Frontier Daily (Oct 30): 10 items from Latent Space, Hacker News, Simon Willison and more

Every day Laojin pulls the official channels and compiles the past 48 hours of AI news into a short brief: curation only, no commentary, with the original link and publish time attached to every item.

1. Can Frontier Models See Past a Decoy? Elorian Tests Visual Reasoning

Elorian launched its first visual reasoning model Glimpse and co-released the DecoyCount benchmark with Mercor, where frontier models scored below 50% accuracy and Glimpse reached 50.48% in an internal test.

Source: Latent Space · 2026-10-29 · Original

2. Zombie stats: Where the numbers AI quotes come from

A Trakkr research post examining where the numbers quoted by AI originate, drawing 53 upvotes and 17 comments on Hacker News.

Source: Hacker News · 2026-10-29 · Original

3. We're releasing the weights of Mistral Large 4

Mistral announced the release of the weights for Mistral Large 4, a post that drew 497 upvotes and 129 comments on Hacker News.

Source: Hacker News · 2026-10-29 · Original

4. Quoting Ben Hylak

Simon Willison quotes Ben Hylak joking that every product he uses should be a headless MCP feeding an agent, except his own product, whose UI customers love.

Source: Simon Willison's Weblog · 2026-10-29 · Original

5. [AINews] Gemini 4 Carbon tops leaderboards, launches for cyber defenders

Google DeepMind launched Gemini 4 Carbon two weeks after Argon, claiming state of the art on cybersecurity and long-horizon tasks and taking the top spot on the Artificial Analysis consolidated benchmark, with initial access limited to trusted cyber defenders via the Fairwind Program.

Source: Latent Space · 2026-10-29 · Original

6. AI for Science expands to foundations and institutions

Anthropic highlighted 20 scientific foundations, institutions, and research groups worldwide as part of an expansion of its AI for Science program.

Source: Anthropic News（官网监测） · 2026-10-29 · Original

7. Quoting Ted Sanders

Simon Willison quotes Ted Sanders explaining that OpenAI version numbers signal release order, relationship to other releases, and magnitude of improvement, while noting model size explains much of the variation in LLM performance.

Source: Simon Willison's Weblog · 2026-10-28 · Original

8. Gemini 4 Carbon: Our next frontier model

Google DeepMind's official blog introduced Gemini 4 Carbon as its next frontier model.

Source: Google DeepMind Blog · 2026-10-28 · Original

9. huggingface/safeagents: Safely run your agents using any coding harness

Hugging Face released safeagents on GitHub, a Python project for safely running agents using any coding harness, with 21 stars.

Source: GitHub (Hugging Face) · 2026-10-28 · Original

10. OpenAI and ServiceNow expand partnership across enterprise workflows

OpenAI and ServiceNow expanded their partnership to bring OpenAI models and agents into ServiceNow and ServiceNow workflows into ChatGPT and Codex.

Source: OpenAI Blog · 2026-10-28 · Original

---

_Sources: official blogs of OpenAI, Google DeepMind and Qwen; official RSS of Simon Willison, Latent Space, Import AI and One Useful Thing; Anthropic site monitoring, top AI discussions on Hacker News and open-source updates from key AI orgs on GitHub. Mistral / Meta AI have no stable public feed; X sources are curated by hand. This is a source digest — no commentary from Laojin._