Morning Singularity Digest - 2026-05-22

Context

The best-benchmarked open-source AI memory system.

What's new

The best-benchmarked open-source AI memory system.

Key details

• Caution MemPalace has NO other official websites.
• The ONLY official sources are: - This GitHub repository - The PyPI package - The docs at mempalaceofficial.com ANY other domain (including .tech , .net , or other .com variants) is an impostor and may distribute malware.
• Do not download executables from untrusted sites.
• Details and timeline: docs/HISTORY.md.

Results & evidence

• Important 🚨 Claude Code sessions expire in 30 days w/out auto-save hooks wired!
• Verbatim storage, pluggable backend, 96.6% R@5 raw on LongMemEval — zero API calls.

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

| Topic | What You'll Learn | |---|---| | Token Optimization | Model selection, system prompt slimming, background processes | | Memory Persistence | Hooks that save/load context across sessions automatically | | Continuous Learning | Auto-extract patterns...

What's new

Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

Key details

• Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
• Language: English | Português (Brasil) | 简体中文 | 繁體中文 | 日本語 | 한국어 | Türkçe | Русский | Tiếng Việt | ไทย 182K+ stars | 28K+ forks | 170+ contributors | 12+ language ecosystems | Anthropic Hackathon Winner Language / 语言 / 語言 / Dil / Язык / Ngôn ngữ English | P...
• From an Anthropic hackathon winner.
• A complete system: skills, instincts, memory optimization, continuous learning, security scanning, and research-first development.

Results & evidence

• Language: English | Português (Brasil) | 简体中文 | 繁體中文 | 日本語 | 한국어 | Türkçe | Русский | Tiếng Việt | ไทย 182K+ stars | 28K+ forks | 170+ contributors | 12+ language ecosystems | Anthropic Hackathon Winner Language / 语言 / 語言 / Dil / Язык / Ngôn ngữ English | P...
• Production-ready agents, skills, hooks, rules, MCP configurations, and legacy command shims evolved over 10+ months of intensive daily use building real products.
• ECC v2.0.0-rc.1 adds the public Hermes operator story on top of that reusable layer: start with the Hermes setup guide, then review the rc.1 release notes and cross-harness architecture.

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:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.

What's new

In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.

Key details

• Existing rule-based labelers struggle with the diverse descriptions in clinical reports, while fine-tuning pre-trained language models (PLMs) requires large amounts of labeled data that are often unavailable in clinical settings.
• In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.
• PromptRad reformulates multi-label classification as masked language modeling and incorporates synonyms from the UMLS Metathesaurus into a multi-word verbalizer to enrich category representations.
• By fine-tuning the PLM without additional classification layers, PromptRad requires substantially less labeled data than conventional fine-tuning.

Results & evidence

• arXiv:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.
• Experiments on liver CT (computed tomography) reports show that PromptRad outperforms dictionary-based and fine-tuning baselines with only 32 labeled training examples, and achieves competitive performance with GPT-4 despite using a much smaller model.
• Computer Science > Computation and Language [Submitted on 19 May 2026 (v1), last revised 20 May 2026 (this version, v2)] Title:PromptRad: Knowledge-Enhanced Multi-Label Prompt-Tuning for Low-Resource Radiology Report Labeling View PDF HTML (experimental)Abs...

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:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.

What's new

arXiv:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.

Key details

• Given an egocentric video timestamp, the task requires anticipating the next human-object interaction, including the future active object's bounding box, noun category, verb category, time-to-contact, and confidence score.
• VISTA follows a StillFast-style design that combines object-centric spatial detection with short-horizon temporal context.
• Specifically, a COCO-pretrained Faster R-CNN ResNet-50 FPN detector generates object proposals from the last observed high-resolution frame, while a frozen V-JEPA 2.1 temporal branch extracts clip-level egocentric context from the observed video.
• The temporal representation is injected into the detection pathway through feature modulation and ROI-level context fusion.

Results & evidence

• arXiv:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.
• Specifically, a COCO-pretrained Faster R-CNN ResNet-50 FPN detector generates object proposals from the last observed high-resolution frame, while a frozen V-JEPA 2.1 temporal branch extracts clip-level egocentric context from the observed video.
• Experimental results on the official challenge server show that VISTA achieves first place in the EgoVis 2026 Ego4D STA Challenge.

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

VSCode Feature Request: Permissions, Sandboxing and Update Management (2018)

What's new

VSCode Feature Request: Permissions, Sandboxing and Update Management (2018)

Key details

• VSCode Feature Request: Permissions, Sandboxing and Update Management (2018)

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

| Topic | What You'll Learn | |---|---| | Token Optimization | Model selection, system prompt slimming, background processes | | Memory Persistence | Hooks that save/load context across sessions automatically | | Continuous Learning | Auto-extract patterns...

What's new

Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

Key details

• Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
• Language: English | Português (Brasil) | 简体中文 | 繁體中文 | 日本語 | 한국어 | Türkçe | Русский | Tiếng Việt | ไทย 182K+ stars | 28K+ forks | 170+ contributors | 12+ language ecosystems | Anthropic Hackathon Winner Language / 语言 / 語言 / Dil / Язык / Ngôn ngữ English | P...
• From an Anthropic hackathon winner.
• A complete system: skills, instincts, memory optimization, continuous learning, security scanning, and research-first development.

Results & evidence

• Language: English | Português (Brasil) | 简体中文 | 繁體中文 | 日本語 | 한국어 | Türkçe | Русский | Tiếng Việt | ไทย 182K+ stars | 28K+ forks | 170+ contributors | 12+ language ecosystems | Anthropic Hackathon Winner Language / 语言 / 語言 / Dil / Язык / Ngôn ngữ English | P...
• Production-ready agents, skills, hooks, rules, MCP configurations, and legacy command shims evolved over 10+ months of intensive daily use building real products.
• ECC v2.0.0-rc.1 adds the public Hermes operator story on top of that reusable layer: start with the Hermes setup guide, then review the rc.1 release notes and cross-harness architecture.

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:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.

What's new

In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.

Key details

• Existing rule-based labelers struggle with the diverse descriptions in clinical reports, while fine-tuning pre-trained language models (PLMs) requires large amounts of labeled data that are often unavailable in clinical settings.
• In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.
• PromptRad reformulates multi-label classification as masked language modeling and incorporates synonyms from the UMLS Metathesaurus into a multi-word verbalizer to enrich category representations.
• By fine-tuning the PLM without additional classification layers, PromptRad requires substantially less labeled data than conventional fine-tuning.

Results & evidence

• arXiv:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.
• Experiments on liver CT (computed tomography) reports show that PromptRad outperforms dictionary-based and fine-tuning baselines with only 32 labeled training examples, and achieves competitive performance with GPT-4 despite using a much smaller model.
• Computer Science > Computation and Language [Submitted on 19 May 2026 (v1), last revised 20 May 2026 (this version, v2)] Title:PromptRad: Knowledge-Enhanced Multi-Label Prompt-Tuning for Low-Resource Radiology Report Labeling View PDF HTML (experimental)Abs...

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

It makes software repositories runnable and trustworthy for human, CI, and AI agents.

The problem we kept seeing is that the real truth of how a repo gets set up and run is usually scattered across READMEs, scripts, CI config, env files, and maintainer me...

What's new

Hi HN, We're building Ota.

Ota is open repo readiness infrastructure.

Key details

• It makes software repositories runnable and trustworthy for human, CI, and AI agents.

  The problem we kept seeing is that the real truth of how a repo gets set up and run is usually scattered across READMEs, scripts, CI config, env files, and maintainer me...

• That slows onboarding, causes local and CI drift, and makes automation brittle.

  Ota gives each repo one explicit operational contract for what it needs, how it becomes ready, and how tasks should run.

  The core flow is:

  - `ota doctor` to diagnose what...

• We think repo readiness is its own layer: something between the repo, the developer, CI, and now agents.

  We'd especially love feedback on:

  - whether this problem feels real in your repos - whether the contract model feels like the right abstraction...

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

arXiv:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.

What's new

In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.

Key details

• Existing rule-based labelers struggle with the diverse descriptions in clinical reports, while fine-tuning pre-trained language models (PLMs) requires large amounts of labeled data that are often unavailable in clinical settings.
• In this paper, we propose PromptRad, a knowledge-enhanced multi-label \textbf{prompt}-tuning approach for \textbf{rad}iology report labeling under low-resource settings.
• PromptRad reformulates multi-label classification as masked language modeling and incorporates synonyms from the UMLS Metathesaurus into a multi-word verbalizer to enrich category representations.
• By fine-tuning the PLM without additional classification layers, PromptRad requires substantially less labeled data than conventional fine-tuning.

Results & evidence

• arXiv:2605.20052v2 Announce Type: replace-cross Abstract: Automatic report labeling facilitates the identification of clinical findings from unstructured text and enables large-scale annotation for medical imaging research.
• Experiments on liver CT (computed tomography) reports show that PromptRad outperforms dictionary-based and fine-tuning baselines with only 32 labeled training examples, and achieves competitive performance with GPT-4 despite using a much smaller model.
• Computer Science > Computation and Language [Submitted on 19 May 2026 (v1), last revised 20 May 2026 (this version, v2)] Title:PromptRad: Knowledge-Enhanced Multi-Label Prompt-Tuning for Low-Resource Radiology Report Labeling View PDF HTML (experimental)Abs...

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:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.

What's new

arXiv:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.

Key details

• Given an egocentric video timestamp, the task requires anticipating the next human-object interaction, including the future active object's bounding box, noun category, verb category, time-to-contact, and confidence score.
• VISTA follows a StillFast-style design that combines object-centric spatial detection with short-horizon temporal context.
• Specifically, a COCO-pretrained Faster R-CNN ResNet-50 FPN detector generates object proposals from the last observed high-resolution frame, while a frozen V-JEPA 2.1 temporal branch extracts clip-level egocentric context from the observed video.
• The temporal representation is injected into the detection pathway through feature modulation and ROI-level context fusion.

Results & evidence

• arXiv:2605.20901v1 Announce Type: cross Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis 2026.
• Specifically, a COCO-pretrained Faster R-CNN ResNet-50 FPN detector generates object proposals from the last observed high-resolution frame, while a frozen V-JEPA 2.1 temporal branch extracts clip-level egocentric context from the observed video.
• Experimental results on the official challenge server show that VISTA achieves first place in the EgoVis 2026 Ego4D STA Challenge.

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

To address these problems, we analyze the failure mechanism of linear scalarization from the perspective of gradient dynamics, utilizing the stochastic differential equation (SDE) framework to characterize it as a "Double Dilemma" of drift term deviation an...

What's new

Based on this, we propose a backbone-agnostic optimizer named Conflict-Averse Magnitude-Enhanced Gradient Descent (CAME-Grad).

Key details

• These strategies cannot effectively balance the hard constraints of discriminative clinical supervision with the smoothness requirements of report generation.
• To address these problems, we analyze the failure mechanism of linear scalarization from the perspective of gradient dynamics, utilizing the stochastic differential equation (SDE) framework to characterize it as a "Double Dilemma" of drift term deviation an...
• Based on this, we propose a backbone-agnostic optimizer named Conflict-Averse Magnitude-Enhanced Gradient Descent (CAME-Grad).
• Through conflict-averse direction rectification and magnitude-enhanced energy injection, the algorithm not only ensures geometric validity, but also avoids local optimal solutions.

Results & evidence

• arXiv:2605.22635v1 Announce Type: cross Abstract: While multi-task learning based automatic radiology report generation (RRG) is widely adopted to ensure clinical consistency, most focus on architectural designs yet remain limited to coarse linear scalariza...
• Experiments show that as a universal plug-and-play optimizer, CAME-Grad brings substantial and consistent improvements across eight diverse RRG methods, elevating overall clinical efficacy performance by an average of 2.3\% on MIMIC-CXR and 1.9\% on IU X-Ray.
• Computer Science > Machine Learning [Submitted on 21 May 2026] Title:The Double Dilemma in Multi-Task Radiology Report Generation: A Gradient Dynamics Analysis and Solution View PDF HTML (experimental)Abstract:While multi-task learning based automatic radio...

Limitations / unknowns

• arXiv:2605.22635v1 Announce Type: cross Abstract: While multi-task learning based automatic radiology report generation (RRG) is widely adopted to ensure clinical consistency, most focus on architectural designs yet remain limited to coarse linear scalariza...
• To address these problems, we analyze the failure mechanism of linear scalarization from the perspective of gradient dynamics, utilizing the stochastic differential equation (SDE) framework to characterize it as a "Double Dilemma" of drift term deviation an...

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

The open-source app everyone uses to manage agents at work Quickstart · Docs · GitHub · Discord · Twitter full-tour.webm If OpenClaw is an employee, Paperclip is the company Paperclip is a Node.js server and React UI that orchestrates a team of AI agents to...

What's new

The open-source app everyone uses to manage agents at work Quickstart · Docs · GitHub · Discord · Twitter full-tour.webm If OpenClaw is an employee, Paperclip is the company Paperclip is a Node.js server and React UI that orchestrates a team of AI agents to...

Key details

• Bring your own agents, assign goals, and track your agents' work and costs from one dashboard.
• It looks like a task manager — but under the hood it has org charts, budgets, governance, goal alignment, and agent coordination.
• Manage business goals, not pull requests.
• | Step | Example | | |---|---|---| | 01 | Define the goal | "Build the #1 AI note-taking app to $1M MRR." | | 02 | Hire the team | CEO, CTO, engineers, designers, marketers — any bot, any provider.

Results & evidence

• | Step | Example | | |---|---|---| | 01 | Define the goal | "Build the #1 AI note-taking app to $1M MRR." | | 02 | Hire the team | CEO, CTO, engineers, designers, marketers — any bot, any provider.
• | | 03 | Approve and run | Review strategy.
• - ✅ You want to build autonomous AI companies - ✅ You coordinate many different agents (OpenClaw, Codex, Claude, Cursor) toward a common goal - ✅ You have 20 simultaneous Claude Code terminals open and lose track of what everyone is doing - ✅ You want agent...

Limitations / unknowns

• When they hit the limit, they stop.

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

A collection of DESIGN.md files inspired by popular brand design systems.

What's new

DESIGN.md is a new concept introduced by Google Stitch.

Key details

• Drop one into your project and let coding agents generate a matching UI.
• Copy a DESIGN.md into your project, tell your AI agent "build me a page that looks like this" and get pixel-perfect UI that actually matches.
• DESIGN.md is a new concept introduced by Google Stitch.
• A plain-text design system document that AI agents read to generate consistent UI.

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

This work thus establishes a foundation for benchmarking and advancing both general-purpose and medical LLMs to produce reliable answers in challenging clinical contexts.

What's new

arXiv:2605.21807v1 Announce Type: new Abstract: Across medical specialties, clinical practice is anchored in evidence-based guidelines that codify best studied diagnostic and treatment pathways.

Key details

• These pathways routinely fall short for the long tail of real-world care not covered by guidelines.
• Most medical large language models (LLMs), however, are trained to encode common, guideline-focused medical knowledge in their parameters.
• Current evaluations test models primarily on recalling and reasoning with this memorized content, often in multiple-choice settings.
• Given the fundamental importance of evidence-based reasoning in medicine, it is neither feasible nor reliable to depend on memorization in practice.

Results & evidence

• arXiv:2605.21807v1 Announce Type: new Abstract: Across medical specialties, clinical practice is anchored in evidence-based guidelines that codify best studied diagnostic and treatment pathways.
• Our experiments reveal that even the best-performing baseline (GPT-5.2) correctly answers only 56% of our benchmark with specialized models only reaching 42%.
• Augmenting models with retrieved medical articles improves this performance to up to 82% (using GPT-5.2) highlighting the importance of evidence-grounding for real-world medical reasoning tasks.

Limitations / unknowns

• Most medical large language models (LLMs), however, are trained to encode common, guideline-focused medical knowledge in their parameters.

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

Steve Wozniak did what other college graduation commencement speakers couldn't this year: earn applause when talking about AI.

What's new

During his speech, Wozniak offered reassurance to new graduates who are entering the workforce at the height of the AI revolution.

Key details

• The Apple cofounder took the stage during Grand Valley State University's graduation ceremony earlier this month.
• During his speech, Wozniak offered reassurance to new graduates who are entering the workforce at the height of the AI revolution.
• "You have AI — actual intelligence," Wozniak said.
• The remark garnered laughs and applause from the audience.

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

Organisations using AI to cut headcount are making a short-term trade with long-term consequences.

What's new

Organisations using AI to cut headcount are making a short-term trade with long-term consequences.

Key details

• The ones holding their teams together and investing in how those teams operate with AI are building something more durable.
• There is a version of AI adoption that looks smart on a spreadsheet.
• Fewer people, lower payroll, same output.
• It is the version being quietly executed in boardrooms right now, dressed up in language about efficiency and transformation.

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

Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality

What's new

Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality

Key details

• Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality

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.