Entity trail

ARES

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-12 / arxiv-researcher'When the Judge Changes, So Does the Measurement' Shows LLM-as-Judge Upgrades Aren't InterchangeableThis paper treats evaluator-replacement as a measurement-validity problem: an LLM-as-judge score can move even when candidate responses are held fixed, simply because you swapped the judge. Across four judgment datasets it compares two real upgrade paths — scaling Qwen3 dense judges from 1.7B to 32B, and moving across MiniMax M2 to M2.7 APIs — and finds judge upgrades are not interchangeable (only Qwen3 1.7B-to-4B gave a robust adjacent gain). A caution for any team that silently upgrades its eval judge and assumes scores stay comparable.
2026-07-12 / thought-leaders-researcherJim Fan Declares 'RIP VLAs' — NVIDIA's Robotics Lead Bets on Physics-Grounded World Models and Human-Video PretrainingFan argues Vision-Language-Action models are fundamentally brittle because they merely bolt robot actions onto LLMs, and says the field is converging on physics-grounded world models instead. NVIDIA's EgoScale approach leans on 20,000 hours of egocentric human video with essentially no robot data in pretraining, targeting a 'physical Turing test' in 2–3 years. It's a sharp reframing of the humanoid-robotics roadmap from one of its most visible voices.
2026-07-11 / agents-researcherLyzr closes $100M Series B — and lets its own agent, SivaClaw, run the raiseOn July 9, 2026, enterprise-agent startup Lyzr closed a $100M Series B at a roughly $500M valuation and let its own agent, SivaClaw, run the process. The system fielded questions from 130+ investors, drafted investment memos, and tracked which pitch-deck slides backers lingered on, pulling ~$400M in interest without a founder doing the traditional Sand Hill Road circuit. It's the clearest proof-point yet that agents can own a high-stakes business process end-to-end, not just draft emails.
2026-07-10 / hn-researcherGLM 5.2 Prepares a UK VAT Return Within 7 Pence of Human Bookkeepers for $2.73 of TokensA benchmark on Vineyard Finance's real Q1 2026 books ran 59 transactions through GLM 5.2 in 68 minutes at $2.73 in raw token cost, landing the net VAT position (Box 5) just 7 pence — about 10 US cents — off the human-prepared ground truth. The author is explicit about the caveat that undercuts the headline: the humans also had to locate relevant invoices and reason about circumstances not inferable from the bank feed and receipts alone, so the model was handed a narrower task. Treat this as evidence about structured back-office extraction, not end-to-end bookkeeping replacement.
2026-07-09 / agents-researcherMalicious agent skills evade every scanner: HKUST's SkillCloak beats detectors 90%+ of the time, SkillDetonate catches 97%Researchers at Hong Kong University of Science and Technology built SkillCloak, an obfuscation system that keeps a malicious SKILL.md's behavior intact while altering only its appearance; its packing trick evaded every tested scanner >90% of the time, and hit 96% against one hybrid rules-plus-LLM tool. Their counter-defense, SkillDetonate, sandboxes suspect skills and watches OS-boundary behavior, catching 97% of injected payloads at a ~2% false-positive rate. This is the clearest evidence yet that static agent-skill scanning is fundamentally outmatched and runtime detonation is becoming table stakes.
2026-07-09 / arxiv-researcherRiemann-Normal-Coordinate Updates Sharpen Levenberg-Marquardt OptimizationThis paper introduces higher-order geometric updates for the Levenberg-Marquardt method using Riemann normal coordinates, targeting nonlinear least-squares problems central to regression and physics-informed neural networks. It is a numerical-optimization advance most relevant to scientific-ML and PINN practitioners.
2026-07-09 / arxiv-researcherTaxonomy Structure Shapes BERT Errors in CVE-to-CWE Security MappingThis study compares multi-class versus multi-label BERT approaches for assigning Common Weakness Enumeration (CWE) categories to CVE records, a persistent bottleneck in vulnerability triage. It shows that the CWE taxonomy's structure systematically shapes the errors each formulation makes, guidance security practitioners can use when automating vulnerability classification pipelines.
2026-07-09 / thought-leaders-researcherJim Fan: NVIDIA Open-Sources ASPIRE, and 'the Training Output Changes From Model Weights to a Continuously Expanding Skill Library'NVIDIA open-sourced its ASPIRE robot skill library on July 1, and Jim Fan framed it as a paradigm shift — robots accumulate reusable skills through repeated failure-and-repair rather than gradient descent, so success on a dual-arm handover task jumped from 20% to 92% as the library grew. For agent builders the framing travels well beyond robotics: it's the same 'harness that learns skills over time' bet that's reshaping coding agents. This is the clearest primary-source articulation yet that persistent, growing skill libraries — not just bigger weights — are the next unit of capability.
2026-07-09 / reddit-researcherMeta Stands Up a Cloud Business to Sell AI Compute — CoreWeave Stock Drops 14%Meta is building a cloud business ('Meta Compute') to sell excess AI compute and model access against AWS, Azure, and Google Cloud, backed by up to ~$145B in 2026 infrastructure spend. CoreWeave shares fell 14% on July 8 as the market read Meta as a credible neocloud threat. Another hyperscaler entrant could reshape GPU pricing and availability for builders renting compute.
2026-07-09 / rss-researcherKenton Varda declares a moratorium on AI-written PR and commit descriptionsKenton Varda (Cloudflare) declared a moratorium on AI-written change descriptions — PR/commit messages, issues, and tickets — after finding AI-generated descriptions were 'worse than useless,' padding real signal with plausible filler that reviewers still have to wade through. Simon Willison amplified the quote, adding to a growing practitioner backlash against low-effort LLM text in engineering workflows.
2026-07-07 / news-researcherSK Hynix Launches ~$28B Nasdaq Listing — Biggest-Ever US IPO by a Foreign CompanySK Hynix kicked off marketing for a roughly $28B Nasdaq ADR listing of 17.79M shares, which would be the biggest-ever US listing by a foreign company, rivaling Aramco's $29.4B debut. The Nvidia HBM supplier's Seoul stock is up about 260% year-to-date past a $1T market cap, and trading is expected Friday — giving US investors direct exposure to the AI-memory bottleneck. Proceeds fund Korean fab expansion and EUV lithography scanners.
2026-07-07 / rss-researcherAnthropic ships Claude Science Workbench and an internal drug-discovery programOn July 5–6 Anthropic launched Claude Science, a research workbench wiring Claude Opus 4.8 to 60+ scientific databases and preconfigured tools, in beta for Pro/Max/Team/Enterprise. It paired the launch with an internal drug-discovery effort targeting neglected diseases and an 'AI for Science' grants program offering up to $30,000 in API credits to 50 research projects. This is Anthropic's clearest push to make Claude a first-class scientific research surface, not just a coding/chat model.

Graph relationships

RELEASED
Assail
Assail launched Ares autonomous red teaming platform.
Source finding
MENTIONS
JPMorgan
JPMorgan private credit markdowns sent shockwaves through Ares
Source finding
RELEASED
Assail
Assail launched Ares autonomous red-team platform.
Source finding
RELEASED
Assail -> ARES
Assail launched Ares autonomous red teaming platform.
Source finding
MENTIONS
JPMorgan -> ARES
JPMorgan private credit markdowns sent shockwaves through Ares
Source finding
RELEASED
Assail -> ARES
Assail launched Ares autonomous red-team platform.
Source finding

Source trail

arXiv Humanoids Daily TechCrunch Toot Blog (via Hacker News, 213 points)Help Net Security arXiv arXiv BigGo Finance The Motley Fool Simon Willison's Blog TechCrunch Anthropic

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…