ax@ax-radar:~/curated $ grep -l 'curated=true' sources/
41 srcsignal 72%cycle 04:32

curated · 2026-06-02

▸ 35 items · updated 3m ago
browse by dayclear filter ✕
May 2026
MTWTFSS
1234567368199211010113212281335143815281617179182419472026213222362372462572625272928342936308316
June 2026
MTWTFSS
138235332422532610788299441029112912151371415161718192021222324252627282930
2026-06-02 · Tue
21:35
10d ago
AI HOT (Curated Pool)· aihot-apiZH21:35 · 06·02
→Anthropic Supports Implementation of U.S. AI Executive Order
Anthropic said it supports implementation of a U.S. AI executive order and expects to work with the White House; the post does not disclose the order’s provisions, implementation timeline, or Anthropic’s specific commitments.
#Safety#Anthropic#White House#Policy
why featured
HKR-R passes because Anthropic-White House cooperation hits regulation and safety nerves. HKR-H/K fail: the post gives no order details, implementation timeline, or concrete Anthropic commitment, so this stays low-value but browseable.
editor take
Anthropic backs the U.S. AI order, with no provisions or commitments disclosed; this reads like positioning, not safety work.
HKR breakdown
hook —knowledge —resonance ✓
→ open source
55
SCORE
H0·K0·R1
21:34
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH21:34 · 06·02
→Google DeepMind open-sources a toolkit for scientific agents
Google DeepMind released Science Skills on GitHub for scientific-discovery agent workflows; the post does not disclose the license, benchmark results, or numeric token-efficiency gains.
#Agent#Tools#Google DeepMind#Open source
why featured
Passes HKR-H/K/R: DeepMind, open source, and science agents make it relevant. Missing license, benchmarks, and efficiency data keep it in the 78–84 band, not P1.
editor take
DeepMind put Science Skills on GitHub, but no license or benchmarks; science agents live or die on reproducibility, not a launch tweet.
sharp
DeepMind is staking a cheap claim here: Science Skills is on GitHub for scientific-discovery agents, but the post gives no license, benchmark, or token-efficiency number. Scientific agent tooling has a higher bar than generic agent scaffolding. Users need reproducible protocols, tool-call boundaries, and failure modes, not a loose claim about better token efficiency. I’m skeptical of the framing. DeepMind has earned credibility in scientific AI after AlphaFold, but agent tooling has already been through the LangGraph, LlamaIndex, and smolagents cycle. Without an eval harness or task suite, an open-source repo becomes a polished example pack fast. The GitHub link is a starting gun, not evidence.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
80
SCORE
H1·K1·R1
21:16
10d ago
● P1AI HOT (Curated Pool)· aihot-apiZH21:16 · 06·02
→Claude Code adds dynamic workflows to coordinate multiple subagents in parallel
Claude Code added dynamic workflows that execute JavaScript files at runtime to create and coordinate multiple subagents; each subagent has its own context window, and the feature is described for research, security analysis, and code review tasks.
#Agent#Code#Tools#Anthropic
why featured
HKR-H/K/R all pass: Claude Code gets runtime JS workflows coordinating isolated-context subagents. Anthropic update earns a bump, but this is a feature release rather than a model or platform launch, so it sits in the 78–84 band.
editor take
Claude Code now spawns parallel sub-agents with task-specific instructions — this is the shift from single-threaded assistant to multi-threaded foreman.
sharp
Anthropic's official blog announced dynamic workflows for Claude Code, and both sources covering it are pulling from the same post — no angle divergence here. The core change: instead of plowing through a complex task serially, Claude Code now decides which subtasks can run in parallel, generates a custom instruction set (they call it a harness) for each, and dispatches multiple sub-agents simultaneously. This is directionally similar to what Devin and Factory have been doing, but Anthropic baked it into the terminal-based Claude Code, which lowers the barrier. The blog doesn't include benchmark numbers, doesn't say how many sub-agents can run concurrently, and doesn't clarify whether parallel execution hits your token budget differently. I'd discount the hype a bit — parallel orchestration sounds great, but the real test is subtask decomposition quality and conflict resolution between agents, neither of which the post digs into. What's confirmed: an architectural step forward. What's not: stability on large-scale projects.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
90
SCORE
H1·K1·R1
20:15
10d ago
AI HOT (Curated Pool)· aihot-apiZH20:15 · 06·02
→NVIDIA DGX Station starts shipping to developers and researchers
NVIDIA DGX Station systems have started reaching developers and researchers, and GB300-equipped units are shipping through partners including ASUS, Dell, Gigabyte, HP, MSI, and Supermicro.
#Inference-opt#NVIDIA#ASUS#Dell
why featured
HKR-K and HKR-R pass: GB300 DGX Station is now shipping via six OEMs, but price, performance, and supply volume are not disclosed. This is a small-to-mid hardware update, below featured threshold.
editor take
NVIDIA DGX Station ships GB300 units; pricing and memory are undisclosed, so local inference hinges on procurement friction.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
68
SCORE
H0·K1·R1
19:59
10d ago
AI HOT (Curated Pool)· aihot-apiZH19:59 · 06·02
→Claude Code self-check and feedback loop tips
The title describes Claude Code self-check and feedback loop tips, and the body only says to encode manual checks before handoff; the post does not disclose steps, examples, parameters, or reproducible conditions.
#Code#Agent#Tools#Claude
why featured
Triggers hard-exclusion-6: no data, case, steps, or parameters beyond a Claude Code tip title. HKR-R passes for developer workflow relevance, but HKR-H/K fail, so importance is capped below 40.
editor take
ClaudeDevs gives only a pre-handoff manual-check idea, with no steps or examples; I don’t buy “tips” without reproducible conditions.
HKR breakdown
hook —knowledge —resonance ✓
→ open source
38
SCORE
H0·K0·R1
19:41
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH19:41 · 06·02
→Runway API adds Aleph 2.0 video editing
Runway API now provides Aleph 2.0 video editing for integration into apps, products, and platforms, supporting precise edits on multi-shot videos up to 30 seconds at 1080p while changing only selected portions; the post does not disclose pricing, rate limits, latency, or model availability by region.
#Multimodal#Vision#Tools#Runway
why featured
Runway is a core AI-video player, and Aleph 2.0 exposes partial video editing via API with 30s and 1080p limits. HKR-H/K/R all pass, but this is a mid-weight product update, not a model-class release.
editor take
Runway putting Aleph 2.0 in the API is a product move; 30s 1080p editing is useful, but no pricing or latency keeps it out of real cost plans.
sharp
Runway is pushing video AI toward controllable editing, which is closer to production than another raw generation demo. Aleph 2.0 through the API supports multi-shot videos up to 30 seconds at 1080p, and edits only selected portions. That covers a lot of real work: ad variants, localization, social cuts, and revision loops. The missing pieces are the ones engineers will price first: no pricing, rate limits, latency, or regional availability. Video APIs fail less on capability slides than on queue time, retry behavior, and unit economics under batch load. Pika, Luma, and Veo keep fighting over generation quality; Runway is making a cleaner grab for the post-production workflow. Until it publishes operational constraints, this is an integrable feature, not a dependable pipeline.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
75
SCORE
H1·K1·R1
19:36
10d ago
AI HOT (Curated Pool)· aihot-apiZH19:36 · 06·02
→OpenRouter launches three new Microsoft models
OpenRouter listed three MicrosoftAI models—MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2; the RSS snippet does not disclose parameters, pricing, rate limits, or access conditions.
#Multimodal#Vision#Audio#OpenRouter
why featured
This is a small distribution-channel product update. HKR-K passes on three MicrosoftAI model names and modalities, while HKR-H/R fail because parameters, pricing, access terms, and benchmarks are not disclosed.
editor take
OpenRouter listed 3 Microsoft MAI models, but no pricing or limits are disclosed; routing multimodal is nice, usability remains unproven.
HKR breakdown
hook —knowledge ✓resonance —
→ open source
62
SCORE
H0·K1·R0
19:26
10d ago
AI HOT (Curated Pool)· aihot-apiZH19:26 · 06·02
→Replit and Microsoft Launch Fabric Integration
Replit and Microsoft announced a Fabric integration for organizations to build internal tools, workflows, or data dashboards in Replit and publish them directly to Microsoft Fabric with built-in security, authentication, and governance; the post does not disclose pricing or launch timing.
#Tools#Replit#Microsoft#Product update
why featured
Mid-low product partnership: HKR-K passes for the Replit-to-Microsoft Fabric publishing path, while pricing, launch timing, and capability limits are missing. It misses the 2/3 HKR bar for featured.
editor take
Replit plugs into Microsoft Fabric; pricing and launch timing are undisclosed. Governance-native deployment is the only enterprise hook here.
HKR breakdown
hook —knowledge ✓resonance —
→ open source
64
SCORE
H0·K1·R0
18:27
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH18:27 · 06·02
→Claude Platform Adds CLI Tool
Claude Platform added a CLI that runs every API endpoint from the terminal, calls the Messages API, launches Claude-hosted agents, and pipes results directly into the shell.
#Agent#Tools#Code#Claude
why featured
Claude Platform CLI clears HKR-H/K/R as a practical developer-tooling update, but the post only gives capability scope; install flow, permissions, safety limits, and pricing are not disclosed.
editor take
Anthropic putting every Claude API endpoint behind a CLI is a distribution move: Claude Code gets a native control plane in the terminal.
sharp
Anthropic is making a practical land grab here: the Claude Platform CLI turns API calls, hosted agents, and shell pipelines into one terminal-native workflow. The concrete hook is broad: every API endpoint, Messages API calls, Claude-hosted agents, and direct piping into the shell. That fits Claude Code better than another IDE surface, because the developer already lives in terminals for tests, logs, deploy scripts, and repo surgery. I like the move, but the missing enterprise details matter. The snippet gives no pricing, permission model, audit trail, or sandbox boundary. A CLI that can launch agents and pipe outputs into shell is powerful; it is also exactly where sloppy credentials and accidental execution become expensive. OpenAI and Google have chased developer surfaces through IDEs and SDKs; Anthropic is pushing closer to the Unix muscle memory.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
74
SCORE
H1·K1·R1
18:00
10d ago
AI HOT (Curated Pool)· aihot-apiZH18:00 · 06·02
→NVIDIA releases self-evolving Hermes agent
NVIDIA released a self-evolving Hermes agent for enterprise AI; the post does not disclose model parameters, training mechanisms, launch timing, or pricing.
#Agent#NVIDIA#Nemotron Labs#Product update
why featured
Hard-exclusion applies for marketing-only substance: it gives NVIDIA’s Hermes agent name and enterprise AI positioning, but no mechanism, availability, or pricing. HKR-H/K/R all fail, so it is capped below 40.
editor take
NVIDIA released Hermes for enterprise AI, with no params or pricing disclosed; “self-evolving” needs mechanics, not vibes.
HKR breakdown
hook —knowledge —resonance —
→ open source
39
SCORE
H0·K0·R0
17:56
10d ago
AI HOT (Curated Pool)· aihot-apiZH17:56 · 06·02
→OpenClaw partners with Microsoft to enter enterprise ecosystem
OpenClaw announced a partnership with Microsoft to bring OpenClaw into the Microsoft and Windows ecosystem; the post does not disclose deployment details, security mechanisms, pricing, or rollout timing.
#Agent#Tools#OpenClaw#Microsoft
why featured
HKR-H and HKR-R pass because Microsoft/Windows distribution matters for agent tools. HKR-K fails: no deployment path, safety mechanism, or pricing, so this stays below featured.
editor take
OpenClaw enters Microsoft and Windows ecosystems; deployment, security, and pricing are undisclosed, so don’t score enterprise readiness yet.
HKR breakdown
hook ✓knowledge —resonance ✓
→ open source
64
SCORE
H1·K0·R1
17:30
10d ago
AI HOT (Curated Pool)· aihot-apiZH17:30 · 06·02
→GitHub Copilot App: An Agent-Native Desktop Experience
GitHub announced the Copilot app at Microsoft Build 2026 and positioned it as an agent-native desktop experience; the RSS snippet does not disclose the feature list, pricing, or release timeline.
#Agent#Tools#Code#GitHub
why featured
HKR-H and HKR-R pass because a GitHub Copilot desktop app changes the coding-agent entry point. HKR-K fails: the body lacks features, pricing, and launch timing, so this stays below featured.
editor take
GitHub announced a Copilot desktop app; no pricing or timeline disclosed, so agent-native is still mostly packaging.
HKR breakdown
hook ✓knowledge —resonance ✓
→ open source
68
SCORE
H1·K0·R1
17:12
10d ago
AI HOT (Curated Pool)· aihot-apiZH17:12 · 06·02
→NVIDIA NemoClaw tutorial for deploying Hermes Agent
NVIDIA’s tutorial shows how to deploy NousResearch Hermes Agent with NemoClaw and OpenShell, connect it to Slack, Outlook, GitHub, and NVIDIA Developer Forums, and convert chat corrections into reusable skills that persist across rebuilds while private data stays behind runtime policies.
#Agent#Tools#Memory#NVIDIA
why featured
HKR-K/HKR-R pass because the post gives a concrete agent deployment path and persistent-skill mechanism. HKR-H fails; this is a vendor tutorial, not a major model or platform release.
editor take
NVIDIA connects Hermes Agent to 4 tools; no evals or isolation details disclosed, and persistent skills can become persistent mess.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
66
SCORE
H0·K1·R1
17:08
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH17:08 · 06·02
→Google DeepMind releases Gemini multi-agent research system
Google DeepMind introduced Co-Scientist, a Gemini-based multi-agent system that generates, debates, and evolves scientific hypotheses; the post does not disclose the Gemini version, benchmark results, access model, or release timeline.
#Agent#Reasoning#Google DeepMind#Gemini
why featured
HKR-H/K/R all pass, but model version, eval results, and availability are not disclosed. This fits a strong research/product release, not the 85+ must-write band.
editor take
Only headline-level detail: Gemini Co-Scientist sounds like a research agent, but no model version, evals, or access date means no discovery credit yet.
sharp
Google DeepMind is wrapping Gemini as Co-Scientist, and the dangerous part is how easily “scientific discovery” gets flattened into an agent demo. The snippet only says it can generate, debate, and evolve hypotheses. It gives no Gemini version, benchmark, expert baseline, access model, or release timeline. Those missing fields matter because research agents do not fail at role-play; they fail at producing testable, reproducible hypotheses that save domain experts real experimental cycles. I like the direction. DeepMind has earned credibility with AlphaFold and AlphaGeometry. But Co-Scientist, as disclosed here, reads like Gemini plugged into a hypothesis loop, not an auditable discovery system. Without wet-lab cycles, hit rates, or negative examples, this is narrative placement rather than evidence.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
80
SCORE
H1·K1·R1
16:45
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH16:45 · 06·02
→Claude Code Team Practice: How Agentic Coding Changes Engineering Organizations and Processes
The Claude Code engineering team described process changes after making agentic coding the default at Code w/ Claude SF 2026: JIT planning, asking Claude first for context collection, Claude handling style and tests in code review, and humans focusing on legal and safety judgments.
#Agent#Code#Tools#Claude
why featured
First-party Claude Code workflow post with concrete engineering mechanisms and strong HKR-H/K/R fit. It is not a model or major product release, so it stays in the 78–84 band.
editor take
Claude Code making agentic coding the default matters more than another SWE-bench bump; once workflow changes, an IDE plugin is too small a product.
sharp
Anthropic’s sharp move here is pushing Claude Code from coding assistant into engineering protocol. The mechanisms are concrete: JIT planning, asking Claude first for context gathering, Claude handling style and tests in review, and humans keeping legal and safety calls. I buy the direction, not the whole Anthropic wrapper. The article gives no team size, defect rate, review latency, or rollback data, so this is not yet a reproducible operating model. Cursor and GitHub Copilot still fight for the editor surface; Claude Code is claiming task slicing, context collection, and PR gating. That moves software engineering pressure from autocomplete into workflow ownership, which hurts the toolchain vendors more than another benchmark chart.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
80
SCORE
H1·K1·R1
16:25
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH16:25 · 06·02
→OpenAI Codex releases Python SDK for direct app integration
OpenAI Codex released a Python SDK with the install command pip install openai-codex, and the snippet says it can reuse the Codex login state; the post does not disclose API pricing, model versions, or rate-limit conditions.
#Agent#Code#Tools#OpenAI
why featured
HKR-H/K/R pass: a Codex SDK for embedded app use is practical and discussable. Sparse sourcing keeps it in the mid-weight product-update band: package and auth are given, but price, model, and rate limits are not.
editor take
The sharp bit is login-state reuse, not the pip package; without pricing, model IDs, or limits, this smells like distribution probing, not a stable API.
sharp
OpenAI is pushing Codex closer to embeddable app infrastructure, but the post does not support the “top coding and image agent” leap. The hard facts are only `pip install openai-codex` and reuse of the Codex login state; pricing, model version, and rate limits are absent. For builders, login-state reuse is the spicy part because it bypasses the usual API-key procurement and permission path, and puts a ChatGPT/Codex session inside local tooling. Cursor and Claude Code own IDE or CLI entry points; OpenAI is testing whether third-party apps can carry Codex as a built-in runtime. I would not treat this as production plumbing until metering and limits are explicit.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
74
SCORE
H1·K1·R1
16:22
10d ago
● P1AI HOT (Curated Pool)· aihot-apiZH16:22 · 06·02
→OpenAI launches Codex Sites feature to turn ideas into interactive websites
OpenAI launched Codex Sites, which turns work, ideas, and plans into an interactive website or app that a team can access through one URL; the feature rolls out first to Business and Enterprise plans, and the post does not disclose pricing or broader availability timing.
#Agent#Code#Tools#OpenAI
why featured
HKR-H/K/R all pass, but the post gives launch framing without pricing, permission boundaries, or quality examples. Treat it as a mid-weight OpenAI product feature, above the featured threshold.
editor take
OpenAI pushed Codex from writing code to generating shareable interactive sites, but it's locked behind Business and Enterprise plans for now.
sharp
OpenAI added a feature called Sites to Codex — you describe an idea, and it spits out an interactive web page you can click around in and share via URL. Both sources are pulling from the same OpenAI blog post, so the core facts are consistent: it's a preview, available to Business and Enterprise users, and pitched as a way to turn static spreadsheets into live dashboards or planning tools. I'd take this with a small grain of salt. Under the hood, this looks like Codex's existing code generation plus a hosting and sharing layer — not a brand-new capability. The CFO-to-scenario-planner example is concrete and tells you who OpenAI is targeting: people inside companies who have ideas but don't write frontend code. What's missing: pricing, any limits on site complexity, and any timeline for individual users. If it's just auto-deploying ChatGPT's code output, the bar is low. If it handles real interactive logic and state management, that's when it gets interesting.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
87
SCORE
H1·K1·R1
16:00
10d ago
AI HOT (Curated Pool)· aihot-apiZH16:00 · 06·02
→DigitalOcean AI cloud service launches on OpenRouter
DigitalOcean’s AI-Native Cloud is now available on OpenRouter, offering inference for popular open-weight models and ranking first in DeepSeek V3.2 output speed and latency, according to Artificial Analysis data cited in the post.
#Inference-opt#DigitalOcean#OpenRouter#Artificial Analysis
why featured
Triggers hard-exclusion-cloud-vendor-promo: this is a managed inference availability post. HKR-K has a concrete speed/latency claim, but no pricing, SLA, or reproducible test conditions, so it is capped at 39.
editor take
DigitalOcean joined OpenRouter and claims No.1 DeepSeek V3.2 speed; pricing and hardware are undisclosed, so treat this as channel expansion.
HKR breakdown
hook —knowledge ✓resonance —
→ open source
39
SCORE
H0·K1·R0
16:00
10d ago
AI HOT (Curated Pool)· aihot-apiZH16:00 · 06·02
→Replit Canvas launches multiple new features
Replit announced multiple Canvas updates and linked to replit.com/canvas; the post does not disclose the specific features, release timing, or plan availability.
#Code#Tools#Replit#Product update
why featured
HKR-H/K/R all fail: the item gives only a Replit Canvas update link, with no feature list, launch conditions, or eligible tiers. Per the 0/3 HKR rule, it is excluded and capped below 40.
editor take
Replit Canvas announced multiple updates, but disclosed no features or plans; this is a link post, not an IDE signal yet.
HKR breakdown
hook —knowledge —resonance —
→ open source
32
SCORE
H0·K0·R0
15:41
10d ago
AI HOT (Curated Pool)· aihot-apiZH15:41 · 06·02
→Gary Marcus: Why Things Will Eventually Fall Apart
Gary Marcus discusses AI system risks in a trustworthy AI column; the RSS snippet cites limits in related mathematical theory and human psychological complexity, but the post does not disclose specific models, experiments, benchmarks, or case details.
#Safety#Gary Marcus#Safety/alignment#Commentary
why featured
Hard-exclusion-zero-sourcing applies: the feed gives an opinion angle with no data, case, experiment, or named system. HKR-H and HKR-R pass, but HKR-K fails, so importance is capped below 40.
editor take
Gary Marcus pins AI economics on no moat and 750,000 X views; I buy the price-war call, not the missing cost curve.
HKR breakdown
hook ✓knowledge —resonance ✓
→ open source
38
SCORE
H1·K0·R1
14:48
10d ago
AI HOT (Curated Pool)· aihot-apiZH14:48 · 06·02
→SenseTime Open-Sources SenseNova-Skills AI Office Skill Suite
SenseTime open-sourced SenseNova-Skills for skill-compatible agents such as OpenClaw and HermesAgent, with four office capabilities: infographic generation, data analysis, PPT creation, and deep research across academic, technical, and social sources.
#Agent#Tools#SenseTime#OpenClaw
why featured
HKR-H and HKR-K pass via the open-source skills-suite angle and 4 named office skills. HKR-R is weak because the post gives no evals, license terms, deployment conditions, or usage data, so this stays a normal product update.
editor take
SenseTime open-sourced SenseNova-Skills with 4 office skills; license and evals are undisclosed, so enterprise permission wiring is the test.
HKR breakdown
hook ✓knowledge ✓resonance —
→ open source
66
SCORE
H1·K1·R0
14:13
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH14:13 · 06·02
→Holo3.1: Fast Local Computer-Use Agents
Holo3.1 releases Qwen-based computer-use agents in 0.8B, 4B, 9B, and 35B-A3B sizes, with FP8, Q4 GGUF, and NVFP4 quantized checkpoints for local inference and a 79.3% AndroidWorld score for the 35B-A3B model.
#Agent#Tools#Inference-opt#H Company
why featured
HKR-H/K/R all pass: Holo3.1 pairs a local computer-use agent with concrete model sizes and quantized checkpoints. It fits the 78–84 band, below major lab model-release weight.
editor take
Holo3.1 makes local computer-use agents feel less like a demo; 79.3% on AndroidWorld is nice, but latency and real-device failures matter more.
sharp
Holo3.1’s move is deployment, not raw intelligence. H Company ships 0.8B, 4B, 9B, and 35B-A3B variants, plus FP8, Q4 GGUF, and NVFP4 checkpoints. That is a serious bid for the local-agent default stack, not another cloud-only demo. The 35B-A3B model posts 79.3% on AndroidWorld, which is strong enough to care about. I still don’t buy the headline until local traces show latency and failure modes. Computer-use agents break on screenshot parsing, click coordinates, app-version drift, and permission popups. OpenAI Operator and Claude Computer Use both hit that wall. The missing data is Q4 end-to-end task time and multi-step crash rate on real devices.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
80
SCORE
H1·K1·R1
14:13
10d ago
AI HOT (Curated Pool)· aihot-apiZH14:13 · 06·02
→Nathan Lambert Leaves Ai2 After 2.5 Years on OLMO and Related Projects
Nathan Lambert announced his departure from Ai2 after more than 2.5 years, during which he worked on open-source model projects including OLMO and Tulu.
#Fine-tuning#Nathan Lambert#Ai2#Allen Institute for AI
why featured
HKR-H/K/R all pass, but the post only confirms the departure and 2.5-year stint; no next role, succession plan, or OLMO/Tulu roadmap change is disclosed. This is relevant open-source AI personnel news, not featured-level impact.
editor take
Nathan Lambert left Ai2 after 2.5 years on OLMO/Tulu; open models lost a rare trainer-writer-evangelist combo.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
68
SCORE
H1·K1·R1
14:02
10d ago
AI HOT (Curated Pool)· aihot-apiZH14:02 · 06·02
→MiniCPM-V 4.6 now supports vLLM v0.22.0
OpenBMB announced that MiniCPM-V 4.6 fully supports vLLM v0.22.0, letting users run it by pulling a prebuilt package without custom branches or extra compilation.
#Multimodal#Vision#Inference-opt#OpenBMB
why featured
HKR-K and HKR-R pass: this is a concrete inference-deployment update with a version number and install condition. HKR-H is weak, and the impact is limited to MiniCPM-V/vLLM users, so it stays in 60-71.
editor take
MiniCPM-V 4.6 runs on vLLM v0.22.0 via prebuilt packages; fewer compile traps matter for open multimodal adoption.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
64
SCORE
H0·K1·R1
13:28
10d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH13:28 · 06·02
→Anthropic Expands Project Glasswing Program
Anthropic expanded Project Glasswing to about 150 new organizations across more than 15 countries, covering electricity, water, healthcare, communications, and hardware infrastructure, after an initial group of about 50 partners.
#Code#Safety#Tools#Anthropic
why featured
Anthropic expanded Project Glasswing to about 150 new organizations across 15+ countries, giving HKR-H/K/R enough substance. No concrete safety mechanism or Claude capability change is disclosed, so it stays in the lower featured band.
editor take
Anthropic added 150 Glasswing orgs, staking out governed cyber-AI before clones arrive; finding bugs scales, patch governance will hurt.
sharp
Anthropic is not just bragging about Claude Mythos Preview finding bugs; it is trying to put critical-infrastructure cyber AI inside a controlled club before the cheap copies arrive. The expansion adds about 150 organizations across 15+ countries, after roughly 50 early partners reported more than 10,000 high- or critical-severity flaws. That is a serious number, but it also exposes the old security bottleneck: triage, disclosure, patch review, and deployment windows do not scale like model inference. The loaded line is Anthropic’s 6-to-12-month forecast that other AI labs will reach Mythos-class cyber capability, perhaps without safeguards. I buy the urgency more than the polish. The article gives no false-positive rate, mean time to patch, or patch acceptance rate. Without those, 10,000 findings are either defensive leverage or a fresh liability dump on maintainers.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
75
SCORE
H1·K1·R1
12:53
11d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH12:53 · 06·02
→StepFun releases Step 3.7 Flash as an open-weight model for agentic coding
StepFun released the open-weight Step 3.7 Flash model for fast agentic coding, with tool calling and multimodal understanding, and the model is already available in Kilo alongside MiniMax M3.
#Agent#Tools#Multimodal#StepFun
why featured
HKR-H/K/R pass on the open-weight agentic-coding angle and Kilo availability. Missing benchmarks, size, license, and pricing keep it at the lower featured threshold.
editor take
Step 3.7 Flash landing in Kilo matters more than the open-weight label; no pricing, benchmarks, or context window means this is distribution first.
sharp
Step 3.7 Flash looks like a play for the agentic-coding entry point, not a proof of model strength. The disclosed hooks are open weights, tool calling, multimodal understanding, and availability in Kilo. Missing are parameter count, context window, SWE-bench, pricing, and license terms. For practitioners, the Kilo integration carries more weight than the model-card language, because it puts the model inside the edit-run-tool loop. MiniMax M3 landing in Kilo at the same time makes StepFun’s claim harder to isolate. Open-weight coding models no longer win by saying they call tools. They win by making fewer repo-level mistakes, burning fewer tokens, and surviving messy local environments. Without those numbers, Step 3.7 Flash is mainly a distribution move.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
73
SCORE
H1·K1·R1
09:43
11d ago
AI HOT (Curated Pool)· aihot-apiZH09:43 · 06·02
→Alibaba Cloud Releases AgentScope Java 1.1, Claw, and Other Features
Alibaba Cloud released AgentScope Java 1.1 with Claw, Builder, workspace-driven evolution, and distributed isolation; the post does not disclose pricing, rollout timelines, benchmark results, or implementation details beyond the RSS snippet.
#Agent#Tools#Code#Alibaba Cloud
why featured
HKR-K/R pass: the post names concrete agent-framework mechanisms and targets developer tooling. Price, timeline, and benchmarks are not disclosed, and the vendor-source format keeps it in the small product-update band.
editor take
Alibaba Cloud shipped AgentScope Java 1.1; no pricing, benchmarks, or internals disclosed, and Claw’s shell access needs isolation scrutiny first.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
63
SCORE
H0·K1·R1
08:59
11d ago
AI HOT (Curated Pool)· aihot-apiZH08:59 · 06·02
→Alibaba Cloud Qwen3.7 Models Arrive on Vercel AI Gateway
Alibaba Cloud added Qwen3.7-Plus and Max to Vercel AI Gateway, and users can test their native agent capabilities for free until June 4.
#Agent#Alibaba Cloud#Qwen#Vercel
why featured
HKR-K and HKR-R pass because the post gives model names, gateway, and a free test window. HKR-H is weak; no pricing, limits, or benchmark data are disclosed, so this stays in the small product-update band.
editor take
Qwen3.7-Plus and Max hit Vercel AI Gateway, free until June 4; pricing and context are undisclosed, test SDK friction first.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
64
SCORE
H0·K1·R1
08:31
11d ago
AI HOT (Curated Pool)· aihot-apiZH08:31 · 06·02
→SK Chairman Chey Tae-won says SK Hynix plans to double wafer capacity within five years
SK Hynix chairman Chey Tae-won said the company plans to double overall wafer capacity within five years, with new fabs requiring at least three years and memory supply tightness expected to last until 2030.
#SK Hynix#Chey Tae-won#SK#Product update
why featured
HKR-H/K/R all pass lightly: the story has a 2x capacity hook and concrete timelines. It stays in the lower band because the body discusses overall wafer capacity, not HBM, AI accelerators, pricing, or customers.
editor take
SK Hynix plans to double wafer capacity in five years; with fabs taking 3+ years, AI memory relief won't be quick.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
67
SCORE
H1·K1·R1
04:42
11d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH04:42 · 06·02
→To Avoid Paying $120, I Turned a Computer Cleaner into an Open-Source Skill
The author open-sourced a cross-platform AI cleaning skill for Mac and Windows, generating interactive HTML reports from file scans; in a test, it freed nearly 120GB, compared with CleanMyMac identifying 15.8GB.
#Agent#Code#Tools#CleanMyMac
why featured
This is not a platform-level release, but HKR-H/K/R all land through the $120 replacement hook, concrete scan/report mechanism, and 120GB test result. It fits the practical open-source tool band near the featured threshold.
editor take
Turning a $120 cleanup app into an open skill is the sane agent-tools path: small scope, transparent output, and user-verifiable actions.
sharp
This local agent skill works because it restores user control before automation. Codex runs a read-only storage analysis, surfaces Bilibili cache and other candidates, then renders an HTML report with green, yellow, and red deletion tiers. Only after that does it offer safe execution buttons. The reported result is nearly 120GB freed, while CleanMyMac found 15.8GB; that gap is not a UI win, it is a transparency win. I don’t buy the “AI cleanup app” framing. It is closer to an auditable local ops script generator. The risk sits there too: cross-platform deletion advice on Mac and Windows has real blast radius. A good ruleset, dry-run mode, and reversible actions matter more than a smarter model.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
73
SCORE
H1·K1·R1
03:45
11d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH03:45 · 06·02
→StepFun releases Step 3.7 Flash for efficient inference
StepFun released Step 3.7 Flash with a 196B MoE architecture, using multi-matrix factorized attention to cut KV-cache cost to about 22% of DeepSeek models.
#Reasoning#Inference-opt#Agent#StepFun
why featured
HKR-H/K/R all pass: Step 3.7 Flash has concrete specs, not just launch copy, with 196B MoE and ~22% KV-cache cost versus DeepSeek. It is below top-lab flagship weight, so 78 featured.
editor take
Step 3.7 Flash is a serving-economics play: 196B MoE, ~22% DeepSeek KV-cache cost, Apache 2.0, and Fireworks distribution.
sharp
Step 3.7 Flash lands where reasoning models hurt: KV-cache cost, not headline parameter count. The concrete claim is sharp: a 196B MoE using multi-matrix factorized attention to bring KV-cache cost to about 22% of DeepSeek models. For long-context agents and tool loops, that memory bill decides whether a model is usable at scale. Apache 2.0 plus Fireworks AI distribution is the practical move. StepFun is not just posting a domestic model card; it is putting the model where developers can test latency, pricing, and stability. The missing pieces are benchmark scores, active parameter count, throughput, and per-million-token pricing. If that 22% number survives Fireworks or vLLM-style workloads, it says more than another leaderboard win.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
78
SCORE
H1·K1·R1
02:00
11d ago
AI HOT (Curated Pool)· aihot-apiZH02:00 · 06·02
→NVIDIA JetPack 7.2 Supports Memory-Efficient Edge Agent Deployment
NVIDIA released JetPack 7.2 with one-click deployment for the open-source NemoClaw stack and agent skills for Jetson devices; the post does not disclose memory-efficiency metrics or supported Jetson hardware lists.
#Agent#Safety#Memory#NVIDIA
why featured
NVIDIA’s product update has a clear version and edge-agent deployment mechanism, so HKR-K/R pass. Memory-efficiency metrics are not disclosed, and the angle stays in the normal 60–71 product-update band.
editor take
JetPack 7.2 adds NemoClaw one-command deploy and MIG for Jetson; memory-efficiency numbers are missing, so call it edge-agent tooling catch-up.
HKR breakdown
hook —knowledge ✓resonance ✓
→ open source
68
SCORE
H0·K1·R1
00:55
11d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH00:55 · 06·02
→Anthropic Developer Shares a Claude Code Understanding-Verification Workflow
An Anthropic developer shared a Claude Code understanding-verification workflow with 8 steps, using incremental teaching, user restatement, checklists, and quizzes to confirm the human can defend the problem, solution, and impact before moving to the next stage.
#Agent#Code#Tools#Anthropic
why featured
HKR-H/K/R all pass: a concrete Claude Code workflow with an 8-step verification loop and a strong oversight hook. It is a practical tutorial, not a product release, so it sits at the lower featured band.
editor take
Claude Code’s 8-step workflow is refreshingly anti-autopilot: capable agents turn humans into rubber stamps unless the process fights back.
sharp
Claude Code’s workflow lands because it treats human understanding as an artifact, not a vibe. The 8 steps force restatement, checklists, and quizzes before the next stage; that is closer to engineering audit than the usual “agent made a diff, please approve” loop. I buy the direction. Cursor, Devin, and Claude Code have spent the last year stretching autonomous coding sessions, but the first thing that breaks in long runs is not always model skill. It is the human losing the thread. The missing part is measurement: no failure rate, time cost, team size, or before/after data is disclosed. Without that, this is a strong Anthropic process pattern, not proof that understanding-verification scales.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
74
SCORE
H1·K1·R1
00:12
11d ago
AI HOT (Curated Pool)· aihot-apiZH00:12 · 06·02
→Karpathy Shares His Learning Methodology
The title says Andrej Karpathy shared a learning methodology, and the RSS body contains only one short sentence; the post does not disclose specific methods, steps, examples, or reproducible conditions.
#Andrej Karpathy#Commentary
why featured
Triggers hard-exclusion-6: the RSS body is one sentence with no method, example, data, or reproducible condition. HKR-H/K/R all fail, so it is treated as noise.
editor take
Andrej Karpathy is only name-checked; no steps are disclosed, so I don’t buy this as methodology.
HKR breakdown
hook —knowledge —resonance —
→ open source
32
SCORE
H0·K0·R0
00:00
11d ago
FEATUREDAI HOT (Curated Pool)· aihot-apiZH00:00 · 06·02
→The Thriving Ecosystem of Open Models
OpenRouter data shows open-weight models generated 69.1% of token usage since 2025, versus 30.9% for closed models, while share leadership shifted across DeepSeek, MiniMax, Kimi, MiMo, Qwen, Tencent Hy3, Alibaba, and Arcee releases.
#Inference-opt#OpenRouter#DeepSeek#Qwen
why featured
HKR-H comes from the 69.1% vs 30.9% contrast, HKR-K has OpenRouter token-share data, and HKR-R hits open-vs-closed competition. It is a data-backed commentary, so featured low band.
editor take
OpenRouter’s 69.1% is not the whole market, but it says API-first developers now treat open weights as the default shortlist.
sharp
OpenRouter’s numbers puncture the lazy claim that closed models own inference: since 2025, open-weight models produced 69.1% of named token volume, versus 30.9% for closed models. The sample is biased toward API routing. It misses enterprise direct deals, cloud-private traffic, and ChatGPT-style consumer usage, so don’t read it as total market share. But it captures the most price-sensitive operators: developers routing by cost, latency, context, and task fit every day. DeepSeek’s lead gave way to MiniMax, Kimi, Qwen, Tencent Hy3, and Arcee appearances. That churn is the point. Open weights are winning here because switching costs are brutally low. Closed labs can still defend premium capability, but default API traffic is getting eaten by price-performance tables.
HKR breakdown
hook ✓knowledge ✓resonance ✓
→ open source
76
SCORE
H1·K1·R1

more

✕

feeds

hot eventsdaily columnall postspodcastscuratedX monitorsavedsourcesagent access

admin

usagesystemcurationiterationsusers