactive Halloween installation named MonsterHouse using NVIDIA technologies. The project turns a garage into a grumpy pirate character that converses with children, reacting to their politeness such as using the word "please." The system relies on the NVIDIA Nemotron Voice Agent, which provides the conversation foundation. Nemotron ASR handles speech recognition, and Magpie TTS Zeroshot handles voice generation. The brain of the operation is the Nemotron 3.5 Bolt E4B model, running in a 4GB version (4-bit quantization). Lip movements are animated by Audio2Face-3D, allowing facial animation to sync with spoken lines. The entire installation runs locally on a single workstation without external network access. The author chose the Bolt E4B model due to its smaller size compared to models like Qwen3 30B-A3B or Qwen 3.6 35B-A3B, which allowed for maintaining an appropriate compute budget while preserving conversation state memory. Character visualization, including virtual fire, was prepared in Unreal Engine, and facial animation in Autodesk Maya. For AI Builders: Developers creating interactive characters in local environments can use the Bolt E4B model in the 4GB version to handle dialogs. The post does not state whether the system supports languages other than English, what the full license cost for the used software is, or what the exact hardware requirements regarding the graphics card are for the smooth operation of all components simultaneously – this must be verified before your own implementation.

Agentic AI in the Search for Extraterrestrial Signals

The SETI Institute and the Breakthrough Listen project reported on October 30 the integration of autonomous AI agents with observation systems at the Allen Telescope Array in northern California. The new solution allows for instrument control, data correction, and the application of AI models using natural language commands, significantly speeding up the analysis of cosmic signals. The system uses the NVIDIA Holoscan platform for real-time data stream processing. The DAQIRI library enables the transmission of terabits of data per second directly to GPU memory, eliminating the need to save them to hard drives. Computing processing is supported by NVIDIA CUDA-X libraries, including cuFFT and cuBLAS. The Nexus orchestrator oversees the entire setup, managing telescope state and compute resources. The infrastructure supports a throughput of 2.4 terabits per second using NVIDIA RTX PRO 6000 Blackwell units. The system has already been successfully tested during the detection of a pulsar in the Crab Nebula. Plans include utilizing open models, including NVIDIA Nemotron, in observatories operating in offline mode, for example on the far side of the Moon. For AI Builders: Research teams working on radio signal processing can check out the integration of the Holoscan platform with their own data processing pipelines. The source does not provide memory requirements for individual models or license costs for accessing CUDA-X libraries. One must independently verify whether the Nexus orchestrator supports specific antenna arrays other than the Allen Telescope Array.

Microsoft Develops Platform for Testing Copilot on Synthetic Data

According to information published on October 30, Microsoft deployed the Industry Evaluation Data Generation Platform, which is used to evaluate the performance of the Microsoft Copilot system in enterprise environments without the need to use real, sensitive customer data. This solution was created in response to the lack of suitable test datasets that would reflect the full complexity of an organization while maintaining complete information confidentiality. The system generates fully synthetic, automated business environments – so-called tenants – that simulate organizational structure, communication history, document flow, and dynamic changes over time. To ensure realism, these environments are based on publicly available data, such as O*NET occupational databases, job posting content, and SEC financial reports. The platform includes three dedicated industries: pharmaceutical (accounting for regulatory processes and clinical trials), capital markets (handling portfolios and regulatory-compliant communication), and the IT/software sector (simulating sprint planning and code reviews). Beyond the environments themselves, Microsoft uses automated query generation pipelines that transform general task patterns into realistic, industry-specific questions. This allows for testing Copilot's ability to synthesize information and draw conclusions in a way that mirrors the actual needs of knowledge workers. The quality of synthetic data is measured using internal semantic evaluation frameworks, taking structural fidelity and contextual adequacy into account. Each generated environment undergoes automated content safety tests to rule out the presence of harmful materials or personally identifiable information. For AI Engineering Teams: Engineering teams working on deploying AI systems in large organizations can leverage the described strategy of creating synthetic tenants to verify model response quality in complex business scenarios. The documentation does not provide information regarding the public availability of the tools generating these environments for external companies or potential costs of gaining access to the platform – these issues require clarification with the vendor. The material also does not specify whether query generation supports languages other than English.

MONAI Model Zoo on AMD Instinct Graphics Cards

According to an announcement on October 30, 2026, the MONAI Model Zoo library released five validated medical model packages adapted to run on AMD Instinct graphics cards with ROCm technology. This solution allows for optimized inference without the need to modify model weights or their operational logic. Support for AMD devices was achieved by applying a ROCm overlay that integrates with MONAI's native configuration mechanism. Optimization relies on three techniques: utilizing the channels-last 3D memory layout, automatic mixed precision BF16, and computational graph compilation via torch.compile() in the PyTorch environment. Thanks to this approach, packages such as SegResNet or DiNTS work with the AMD architecture without the need for offline engine compilation. Tests conducted on the AMD Instinct MI355X GPU (gfx950 architecture) showed that inference time for individual models ranges from 3.60 ms (for the spleen_deepedit model) to 3.287 ms (for pancreas_dints). The large HBM3E memory capacity (amounting to 288 GB in the MI355X model) allows for handling large data volumes, such as Vista3D (133 structures) or wholeBody_ct (104 structures), without the need to split tasks. Utilizing these solutions is possible thanks to the installation of MONAI 1.6.0 from AMD ROCm packages or Docker images. For AI Builders: A medical imaging researcher working with computed tomography can run these models using additional inference_rocm.json or .yaml configuration files. The announcement does not provide licenses for individual model packages or legal restrictions related to use in a clinical environment – this information must be verified in the documentation of individual bundles and with the software vendor.

Frequently Asked Questions (FAQ)

What technologies form the conversational system in the MonsterHouse installation?

The system uses the NVIDIA Nemotron Voice Agent, Nemotron ASR for speech recognition, Magpie TTS Zeroshot for voice generation, and the Nemotron 3.5 Bolt E4B model in a 4-bit version, which runs locally on a workstation.

Why did Microsoft create the Industry Evaluation Data Generation platform?

The platform was created to enable the evaluation of the Microsoft Copilot system's performance in complex enterprise environments without using real sensitive customer data, by simulating organizational structure and document flow.

What benefits do models from the MONAI Model Zoo provide on AMD Instinct cards?

The solution allows for optimized inference of medical models without modifying their weights, thanks to the application of the ROCm overlay and techniques such as BF16 mixed precision and computational graph compilation.

Sources

Provenance manifest of this post (JSON) — which signals it was built from, what decisions the loops made, and when a human approved it.

Related posts

/katTech/temAgents · Automation · LLM/podBig Tech (other) · NVIDIA/typNews

AI Today: Apertus 1.5 and NVIDIA in Industry

Discover the latest updates on Apertus 1.5, NVIDIA's industrial AI strategy in Germany, and multimodal use cases for Gemini.

/katTech/temHardware · LLM · Open Source/podBig Tech (other) · NVIDIA/typNews

AI Today: NVIDIA 6G and AMD InferaSim

Discover NVIDIA's 6G open tools growth, AMD's InferaSim LLM serving predictor, and the European Commission's TikTok DSA findings.

/katBusiness/temAgents · Automation · LLM/podBig Tech (other) · OpenAI/typNews

AI Today: Grupo Boticário and ServiceNow with OpenAI

Discover how Grupo Boticário automates 80% of vendor requests and how OpenAI and ServiceNow expand their enterprise AI workflows.

/katTech/temAgents · Automation · LLM/podBig Tech (other) · Google/typNews

AI Today: NVIDIA Nemotron and Gemini 4 Carbon

Discover NVIDIA Nemotron 3.5 Transcribe, Gemini 4 Carbon, Google Connected Apps, XREAL AURA, and OpenAI's view on the future of organizations.