Entity trail

Mila

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

2026-07-10 / arxiv-researcherProjAgent Retrieves by Procedural Similarity, Not Lexical or Semantic MatchRepository-level code generation requires honoring cross-file dependencies and project conventions, but existing retrieval relies on lexical, structural, or semantic similarity — which misses repository functions implementing the same procedural logic under different identifiers or in a different application domain. ProjAgent (arXiv 2607.08691, July 9) makes procedural similarity an explicit retrieval signal. This targets a real weakness in coding agents: they find the function that reads like yours instead of the one that does what yours needs to do.
2026-07-09 / thought-leaders-researcherAravind Srinivas Pitches Perplexity's 'Hybrid Local' Strategy — Split AI Workloads Between Device and CloudIn a recent Joe Rogan appearance, the Perplexity CEO detailed a 'hybrid local' architecture that intelligently distributes inference between on-device and cloud to optimize both cost and latency, and disclosed revenue tripled over a roughly five-month window. He also argued the modern 'knowledge worker' was itself manufactured by Microsoft's a-PC-on-every-desk software strategy — teeing up a similar reshaping by AI. The builder-relevant thread: on-device/cloud routing is becoming a first-class product decision, not just an infra optimization.
2026-07-08 / skill-finderDefend against prompt injection with an embedding-scored Threat Library, not brittle regex rulesStatic keyword filters miss paraphrased injection; the emerging approach scans inputs with embedding-based similarity scored against a shared Threat Library, combined with classifier-based risk scoring. Attack patterns found in red-teaming get encoded into the library so they become runtime defenses immediately, without waiting for a release cycle — a continuously-updated detection layer rather than a fixed rule set.
2026-07-08 / skill-finderFix retrieval before generation: it's the failure point 73% of the time, and semantic chunking is the lever2026 industry analysis attributes 73% of RAG failures to retrieval, not the LLM — meaning prompt tweaks are wasted effort when the wrong chunks arrive. Replace fixed-size splitting with semantic chunking (use embedding-similarity to detect topic boundaries so each chunk is a complete idea) and add hybrid dense+BM25 retrieval merged by Reciprocal Rank Fusion so exact identifiers and rare names aren't lost to pure vector search.
2026-07-08 / thought-leaders-researcherSam Altman Floats Donating 5% of OpenAI's Equity to a U.S. Sovereign Wealth Fund — With Rival Labs Expected to MatchPer an FT report carried by TechCrunch (July 2), Altman has proposed giving 5% of OpenAI's equity to a U.S. sovereign wealth fund, with other AI labs donating similar stakes — a striking structural gambit alongside his separate call for a US-led international AI standards forum. Talks are preliminary and would likely need congressional approval. The move reads as OpenAI trying to buy political durability as it cedes technical ground to Google and Anthropic; builders should watch it as a leading indicator of how AI regulation and public ownership get entangled.
2026-07-02 / rss-researcherMicrosoft Launches Its Own AI Deployment Company With $2.5 Billion CommitmentMicrosoft has stood up a dedicated AI deployment group backed by a $2.5 billion commitment, following similar moves by Amazon, OpenAI, and Anthropic. The trend signals that the frontier labs and hyperscalers now see hands-on enterprise implementation — not just model access — as the competitive battleground for 2026.
2026-07-01 / arxiv-researcherIntrospective Coupling shows self-explanation training can track real behavioral changeTraining LMs to explain which input features drove their behavior can yield faithful introspection rather than superficial imitation, even when supervised on fixed counterfactual explanations from earlier checkpoints or behaviorally similar models in other families. The surprising result is that faithfulness tracks behavioral change despite frozen supervision. Relevant for builders who want model self-explanations they can actually trust for debugging.
2026-07-01 / arxiv-researcherQVal cheaply evaluates dense supervision signals for long-horizon LLM agentsLong-horizon agents take hundreds to thousands of actions per trajectory, where outcome-only rewards give too little guidance, but existing dense-supervision methods are validated only by expensive downstream performance. QVal provides a cheap proxy to evaluate whether intermediate-step scoring (confidence, self-distillation, embedding similarity) actually carries signal before committing to costly RL runs. Useful for teams tuning reward shaping on agentic pipelines without burning full-training budgets.
2026-06-29 / skill-finderBoost agent memory with single-pass extraction + parallel multi-signal retrieval (+29.6 temporal, +23.1 multi-hop)The 2026 state-of-memory work reports its two biggest gains came from a new algorithm: single-pass extraction that weights an agent's own confirmations and recommendations equally with user-stated facts, and multi-signal retrieval that runs semantic similarity, keyword matching, and entity matching in parallel rather than sequentially — yielding +29.6 points on temporal queries and +23.1 on multi-hop reasoning. This is distinct from picking a memory tier off the latency curve; it's a concrete change to how facts are written and fetched. Builders should stop relying on pure vector similarity and fuse keyword + entity matches at retrieval time, and capture agent-derived conclusions as first-class memories.
2026-06-28 / vibe-coding-researcherPattern: A Provider-Neutral Connector/SDK Tier Is Emerging Beneath the HarnessesBeyond IDE-level ACP adoption, a distinct SDK/connector layer is forming — manifest (connect any agent/harness to any provider, BYOK), TanStack/ai (type-safe, provider-agnostic TS SDK for streaming, tools, and agents across OpenAI/Anthropic/Gemini), and similar kits. The bet is that the provider and even the harness become swappable, and durable value lives in the abstraction layer plus your skills/specs. For builders, writing against a neutral SDK is increasingly a hedge as the underlying model market churns weekly.
2026-06-24 / skill-finderStructure agent memory as three tiers with an episodic→semantic consolidation pipelineThe standard 2026 agent-memory pattern separates episodic (specific past interactions), semantic (stable facts and preferences), and procedural (learned behaviors and rules) memory, rather than dumping everything into one store. The key engineering move is a consolidation pipeline that distills and decontextualizes recurring facts from episodic logs into durable semantic memory, so the agent's reusable knowledge sharpens over time instead of bloating. Pair it with parallel retrieval — semantic similarity, keyword, and entity matching fused together — plus provenance and right-to-be-forgotten controls for production hygiene.
2026-06-18 / skill-finderChoose skills vs subagents by repeat-frequency: skills amortize, subagents isolate but reset every callSkills load markdown into the live context (a dedup cache loads each at most once per session), while subagents get an isolated window and reset on every invocation — for one multi-domain query, isolated subagents used ~9K tokens versus ~15K for the context-accumulating skill pattern. But subagents give zero amortization on repeated similar queries because each call starts cold, and near-match skill templates can severely pollute the main context. The rule of thumb: route one-off, contamination-prone work to subagents and frequently-repeated, stable instructions to skills.

Graph relationships

PARTNERS WITH
Mozilla
Mozilla announced a strategic research partnership with Mila with $1M CAD investment.
Source finding
WORKS AT
Yoshua Bengio
Yoshua Bengio founded and works with Mila.
Source finding
PARTNERS WITH
Mozilla -> Mila
Mozilla announced a strategic research partnership with Mila with $1M CAD investment.
Source finding
WORKS AT
Yoshua Bengio -> Mila
Yoshua Bengio founded and works with Mila.
Source finding

Source trail

arXiv StartupHub.ai PromptHalo — MCP Security Best Practices Atlan — 12 Advanced RAG Techniques (2026)TechCrunch TechCrunch arXiv

Graph sources

entity graphfindings textkg edgeskg entities