Friday, October 16, 2026

LLMs miss key decision factors when asked in advance, preprint reports

Comparing multiple generated answers exposed consequential disagreements that advance predictions and single-response analysis failed to capture, the authors say.

• Can LLMs Anticipate Their Cruxes?

  arxiv.orgOct 16, 2026, 10:08 PMUTC

Source details

Source

arxiv.org

Updated

Oct 17, 2026, 1:20 AMUTC

Sourcegraph deprecates Deep Search’s old context-filter key

AI-generated illustration

The replacement serves both Deep Search and Agentic Batch Changes. Existing rules still work, though the company’s pages give different migration guidance.

• Sourcegraph, Deep Search context filters move to agents.contextFilters, October 15, 2026

  sourcegraph.comOct 16, 2026, 2:07 PMUTC

• Sourcegraph docs, Agent Context Filters (undated)

  sourcegraph.comOct 16, 2026, 2:07 PMUTC

Source details

Source

sourcegraph.com

Updated

Oct 17, 2026, 1:22 AMUTC

Bad agent skills did not trip up every model equally, preprint finds

AI-generated illustration

Repairs improved all three models tested on 49 tasks, but their responses to flawed instructions differed sharply.

• arXiv abstract page, Why Static Skill Inconsistencies Do Not Equate to Runtime Risks: A Behavioral Analysis of LLM Agent Execution

  arxiv.orgOct 16, 2026, 11:11 AMUTC

• Same paper, HTML full text (same authors, not an independent source)

  arxiv.orgOct 16, 2026, 11:11 AMUTC

Source details

Source

arxiv.org

Updated

Oct 17, 2026, 1:22 AMUTC

Google reports 40-fold error reduction in Willow quantum experiment

AI-generated illustration

The test targets the costly preparation of magic states, but Google says the technique still needs to be demonstrated at larger scale.

• Google (blog.google), Bringing the magic: Google breaks open a bottleneck in quantum computing

  blog.googleOct 16, 2026, 9:13 AMUTC

Source details

Source

blog.google

Updated

Oct 17, 2026, 1:22 AMUTC

Agent harnesses can lose more correct answers than they gain, preprint finds

AI-generated illustration

AlphaDiana compares models with and without agent software, tracking accuracy alongside the output tokens spent on unsuccessful attempts.

• arXiv, AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks (abstract page)

  arxiv.orgOct 16, 2026, 8:08 AMUTC

• arXiv, same paper, full text (v1)

  arxiv.orgOct 16, 2026, 8:08 AMUTC

Source details

Source

arxiv.org

Updated

Oct 17, 2026, 1:22 AMUTC

GitHub Copilot in Slack and Teams can now work across several repositories at once

AI-generated illustration

GitHub says its Copilot assistant in Slack and Microsoft Teams can now handle work that spans more than one repository inside a single conversation.

• GitHub Changelog, Multi-repo support in Slack and Teams, 15 October 2026

  github.blogOct 16, 2026, 4:06 AMUTC

Source details

Source

github.blog

Updated

Oct 16, 2026, 9:41 PMUTC

Clerk’s impersonation guide lists five free uses a month—and an Electron caveat

AI-generated illustration

The Electron page directs developers to adapt a Next.js example. Its October 15 update label does not establish a feature launch.

• Clerk, Build a custom flow for handling user impersonation

  clerk.comOct 16, 2026, 1:07 AMUTC

Source details

Source

clerk.com

Updated

Oct 17, 2026, 1:22 AMUTC