aipoc-lab
/katEcosystem/temAgents · LLM · Open Source/podMicrosoft · NVIDIA/typNews

AI Today: Nemotron and PDD Kit

Authorship and approval

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

Published and updated

Published: Updated:

State of knowledge as of 2026-11-09. Source material collected on 2026-11-06. These are two independent topics related to working with AI. Chris Alexiuk from NVIDIA promotes open Nemotron models and shows developers how to build and deploy applications and agents. Microsoft Research describes PDD Kit, which helps formally verify the code of applications created by agents, and when a proof is impossible, prepares tests marked as empirical.

Highlights Today (TL;DR)

  • NVIDIA Nemotron: Chris Alexiuk promotes open models, including Lightning for frequent, repetitive tasks and Ultra for long-running agents.
  • PDD Kit: Microsoft Research tested the tool on a full-stack Rust application with approximately 5000 lines.

Chris Alexiuk promotes NVIDIA open Nemotron models

In a material published on November 5, NVIDIA introduced Chris Alexiuk, a senior product research engineer in deep learning, who teaches developers how to use open language models. Alexiuk publishes YouTube content on building, fine-tuning, and evaluating LLM-based applications and agents, as well as deploying them to production. He believes that full openness should include model weights, the data used to train it, and training recipes. In his view, such access is a prerequisite for the democratization of AI. The engineer promotes the NVIDIA Nemotron family, described in the material as open, multimodal models designed for long-running, self-improving, and specialized agents. Among application examples, Alexiuk mentions Nemotron 3.5 Lightning for repetitive tasks executed frequently and Nemotron Ultra for long-running autonomous agents, deep research, and enterprise tasks. The article does not provide technical details of these models. Alexiuk became interested in machine learning during computational physics classes. His first project was an image classifier that recognized a friend's distinctive parrot on a camera. Currently, he also hosts streams dedicated to Nemotron in the NVIDIA Nemotron Labs series. For AI builders: An agent-building team can check Nemotron variants described for execution tasks and long-running work, and then test them on their own task. The material does not provide memory or hardware requirements, license, price, or Polish language support — these details must be verified before deployment.

PDD Kit verifies agent-written application code

Microsoft Research described PDD Kit — a tool for formally checking applications created by AI agents — in a November blog post. The tool helps the team write down requirements as specifications in the Verus language, and then generates proofs that the code meets this specification for every input. Where formal proof is not possible, it prepares tests; their results are marked as empirical rather than as proof of correctness. The team approves requirements in a document acting as an intermediary between the intent description and the specification and tests. PDD Kit also points out elements whose correct operation is taken on trust and asks for approval of each one. Microsoft tested the tool on Pet Store, a full-stack application written in Rust, numbering approximately 5000 lines. The backend uses Axum and SQLite, and the interface uses Leptos compiled to WebAssembly. In the described example, the team focused on logging and two requirements: a token is to be issued only after providing the correct password, and an admin token exclusively to administrator accounts. According to the post, formal verification helped strengthen the code and detected existing bugs; the proof showed no login errors against the established specification. Work starts with questions about intent before the agent modifies the code. PDD Kit analyzes existing code, explains elements related to the requirement, and helps clarify its meaning. Microsoft emphasizes that testing alone does not cover all possible inputs, and tests written by the agent may check what the agent built instead of what the team asked for. However, the formal proof applies exclusively to the recorded specification; the part of the application that cannot be covered by it remains verified by tests. For AI builders: a team maintaining a Rust application can select login logic, write down requirements, and check which of them can be formally verified. The post does not provide the price, license, or availability of PDD Kit outside of integration with Chopin — these details must be determined before planning the use of the tool.

Frequently Asked Questions (FAQ)

What Nemotron model applications does Chris Alexiuk point out?

Alexiuk believes that full openness includes model weights, training data, and training recipes. In the material, he promotes the Nemotron family and points to Nemotron 3.5 Lightning for frequent, repetitive tasks, and Nemotron Ultra for long-running autonomous agents, deep research, and enterprise tasks.

How does PDD Kit verify the code of applications created by agents?

PDD Kit records requirements in the Verus specification and generates proofs that the code satisfies it for every input. When proof is not possible, the tool prepares tests whose results are marked as empirical rather than as a proof of correctness.

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 · Open Source · Tooling/podCommunity · NVIDIA/typNews

AI Today: llama.cpp 1.0.0 and Shield

Discover the latest AI updates including llama.cpp 1.0.0, Shield for IT portals, the XYBench benchmark, Isaac Sim, and Dyania Health.

/katBusiness/temAgents · AI in Business · LLM/podAnthropic · NVIDIA/typNews

AI Today: Claude Fable 5.2 and Mythos 5.2

Discover Claude Fable 5.2 with double the task completion rate, new Anthropic research programs, and global AI partnerships.

/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.

/katEcosystem/temAutomation · LLM · Open Source/podCommunity · NVIDIA/typNews

AI Today: Kolibri and NVIDIA Kumo Relational

Discover Kolibri, a sovereign German mixture-of-experts model, and NVIDIA Kumo Relational for direct database prediction.