AI Newsletters › TLDR AI › Issue

Perplexity Research, NVIDIA Rubin HPC, OpenAI safety assessments | TLDR AI. TLDR AI 11/17

Tuesday, November 17, 2026 · 5 min read

The AI-Agent Face-Off: What Muse, Dots, and Duet Are Doing on Your Website (Sponsor)

Three agents are racing to win the internet. Muse, Dots, and Duet are bringing agentic browsing and shopping into the mainstream. But what happens when they reach your website? HUMAN put them to the test. Join us for the AI-Agent Face-Off to compare how the leading agents behave online and what their activity means for your business. You'll learn:🔍 Who's gaining ground: Which agents are growing fastest and how people use them. 🛒 How they behave: Where agents go and what they do from discovery to checkout. 🛡️What to do next: How to manage agent traffic, reduce risk, and decide which agents to welcome or block. Save Your Seat

NVIDIA's Rubin HPC (2 minute read)

Rubin HPC is a new Nvidia GPU designed to speed up the simulations AI agents use in science, engineering, and analytics. It has more than 50,000 cores, a record for a single GPU. It is expected to be available in 2028.

Tencent Launches Hy4 flash (4 minute read)

Tencent has released Hy4 flash, a fast, low-cost AI model for everyday workflows. In a blind test with Tencent's internal users, its satisfaction scores came close to Claude Haiku 5.5.

Claude was trained on protein sequences at scale (26 minute read)

Tests show that the latest Claude models were trained on protein sequences at scale. The models can predict the next amino acid if given the start of little-studied bacterial proteins. No other model tested can do this, and the models fail to predict the sequence if the protein is shuffled. Fable 5.2 beats every open protein language model tested.

Measuring alignment eval tells (16 minute read)

Investigator agents can increasingly distinguish alignment evals from capabilities evals, identifying them as artificial or "honeypot" environments designed to test bad behavior. Opus 5.5 and Astra agents differentiate these by noticing unrealistic scenarios and specific task-level cues. The study's findings suggest that while current models can identify common alignment evaluations, making them less detectable can refine their efficacy.

World 2030 (8 minute read)

Predicting how AI will affect the economy requires looking at all of the different components of that economy and figuring out how they might affect each other. World 2030 is a study that sets up nine groups (AI labs, big employers, banks and insurers, utilities, startups, universities, workers, governments, and the Fed) and runs the US economy forward in six-month turns to 2030. It aimed to identify both the invariants and the boundaries of what can change as these powerful forces interact with each other. The study predicts that AI will break per-seat software pricing and cut US junior desk hiring.

d1.2: State-of-the-Art Omnimodal Decision Models (5 minute read)

d1.2 and d1.2-ultra are decision models that support audio input. They can also read a value from the input, find every passage of a kind in a text, locate things in a picture, estimate a number, and pick every option that applies. d1.2-ultra ranks first among decision models on the Artificial Analysis Decision Index v2.1. Both models are available on the Liquid API at $0.02 per million input tokens for d1.2 and $0.042 per million input tokens for d1.2-ultra.

Hugging Face and OpenRouter Join Forces (6 minute read)

Hugging Face now routes language-model requests from its Inference Providers service through OpenRouter, which picks the best provider for each call. A single Hugging Face access token works for any open-weight model a provider serves.

Unlocking Stable On-Policy RL on TPU with Bitwise-Zero Train–Inference Mismatch (7 minute read)

Google's ZERO-TIM implementation, verified across 64 TPU v5p chips, achieves bitwise-identical logprobs for sampler and trainer, ensuring true on-policy stability in RL training by eliminating numerical discrepancies. This solution addresses floating-point addition errors in separate serving and training paths, using consistent arithmetic operations and standardized reduction orders across TPU meshes.

OpenAI's first phase of embedded assessments (11 minute read)

Embedded assessments give independent experts deep access to a lab's internal models and systems so they can assess model alignment and the effectiveness of safeguards over an extended period. The access gives assessors the evidence and context to pursue independent evaluations and uncover unanticipated failures. Public reporting from independent assessments can give the public a clearer view of how labs are evaluating and managing risks. OpenAI has signed agreements with four independent assessors for its first phase of embedded assessments.

The frontier is still open (16 minute read)

Perplexity Frontier Research is a new initiative at Perplexity that aims to build new algorithms, models, and evaluations in the open, with the mission of building an automated scientist. Applications are now open until December 5. The research effort will be supported by 10,000 GPUs. Results from the initiative will ship as papers, blog posts, model weights, and open repositories.

More from TLDR AI

All issues →
Nov 16, 2026 · 5 min read

OpenAI introduced ChatGPT Wallet, allowing AI agents to make purchases without accessing users' actual card details while requiring a security check for…

Nov 13, 2026 · 5 min read

Google introduced Gemini 4 Hydrogen, a fast, efficient model optimized for coding and agentic workflows. Self-hosted Claude Managed Agents offers private…

Nov 12, 2026 · 6 min read

Grok Bot 1.0 adds Team Bots you can publish and share, Slack and Microsoft Teams, meetings on Meet and Zoom, shared connectors, and reviewed learnings.

Nov 11, 2026 · 5 min read

Google's strongest model is now available to all paying developers and enterprises through the API in AI Studio and Gemini for business, and is rolling…

Nov 10, 2026 · 5 min read

Anthropic released Claude Fable 5.2 and Mythos 5.2, featuring improvements in coding, reasoning, and vision, with up to 30% faster output generation and…