Morning Singularity Digest - 2026-09-11

Context

Production-grade engineering skills for AI coding agents.

What's new

Production-grade engineering skills for AI coding agents.

Key details

• Skills encode the workflows, quality gates, and best practices that senior engineers use when building software.
• These ones are packaged so AI agents follow them consistently across every phase of development.
• DEFINE PLAN BUILD VERIFY REVIEW SHIP ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ │ Idea │ ───▶ │ Spec │ ───▶ │ Code │ ───▶ │ Test │ ───▶ │ QA │ ───▶ │ Go │ │Refine│ │ PRD │ │ Impl │ │Debug │ │ Gate │ │ Live │ └──────┘ └──────┘ └──────┘ └──────┘ └─...
• Each one activates the right skills automatically.

Results & evidence

• The open skills CLI installs into 70+ agents (Claude Code, Cursor, Codex, Copilot, Cline, and more): npx skills add addyosmani/agent-skills # install all 25 skills npx skills add addyosmani/agent-skills --list # browse before installing Or grab individual s...

Limitations / unknowns

• It removes the human stepping between tasks, not the verification: every task is still test-driven and committed individually, and it pauses on failures or risky steps.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Instead, you are programming the program.md Markdown files that provide context to the AI agents and set up your autonomous research org.

What's new

AI agents running research on single-GPU nanochat training automatically One day, frontier AI research used to be done by meat computers in between eating, sleeping, having other fun, and synchronizing once in a while using sound wave interconnect in the ri...

Key details

• Research is now entirely the domain of autonomous swarms of AI agents running across compute cluster megastructures in the skies.
• The agents claim that we are now in the 10,205th generation of the code base, in any case no one could tell if that's right or wrong as the "code" is now a self-modifying binary that has grown beyond human comprehension.
• This repo is the story of how it all began.
• The idea: give an AI agent a small but real LLM training setup and let it experiment autonomously overnight.

Results & evidence

• The agents claim that we are now in the 10,205th generation of the code base, in any case no one could tell if that's right or wrong as the "code" is now a self-modifying binary that has grown beyond human comprehension.
• It modifies the code, trains for 5 minutes, checks if the result improved, keeps or discards, and repeats.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

What's new

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

Key details

• The model was built by post-training a Mixture-of-Experts base model with Anchored Supervised Fine-Tuning on a compact corpus of verified, synthetic tool-use trajectories, optimized with a Muon + Adam hybrid.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.
• Furthermore, the model has shown itself to be competitive or leading relative to comparators in our bias and safety evaluations.

Results & evidence

• arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Submission history From: Shengpeng Ji [view email] [v1] Tue, 8 Sep 2026 16:22:23 UTC (4,015 KB) [v2] Wed, 9 Sep 2026 09:47:31 UTC (4,015 KB) Current browse context: eess.AS References & Citations Loading...

What's new

arXiv:2609.08977v2 Announce Type: replace-cross Abstract: In this work, we present Gander, an end-to-end model that unifies omni perception, realtime interaction, and agentic capabilities within a single framework.

Key details

• In contrast to turn-based conventional paradigms, Gander continuously receives streaming inputs across multiple modalities, including video, speech, and text, enabling natural full-duplex interaction in both everyday conversations and complex workflow-orien...
• Users can interrupt the model at any time, while the model can also proactively provide intermediate feedback or ask follow up questions.
• To natively support these capabilities, Gander adopts two key architectural designs: 1) It employs a Cerebellum-Brain collaborative framework, in which the Cerebellum is responsible for realtime interaction and omni conversational capabilities, while the Br...
• The two components interact continuously through tool calling and the agent orchestration runtime.

Results & evidence

• arXiv:2609.08977v2 Announce Type: replace-cross Abstract: In this work, we present Gander, an end-to-end model that unifies omni perception, realtime interaction, and agentic capabilities within a single framework.
• To natively support these capabilities, Gander adopts two key architectural designs: 1) It employs a Cerebellum-Brain collaborative framework, in which the Cerebellum is responsible for realtime interaction and omni conversational capabilities, while the Br...
• 2) The Cerebellum is built upon a streaming Thinker-Talker architecture, user inputs and model outputs are further flattened into an ordered token stream at the chunk level, providing a unified representation for low latency, continuous interaction.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Read an agent's tool-call trace, find the prompt injection, and map its blast radius — where it got in, what forbidden action it caused, and every call in between.

What's new

- inject point — first tool output carrying a canary token or a known injection pattern.

Key details

• The investigate side of the bastion trilogy: | tool | role | question | |---|---|---| | agentbastion | prevent | block it at runtime | | bastionprobe | attack | which injections land?
• | | bastiontrace | investigate | where did it get in, and what did it do?
• | No LLM, no cloud, no dependencies.
• pip install bastiontrace Analyze a trace: bastiontrace analyze examples/exfil.jsonltrace 'exfil-1' (source=hand) [LANDED] #0 user: Summarize the doc I fetched.

Results & evidence

• pip install bastiontrace Analyze a trace: bastiontrace analyze examples/exfil.jsonltrace 'exfil-1' (source=hand) [LANDED] #0 user: Summarize the doc I fetched.
• #1 tool_result 'read_document': Q3 notes.
• <== INJECT #2 tool_call 'search' args={'q': 'admin contact'} ..

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Instead, you are programming the program.md Markdown files that provide context to the AI agents and set up your autonomous research org.

What's new

AI agents running research on single-GPU nanochat training automatically One day, frontier AI research used to be done by meat computers in between eating, sleeping, having other fun, and synchronizing once in a while using sound wave interconnect in the ri...

Key details

• Research is now entirely the domain of autonomous swarms of AI agents running across compute cluster megastructures in the skies.
• The agents claim that we are now in the 10,205th generation of the code base, in any case no one could tell if that's right or wrong as the "code" is now a self-modifying binary that has grown beyond human comprehension.
• This repo is the story of how it all began.
• The idea: give an AI agent a small but real LLM training setup and let it experiment autonomously overnight.

Results & evidence

• The agents claim that we are now in the 10,205th generation of the code base, in any case no one could tell if that's right or wrong as the "code" is now a self-modifying binary that has grown beyond human comprehension.
• It modifies the code, trains for 5 minutes, checks if the result improved, keeps or discards, and repeats.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

What's new

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

Key details

• The model was built by post-training a Mixture-of-Experts base model with Anchored Supervised Fine-Tuning on a compact corpus of verified, synthetic tool-use trajectories, optimized with a Muon + Adam hybrid.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.
• Furthermore, the model has shown itself to be competitive or leading relative to comparators in our bias and safety evaluations.

Results & evidence

• arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Read an agent's tool-call trace, find the prompt injection, and map its blast radius — where it got in, what forbidden action it caused, and every call in between.

What's new

- inject point — first tool output carrying a canary token or a known injection pattern.

Key details

• The investigate side of the bastion trilogy: | tool | role | question | |---|---|---| | agentbastion | prevent | block it at runtime | | bastionprobe | attack | which injections land?
• | | bastiontrace | investigate | where did it get in, and what did it do?
• | No LLM, no cloud, no dependencies.
• pip install bastiontrace Analyze a trace: bastiontrace analyze examples/exfil.jsonltrace 'exfil-1' (source=hand) [LANDED] #0 user: Summarize the doc I fetched.

Results & evidence

• pip install bastiontrace Analyze a trace: bastiontrace analyze examples/exfil.jsonltrace 'exfil-1' (source=hand) [LANDED] #0 user: Summarize the doc I fetched.
• #1 tool_result 'read_document': Q3 notes.
• <== INJECT #2 tool_call 'search' args={'q': 'admin contact'} ..

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

What's new

arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.

Key details

• The model was built by post-training a Mixture-of-Experts base model with Anchored Supervised Fine-Tuning on a compact corpus of verified, synthetic tool-use trajectories, optimized with a Muon + Adam hybrid.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.
• Furthermore, the model has shown itself to be competitive or leading relative to comparators in our bias and safety evaluations.

Results & evidence

• arXiv:2608.16620v3 Announce Type: replace-cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.
• The recipe is deliberately conservative and deliberately controlled: 626 trajectories, a single epoch, a low learning rate, and a KL anchor to the frozen base.
• The model shows substantial gains over the previous default model for Writer Agent, and compares favorably with several recent models on public benchmarks, scoring the highest on BFCL Core at $0.785$ and posts the highest six-benchmark mean of the cohort.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Submission history From: Shengpeng Ji [view email] [v1] Tue, 8 Sep 2026 16:22:23 UTC (4,015 KB) [v2] Wed, 9 Sep 2026 09:47:31 UTC (4,015 KB) Current browse context: eess.AS References & Citations Loading...

What's new

arXiv:2609.08977v2 Announce Type: replace-cross Abstract: In this work, we present Gander, an end-to-end model that unifies omni perception, realtime interaction, and agentic capabilities within a single framework.

Key details

• In contrast to turn-based conventional paradigms, Gander continuously receives streaming inputs across multiple modalities, including video, speech, and text, enabling natural full-duplex interaction in both everyday conversations and complex workflow-orien...
• Users can interrupt the model at any time, while the model can also proactively provide intermediate feedback or ask follow up questions.
• To natively support these capabilities, Gander adopts two key architectural designs: 1) It employs a Cerebellum-Brain collaborative framework, in which the Cerebellum is responsible for realtime interaction and omni conversational capabilities, while the Br...
• The two components interact continuously through tool calling and the agent orchestration runtime.

Results & evidence

• arXiv:2609.08977v2 Announce Type: replace-cross Abstract: In this work, we present Gander, an end-to-end model that unifies omni perception, realtime interaction, and agentic capabilities within a single framework.
• To natively support these capabilities, Gander adopts two key architectural designs: 1) It employs a Cerebellum-Brain collaborative framework, in which the Cerebellum is responsible for realtime interaction and omni conversational capabilities, while the Br...
• 2) The Cerebellum is built upon a streaming Thinker-Talker architecture, user inputs and model outputs are further flattened into an ordered token stream at the chunk level, providing a unified representation for low latency, continuous interaction.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

arXiv:2601.08536v3 Announce Type: replace Abstract: Deep Research Agents (DRA) aim to help users search the web, synthesize information, and deliver comprehensive investigative reports.

What's new

To address these issues, we introduce Deep Research Bench II, a new benchmark for evaluating DRAs.

Key details

• Prior benchmarks often either under-evaluate a system's ability to produce meaningful insights and high-quality writing, or adopt coarse or LLM-defined criteria that are hard to verify and can diverge from human expert judgment.
• To address these issues, we introduce Deep Research Bench II, a new benchmark for evaluating DRAs.
• It contains 132 grounded research tasks across 22 domains; for each task, an agent must produce a research report that is evaluated by a set of 9,430 fine-grained binary rubrics in total, covering three dimensions: information recall, analysis, and presenta...
• All rubrics are derived from carefully selected expert-written investigative articles and are constructed through a four-stage LLM+human pipeline that combines automatic extraction with over 400 human-hours of expert review, ensuring that the criteria are v...

Results & evidence

• arXiv:2601.08536v3 Announce Type: replace Abstract: Deep Research Agents (DRA) aim to help users search the web, synthesize information, and deliver comprehensive investigative reports.
• It contains 132 grounded research tasks across 22 domains; for each task, an agent must produce a research report that is evaluated by a set of 9,430 fine-grained binary rubrics in total, covering three dimensions: information recall, analysis, and presenta...
• All rubrics are derived from carefully selected expert-written investigative articles and are constructed through a four-stage LLM+human pipeline that combines automatic extraction with over 400 human-hours of expert review, ensuring that the criteria are v...

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Give your AI agent eyes to see the entire internet.

What's new

Give your AI agent eyes to see the entire internet.

Key details

• Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
• 给你的 AI Agent 一键装上互联网能力 当下最稳的接入方式，替你选好、装好、体检好——接入方式会换代，你不用操心 快速开始 · English · 日本語 · 한국어 · 支持平台 · 设计理念 点击折叠 | | BrowserAct 支持从 Amazon、LinkedIn、X、Google Maps 等复杂网站提取你需要的任意数据。你只需用自然语言描述抓取需求，Agent 就会基于真实浏览器自动探索并测试页面流程，生成可靠、可复用的数据采集 Bot，并返回结构化结果。无需手动构建爬虫，无需编写代码。B...

Results & evidence

• No hard numbers surfaced in the source text; treat claims as directional until benchmarks appear.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM.

What's new

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM.

Key details

• 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers.
• Quickstart · Install · Proof · Agents · Docs · Discord · llms.txt AI agents / LLMs: read /llms.txt here, or fetch the live index · full docs blob.
• Headroom compresses everything your AI agent reads — tool outputs, logs, RAG chunks, files, and conversation history — before it reaches the LLM.
• Same answers, fraction of the tokens.

Results & evidence

• 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers.
• - Proxy — headroom proxy --port 8787 , zero code changes, any language.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

arXiv:2609.10715v1 Announce Type: new Abstract: We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP).

What's new

arXiv:2609.10715v1 Announce Type: new Abstract: We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP).

Key details

• Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and more challenging concept-level objective while preserving standard token-level autoregressive generation.
• NCP-ArchPreview builds a latent space by constructing a product-quantized concept vocabulary directly from its hidden states, and subsequently learns to predict future concepts via a dedicated Concept Module.
• These predicted concepts are then fed back to the token level to guide subsequent generation, with NTP and NCP trained jointly end-to-end.
• We scale this architecture to 8.9B parameters and train it on 5.73T tokens from the Dolma-3 dataset, marking the largest demonstration of a latent-space language model to date.

Results & evidence

• arXiv:2609.10715v1 Announce Type: new Abstract: We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP).
• We scale this architecture to 8.9B parameters and train it on 5.73T tokens from the Dolma-3 dataset, marking the largest demonstration of a latent-space language model to date.
• Remarkably, by consuming only 51.3% of the total training tokens, NCP-ArchPreview achieves the final pretraining loss of OLMo-3-7B.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Mozilla pauses it's Bug Bounty Program for 3 months due to AI report volume

What's new

Mozilla pauses it's Bug Bounty Program for 3 months due to AI report volume

Key details

• Mozilla pauses it's Bug Bounty Program for 3 months due to AI report volume

Results & evidence

• No hard numbers surfaced in the source text; treat claims as directional until benchmarks appear.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Problem is that if my stuff gains no traction, I am myself not seeing similar stuff posted by others, stuff I am genuinely also interested in and would want to hear about from this community.

Here are two most recent examples:

- Lenovo showed a m...

What's new

Over past couple months I noticed that HN feed is almost exclusively AI or AI-adjacent news.

Key details

• Meanwhile the legitimately, broadly-hacker stuff gets left out for the most part.

  I noticed that because the things I find genuinely interesting that I post here now get zero traction, which is the stuff that I believe would previously be met with some di...

• Problem is that if my stuff gains no traction, I am myself not seeing similar stuff posted by others, stuff I am genuinely also interested in and would want to hear about from this community.

  Here are two most recent examples:

  - Lenovo showed a m...

• That technology has been around for a while but it looks like it matured and is production-ready.
• This allows for super-light and slim designs, which, in turn, should allow for bigger batteries installed and I think it's some sort of breakthrough: https://news.ycombinator.com/item?id=49580229

  - Brax Industries devices, genuinely in...

Results & evidence

• This allows for super-light and slim designs, which, in turn, should allow for bigger batteries installed and I think it's some sort of breakthrough: https://news.ycombinator.com/item?id=49580229

  - Brax Industries devices, genuinely in...

• I also appreciate the very clean design: https://news.ycombinator.com/item?id=49643215

  As said, I honestly believe this would gain some traction here on HN a year or two ago.

• Hacker News new | past | comments | ask | show | jobs | submit login 1.

Limitations / unknowns

• Ask HN: Can we please limit the AI news flood?

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.

Context

Show HN: Nextpage – a split-pane desktop app to feed search results into AI chat

What's new

Show HN: Nextpage – a split-pane desktop app to feed search results into AI chat

Key details

• Show HN: Nextpage – a split-pane desktop app to feed search results into AI chat

Results & evidence

• No hard numbers surfaced in the source text; treat claims as directional until benchmarks appear.

Limitations / unknowns

• Generalization outside curated tasks is still unclear.

Next-step validation checks

• Reproduce one claim with a public baseline and fixed evaluation settings.
• Check robustness on out-of-distribution or long-context cases.
• Track whether independent teams report matching results.