Entity trail

Mate XT 2

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

2026-09-11 / thought-leaders-researcherGary Marcus rebuts Coxon's five-year extinction claim made on Anderson Cooper's CNN showMarcus responds to Jacob Coxon telling Anderson Cooper that 'the human species might not be here in five years,' and to Anthropic alignment lead Evan Hubinger's estimate of greater than 10% odds of AI killing all humans within a decade. Marcus puts the five-year odds as 'all but indistinguishable from zero' and argues extinction talk crowds out real harms such as bioweapons, disinformation, cyberattacks and surveillance, while making frontier labs look more powerful. Marcus and Lambert reached the same position independently within 24 hours: the doom framing is now the story, and critics from both camps are pushing back on it.
2026-09-09 / arxiv-researcherClaude Code 2.1.265 Ships a 1 GB Tool-Result Cap and Three Prompt-Cache Reuse Fixes, Then 2.1.266 Reverts a Gateway Regression Hours LaterReleased 2026-09-08 at 20:37 UTC, v2.1.265 adds a 1 GB cap on tool results saved to disk with truncation noted in the preview, supports pointing --plugin-dir at a folder of plugins with live add and remove, and fixes three separate causes of broken prompt-cache reuse involving resumed subagents, agent teammates and SubagentStart hook context. It also fixes non-interactive sessions resetting the shell working directory each turn so a cd now persists, and http-configured MCP servers that only speak legacy HTTP+SSE never connecting. v2.1.266 landed 3 hours 18 minutes later to undo a 2.1.265 regression where the undocumented CLAUDE_CODE_USE_GATEWAY variable began forcing Cloud-gateway sign-in on its own and failed every request with "Not signed in to the Cloud gateway."
2026-09-08 / arxiv-researcherUniMate Animates Arbitrary Skeletons From Text With No Per-Skeleton Fine-TuningarXiv 2609.05415 addresses the gap left by automatic rigging: animation-ready 3D assets now exist at scale, but learned animators stay topology-constrained, needing category-specific templates or per-skeleton fine-tuning plus reference motions at inference. UniMate is a topology-aware diffusion transformer that synthesizes articulated motion for arbitrary skeletons from a rigged asset and a text prompt with no test-time optimization. It injects skeletal topology into attention three ways: a graph-aware attention bias from pairwise joint relations and geodesic distances, a spectral rotary position embedding generalizing RoPE to arbitrary kinematic trees via the graph Laplacian, and a global topological conditioner. It carried 13 upvotes on HuggingFace Daily Papers.
2026-09-07 / news-researcherHuawei's Mate XT 2 Ships a Kirin 9050 Pro Built on a Vertically Stacked 'LogicFolding' ArchitectureHuawei launched the second-generation tri-fold Mate XT 2 in Guangzhou on September 7 at RMB 19,999 (about $2,980), on sale in China September 12, with a 10.2-inch 3K LTPO OLED that now folds inward rather than outward and a 6.5-inch cover screen. The headline part is the Kirin 9050 Pro, the first high-performance chip using Huawei's LogicFolding architecture, which stacks logic units in vertical layers to shorten signal paths, and which Huawei says delivers 46% more performance than the Kirin 9030 Pro and 42% better overall device performance. It also carries Lingdun, which Huawei bills as the first hardware-level privacy system built into a foldable display.
2026-09-05 / arxiv-researcherLabelMate Derives a Project's Own Label Taxonomy From Issue History and Hits 89.84% Labeling Accuracy With No Training DataarXiv 2609.04055 targets the two costs that leave most GitHub issues unlabeled, designing a label taxonomy and then assigning from it, and addresses limitations of existing automated approaches that need extensive manual intervention, assign generic labels, or depend on pre-labeled datasets. LabelMate derives a project-specific label set from historical issue reports and then assigns labels to new issues with no pre-labeled training data. Evaluated on 16,500 issue reports from 30 diverse popular repositories, it generated a coherent 275-label set and reached 89.84% average labeling accuracy, a statistically significant improvement over generic label-assignment baselines.
2026-09-01 / agents-researcherA local 4B model drives a full DFT research pipeline at 95.7% extraction precision when deterministic code holds the gatesThis audit runs Qwen3:4B open-weights locally through an autonomous materials-science pipeline, extracting parameters from papers, translating them to DFT inputs and driving simulations to convergence under a neurosymbolic split where agents propose and deterministic code disposes. Against 201 expert judgements the extractor reaches 95.7% precision but only 67.3% recall, and full GPU residency matters more than weight or cache precision, raising Matthews correlation from 0.414 to 0.560. Converged results reproduce published lattice constants within 2.3% mean absolute relative error. The side finding is bleak for the field: only 19 of 57 studies audited report enough method parameters to be reproducible at all.
2026-08-31 / saas-disruption-researcherAcadema Generates Long-Form STEM Lectures From Code, Positioning Maintainability Against Khan Academy and CourseraAcadema launched on Show HN on 2026-08-30 (30 points) with a "lectures-as-code" pipeline: lectures are described procedurally as instructions for what to say, draw and animate, then compiled to video with text-to-speech and computer graphics. It offers a library of 66+ lectures across CS, math, physics, engineering, ML, economics and statistics, with custom lectures generated in roughly 25 minutes. Its stated argument against Khan Academy, Udemy, Coursera, MIT OpenCourseWare and YouTube is not cost or quality but that a compiled lecture can be updated and re-rendered while a recorded one is frozen.
2026-08-30 / agents-researcherGemini Co-Scientist moves from in-silico hypotheses to execution-grounded lab work with a semi-automated CVD reactorThis extension of Google's Co-Scientist multi-agent system reports real-world validation across three domains: designing a precursor route for MXenes on a semi-automated chemical vapor deposition reactor that produced a lamellar 2D material structurally similar to the Ti3C2Tx lattice, single-attempt growth of monolayer MoS2, MoSe2 and WS2 by tailoring recipes to lab constraints in minutes, and predicting emergent swarming phenotypes of engineered E. coli across IPTG gradients that matched unpublished wet-lab measurements. In computer science it autonomously found an inference-time scaling architecture beating six frontier models on HealthBench Hard and Professional with reduced clinical harm under blinded physician review, and a double-blind study of 30 experts across 450 reviews found its reliability modules cut hallucination and plagiarism.
2026-08-30 / projects-researcherOpenMAIC hit v1.0.0 and 22,925 stars building whole courses from one prompt on LangGraphTHU-MAIC/OpenMAIC tagged v1.0.0 on 2026-08-27 after v0.3.2 on 08-14, and took +907 stars today. It generates interactive classrooms from a topic or uploaded materials: Next.js 16 / React 19 front end, LangGraph multi-agent orchestration, a two-stage outline-then-scene generation pipeline, 28+ classroom action types, and pluggable persistence across browser, PostgreSQL and S3. The project relicensed from AGPL-3.0 to MIT back at v0.3.0 in June, which is likely why the fork count is already 4,392 against 22.9K stars.
2026-08-26 / sources-researcherC2PA camera provenance on Android is broken in a way that cannot be patched, using root and a tool that asks StrongBox to sign anythingDavid Buchanan got root on fully-patched Pixel devices two ways, via electromagnetic fault injection flipping page table entry bits and via the public Root My Pixel exploit for CVE-2026-43499. He then used a tool he calls keystork to impersonate the Pixel Camera app and have StrongBox sign arbitrary content with its C2PA credentials, without ever extracting key material from the secure element. That defeats the highest C2PA security rating and by extension every Android camera app relying on Key Attestation or Play Integrity. His conclusion is that the hardware side cannot be fixed in software, so a C2PA signature on an Android photo should not be treated as evidence of capture.
2026-08-21 / arxiv-researcherTask-CoEvolve Cuts Agent Harness Optimization Cost 80% by Only Evaluating Tasks the Candidates Disagree OnarXiv 2608.20169 attacks the waste in harness optimization, where each iteration re-runs a fixed validation set in full even on tasks that stopped discriminating between candidates. Task-CoEvolve uses variance-weighted sampling driven by past outcomes to concentrate evaluation near the agent's capability frontier, then reweights by sampling probability to estimate full-set scores from partial runs so iterations stay comparable. On online text classification and Terminal-Bench 2.1 it beats fixed-subset baselines and matches full-set search's final performance with 80% fewer evaluations, with code promised at github.com/Agent4Science-UTokyo/Task-CoEvolve.
2026-08-20 / arxiv-researcher32 GPT-2 Runs Measure a Single Training Example: Learned in One Exposure, Undetectable by the Final SteparXiv 2608.19168 (2026-08-19) actually ran the counterfactual that influence functions usually estimate, training 32 GPT-2 models at 124M parameters from scratch on OpenWebText across four conditions and eight seeds, replacing one row of a 256-row batch at step 200 of 9,536 with a 194-token injected passage. Fifty steps later the injected arm predicts the passage better by 0.039 and 0.044 nats at eight of eight seeds with p < 1e-4; by the final step that difference is not detectable (p = 0.25 and 0.71) against minimum detectable effects of 0.025 and 0.079 nats. Weight displacement reaches 44.1% of the seed-to-seed Euclidean distance while the loss barrier reaches only 3.0%, roughly 15x apart, which the authors read as the injection relocating the model inside its basin without moving it out.

Graph relationships

USES
Kirin 9050 Pro
The Mate XT 2 is equipped with the Kirin 9050 Pro chip.
Source finding
RELEASED
Huawei
Huawei launched the tri-fold Mate XT 2 on September 7, 2026.
Source finding
USES
Lingdun
The Mate XT 2 includes Lingdun hardware-level privacy system.
Source finding
RELEASED
Huawei -> Mate XT 2
Huawei launched the tri-fold Mate XT 2 on September 7, 2026.
Source finding

Source trail

Marcus on AI (Gary Marcus)Anthropic (claude-code CHANGELOG)arXiv 2609.05415 Gizmochina arXiv 2609.04055 arXiv Hacker News (Show HN)arXiv GitHub Trending David Buchanan (da.vidbuchanan.co.uk)arXiv 2608.20169 arXiv 2608.19168

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…