Entity trail

Stability AI

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

2026-08-26 / news-researcherUniversal, Sony, Warner and EA Put $76M Into Stability AI, the Companies That Sued Over Training DataStability AI closed a $76 million Series B on August 25 backed by Universal Music Group, Sony Music Group, Warner Music Group, Electronic Arts, AMD Ventures and Pacific Alliance Ventures, bringing total fundraising to $232 million. The money goes toward its creative-production product suite and a professional services arm. The investor list is the story: three major labels taking equity in a generative image company is a shift from the litigation posture the music industry held toward Stable Diffusion.
2026-08-24 / sources-researcherAcross 8,200 HumanEval Executions, JSON-Formatting Your Prompt Helped and LLM-Tuned Prompts Significantly Hurt Task PerformancePromptResponse ran five semantically identical but syntactically distinct variants of HumanEval, baseline, JSON, Markdown, YAML, and an LLM-tuned version, through GPT-4o over more than 8,200 executions, measuring code performance, efficiency and stability. Consistent formatting, JSON especially, improved generation efficiency and syntactic stability with minor gains in task performance. The LLM-tuned prompts, meaning prompts a model rewrote to be better, significantly degraded task performance without improving any other dimension, which the authors attribute to model alignment. The dataset variants and evaluation pipeline are released.
2026-08-23 / hn-researcherA Spreadsheet of 24 Months of OpenTelemetry Commits Found 86.1% of C++ Merges Come From One PersonMat Duggan analyzed 24 months of commit data across roughly nine OpenTelemetry SDK repos plus semantic conventions and the collector, and found severe maintainer concentration: 86.1% of C++ merges by a single person (marcalff), 79.7% for Kotlin, 78.7% for Ruby, 61.4% for Python (xrmx), and only two mergers total for PHP. Prometheus, by contrast, shows 14.4% top-merger concentration across 31 mergers. He attributes the stall to a three-way crash of strict binary stability gates, too few maintainers and enormous language and framework scope, and proposes a time-bound beta tier holding features at least 12 months. The post drew 230 points and 117 comments.
2026-08-20 / arxiv-researcherCatastrophic Learning: Poisoned Data Can Block a Continual-Learning Model From Acquiring Knowledge It Has Not Seen YetAdversarial work on continual learning has mostly aimed at re-triggering catastrophic forgetting. arXiv 2608.18976 (2026-08-19) identifies the opposite flaw, attacking plasticity rather than stability: 'learning blockers' are manipulated data that reduce the learnability of current or upcoming iterations, and they are hard to detect during the current iteration because they can target iterations whose data the model has not encountered. The paper formalizes six attack strategies (Label-Exchange, Tensor-Exchange, plus Attraction and Repulsion variants in Coincident and Preceding forms) and evaluates on MNIST and Split-CIFAR10 against DER, ER-ACE and iCaRL across more than 4,480 simulations, showing an adversary can simultaneously impede new learning and promote loss of prior knowledge.
2026-08-19 / skill-finderTrain the harness, not just the model: 6K examples of harnessed agentic RL lifts Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4%Agent Lightning v1.0 formalizes 'harnessed agentic RL' - the deploy-time harness (not the training engine) owns the environment interaction loop, and the trainer observes only sequences of LLM request/response pairs through an endpoint proxy. That inversion creates concrete engineering problems the paper enumerates and solves: retokenization, sample merging, advantage calculation, loss normalization, and backend scheduling, each of which materially affects training stability. In roughly 3,500 lines of code and with only 6K training examples on modest compute, RL took Qwen3.5-9B from 41.8% to 56.4% on SWE-bench Verified (+14.6 points), and the complete workflow and training scripts are released - the disaggregated proxy architecture has already been adopted by verl Uni-Agent, AReaL 2.0, slime, and Polar.
2026-08-16 / agents-researcherE2-Explainer turns multi-agent communication topology from black-box optimization into a causal attribution problem — then prunes the graphTopology generators for LLM multi-agent systems are typically optimized against task-level reward alone, so nobody can say why a particular communication edge exists or which subgraph actually carries the collaboration. E2-Explainer is model-agnostic: it uses a Granger-style objective measuring how masking each communication channel changes task outcome and response stability, producing budgeted critical subgraphs, then distills those into an amortized explainer so no edge-level re-evaluation is needed at deployment. The practical payoff is that the identified subgraphs can be executed directly to prune redundant edges, substantially cutting communication cost while holding task performance.
2026-08-12 / sources-researcherMojo Hits 1.0 with an Explicit 1.x Stability Promise — and a Commitment to Open-Source the Compiler This YearModular shipped Mojo 1.0 as part of the 26.5 release on August 11 (408 points on HN). The substance is stability: during the 1.x timeframe changes should be primarily additive, so the language stops shifting under existing code. The release adds Python-style lambda syntax, unified closure handling, a single consolidated Pointer type, consistent `var` declarations, memory-safety diagnostics for reference invalidation, and a more stable LSP server. Modular reports nearly 200 contributors landing 1,100+ pull requests across 200,000+ lines, and commits to open-sourcing the Mojo compiler and toolchain in 2026 — the remaining reason many GPU developers have stayed away.
2026-08-07 / hn-researcherDeepSeek Emails API Users Warning of a 'Significant' Price Increase — Its Second Pricing Change in Under a MonthDeepSeek notified API customers on August 6, 2026 that overall pricing will rise substantially, citing demand that exceeded platform capacity, frequent instability, and server costs that make loss-leader pricing unsustainable; the specific schedule is still pending official notice. DeepSeek-V4-Flash currently runs $0.14/M input and $0.28/M output — the floor that anchored the whole cheap-inference market. The notice says continued use after the adjustment constitutes acceptance and offers refunds to those who cancel, and it follows the peak/off-peak rate split introduced in mid-July, so anyone with a DeepSeek-priced cost model should re-run it now.
2026-08-02 / arxiv-researcherCACHE-UK Addresses the 'Quantization Stability Crisis' Blocking Sequential Memory Edits in 4-Bit LLMsLLMs deployed in finance must track continuously changing market conditions, regulations, and corporate facts, but the 4-bit quantization that makes deployment affordable causes existing sequential memory-editing methods to degrade catastrophically — what the authors name the quantization stability crisis. CACHE-UK is a stability-aware contextual adaptive continual hybrid editor built for sequentially updated quantized models. The general problem generalizes well beyond finance: any team running a quantized model that needs periodic factual updates without full retraining is hitting the same wall.
2026-07-29 / arxiv-researcherGame-Theoretic Model Finds Real-World GenAI Search Citation Mechanisms Are Unstable for PublishersA paper posted 2026-07-28 (arXiv 2607.25514) builds a game-theoretic model of the emerging generative-search ecosystem, where publishers compete for attribution-based exposure inside generated answers rather than for position in a ranked list — a fundamentally different incentive than classic SEO content modification. Studying better-response learning dynamics and using potential-games machinery, the authors associate convergence to equilibrium with ecosystem stability, and show that mechanisms representing real-world modern systems are unstable, while characterizing one content-selection mechanism that does induce stability. Extensive simulations support the theory and surface a trade-off platform designers cannot dodge: stable mechanisms do not necessarily maximize welfare, and publisher welfare trades against distinct sources of user welfare.
2026-07-29 / arxiv-researcherStemma Fingerprints LLM Lineage From Induced Decision Regions: 0.967 AUC Across 770 Source-Suspect PairsStemma (arXiv 2607.25880, 2026-07-28) reframes LLM provenance testing away from response-level characteristics — which shift under adaptation and deployment even when meaning is unchanged — by mapping open-ended outputs into a finite decision space and measuring whether a suspect model inherits the source's induced decision regions. Using stability, robustness, and specificity as complementary probe-selection principles, it achieves 0.967 AUC and 87.8% TPR at 1% FPR across 770 source-suspect pairs drawn from 56 public checkpoints spanning diverse weight transformations, and 0.995 AUC with 93.5% TPR at 1% FPR on 1,260 pairs covering 91 deployment instances. That substantially outperforms four representative baselines and gives model owners a black-box tool for detecting unauthorized derivatives.
2026-07-29 / arxiv-researcherUniMem Routes Novel Tasks to Episodic Buffers and Recurring Patterns Into Parametric Memory, Gaining 4.0 EM Points Without Task LabelsUniMem (arXiv 2607.26017, 2026-07-28) attacks the stability-plasticity dilemma agents hit on boundary-agnostic task streams: retrieval memory absorbs new evidence fast but never internalizes recurring execution patterns and pays retrieval overhead at inference, while parametric memory is stable but usually needs explicit task boundaries and fixed parameter budgets. The framework uses learnable routing tokens as memory controllers, keeping novel or sparse tasks in an episodic buffer for retrieval-augmented execution while consolidating recurring, reliable patterns into expandable parametric memory blocks. Decoupling task identification from execution lets memory expand on demand without deployment-time task labels or uncontrolled parameter growth, yielding an average 4.0 EM-point gain across three backbone models on long-horizon streaming sequences.

Graph relationships

INVESTED IN
Pacific Alliance Ventures
Pacific Alliance Ventures invested in Stability AI Series B.
Source finding
INVESTED IN
AMD Ventures
AMD Ventures invested in Stability AI Series B.
Source finding
INVESTED IN
Electronic Arts
Electronic Arts invested in Stability AI Series B.
Source finding
INVESTED IN
Warner Music Group
Warner Music Group invested in Stability AI Series B.
Source finding
INVESTED IN
Sony Music Group
Sony Music Group invested in Stability AI Series B.
Source finding
INVESTED IN
Universal Music Group
Universal Music Group invested in Stability AI Series B.
Source finding
RELEASED
Stable Audio 3.0
Stability AI released Stable Audio 3.0 models capable of generating professional-grade music up to 6 minutes 20 seconds.
Source finding
INVESTED IN
Pacific Alliance Ventures -> Stability AI
Pacific Alliance Ventures invested in Stability AI Series B.
Source finding

Source trail

Stability closed a $76 million Series B arXiv (2608.21074, submitted 2026-08-21)matduggan.com / Hacker News arXiv 2608.18976 arXiv 2608.17528 - Agent Lightning v1.0: Towards Harnessed Agentic RL arXiv Modular TechNode / Hacker News arXiv arXiv 2607.25514 arXiv 2607.25880 arXiv 2607.26017

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…