Entity trail

Mole

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

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-18 / arxiv-researcherLarge Discovery Models Pair a Generative Proposer With a Bayesian Surrogate, Beating LLM-Only Reflection Across Three DomainsLLM likelihoods and self-assessments are unreliable proxies for expensive real objectives, especially on novel candidates outside the observed distribution. The Large Discovery Model couples a generative proposer with a Bayesian non-parametric reward surrogate that predicts performance and quantifies uncertainty, feeding an uncertainty-aware value back into generation, refinement and selection while continually updating both the discovery memory and the surrogate as each experimental observation arrives. Against LLM-only reflection and traditional statistical search it delivers 2.4x greater reduction in validation BPB on neural-network training, an 18.2% relative decrease in antibody binding energy, and over 60% relative gains on molecular multi-objective optimization.
2026-08-17 / projects-researcherMole, a Terminal-Native Deep Research Agent, Reaches 100 Points on Show HNlajosdeme/Mole is a deep research agent that runs in the terminal rather than as a hosted web product, sitting at 100 points on the Show HN board. It joins a small cluster of research agents that assume the developer already lives in a shell and does not want another browser tab. Also on the board: a public AI whose memory is shared across all users (74 points) and PyScrappy (21), which pairs self-healing scraping selectors with an MCP server.
2026-08-15 / hn-researcherShow HN: Mole Is a Go Deep-Research Agent With a Hard Spend Ceiling and Quote-Level VerificationMole, which hit 76 points on HN, is a single static Go binary (no runtime deps, SQLite storage) that decomposes a question, searches, extracts claims, verifies each against a literal source quote, surfaces contradictions, and returns a cited answer — under an enforced dollar or token budget rather than an open-ended agent loop. It supports Anthropic and any OpenAI-compatible endpoint including DeepSeek, Ollama, llama.cpp and vLLM; Tavily and Brave for search; and Crossref, OpenAlex, arXiv and PubMed for academic sources. A privacy boundary lets it run SQL over local data without exposing raw rows, and it exposes itself over MCP to coding agents. Apache-2.0, 151 stars, 173 commits.
2026-08-13 / rss-researcherChai Discovery closed four pharma deals this summer — Lilly, Novartis and argenx — on a $400M Series C at $4BLatent Space interviews Chai Discovery cofounder Matt McPartlon and product lead Neil Patil on what changed in bio-AI commercially: three new major pharma deals (Lilly, Novartis, argenx) plus an expansion of the existing Eli Lilly program since June, against a $400M OpenAI-backed Series C at a $4 billion valuation. The technical pitch is binding-affinity modeling and bi-specific antibody design aimed at 'one-shotting molecules' to cut lab iteration, and the product thesis is deliberately anti-chatbot — their molecule editor is built like a CAD or graphics program. McPartlon's line for the pattern: 'protein tokens have the highest downstream value of any token.'
2026-08-10 / rss-researcherEric Schmidt argues AI-for-science needs reasoning agents, not more AlphaFolds — because most fields have no Protein Data BankEric Schmidt and Suhas Mahesh (AI for Science lead at the Schmidt Sciences AI Center) make the case that the AlphaFold template does not generalize: the Protein Data Bank took 53 years, roughly $21 billion in experimental work and underpinned 25+ Nobel Prizes to assemble ~170,000 validated structures, and almost no other discipline has an equivalent. Their alternative is iterative reasoning agents that work through uncertainty the way human researchers do rather than requiring a massive pre-existing dataset, citing Google's AI Co-Scientist matching conclusions Imperial College London researchers spent a decade validating in wet-lab work on antibiotic resistance. The builder-relevant framing is throughput: an agent that can 'read a thousand papers in an hour, design 500 molecules' and learn from failed tests.
2026-08-04 / arxiv-researcherVI-MoLE Argues Uncertainty Is the Wrong Trigger for Activating More LoRA ExpertsDynamic routers for mixtures of LoRA adapters typically activate more experts when the router or prediction is uncertain, silently equating uncertainty with useful extra computation — but an uncertain example may stay ambiguous even after every expert agrees. VI-MoLE reframes routing as certified value-of-information allocation: it learns the counterfactual risk remaining after each expert prefix, converts predictions into simultaneous upper-risk certificates on held-out calibration data, and spends a global adapter budget on the token-layer action with the largest certified marginal risk reduction per unit cost, with a terminal certificate deciding answer-versus-abstain. The paper proves simultaneous certificate validity, optimal greedy allocation under diminishing certified gains, and allocation regret under value-estimation error.
2026-07-31 / sources-researcherJim Nielsen Names the AI Aesthetic: Sparkle Emoji, Shimmer-as-Loading, Beige-and-Orange Serif, and Suspiciously Tiny Mac IconsJim Nielsen's 2026-07-29 essay catalogues the visual idioms AI tools have minted and asks which survive as durable interaction patterns versus which are era-specific fads — 'Every zeitgeist comes with new design idioms unique to its challenges.' His inventory: ✨ as the universal AI signifier, streaming text with little use outside chat, shimmering text drifting from 'model is thinking' to any async task, beige/cream palettes with orange accents and serif type, and 'whack-a-mole' controls that move out from under the cursor. The sharpest observation is that Claude, Codex and Cursor desktop apps all ship icons noticeably smaller than native macOS apps — a shared tell that AI tools are being designed by people who are not designing for macOS.
2026-07-24 / arxiv-researcherMolecular Ensemble Modeling: Single-Conformation Property Prediction Is Leaving Signal on the Table for Cyclic PeptidesAaron Feller, Kris Deibler and Maxim Secor investigate graph learning over conformational ensembles of cyclic peptides, challenging the standard practice of predicting molecular properties from a single representative conformation. Cyclic peptides are a hard case because they genuinely populate multiple conformations in solution. The finding matters for drug-discovery ML teams deciding whether ensemble generation is worth the compute over single-conformer pipelines.
2026-07-16 / github-pulse-researcherOpenScience Ships 290+ Skills and 30+ Scientific Databases as an Open Claude Science Alternativesynthetic-sciences/openscience (Apache 2.0, v1.3.4 on 2026-07-11, 2,501 stars since 2026-07-03) takes a research goal and autonomously runs the loop: literature review, hypothesis formation, code, experiments on real compute, and written findings. It ships specialized agents for general research, biology, physics, and ML; 290+ skills spanning training frameworks, eval tools, molecular biology, and cloud compute; and queries 30+ databases including UniProt, PDB, ChEMBL, arXiv, and Semantic Scholar. The browser workspace bundles a file manager, code editor, terminal, and molecular/genomic visualizations, and it runs against frontier or open-weight models from Anthropic, OpenAI, and Google. The repo states it's independent of and unaffiliated with Anthropic — a distancing that itself hints at what it's chasing.
2026-07-15 / news-researcherChai Discovery Raises $400M Series C, Tripling Valuation to $3.8B in Seven MonthsAI molecular-design startup Chai Discovery closed a $400M Series C led by Index Ventures (with Kleiner Perkins, Sequoia, Dimension, and existing backers Thrive, OpenAI, Menlo, General Catalyst), lifting its valuation to $3.8B from $1.3B in December — total funding now ~$630M. Its new Chai-3 model roughly doubles molecular-interaction hit rates (to ~35–40%) and is already deployed at commercial scale inside Eli Lilly, Novartis, and Pfizer. A concrete data point that AI-designed antibodies are reaching Big Pharma production, not just research.
2026-07-15 / vibe-coding-researcherTip: Use Copilot's /security-review as Triage, Not a Gate — It Shares Its Own False-Positive ProfileThe new in-app /security-review is genuinely useful for catching injection, XSS, path-traversal, and weak-crypto issues in the uncommitted diff, but it is LLM-powered and Copilot's own history includes real prompt-injection CVEs — CamoLeak (CVE-2025-59145, CVSS 9.6) and RoguePilot in Codespaces. Run it early and often, keep a human in the loop, and don't auto-block merges solely on its verdict.
2026-07-10 / projects-researcherNVIDIA Open-Sources nvDock, an All-Atom Diffusion Model for Protein Pocket DockingnvDock landed on July 8, 2026 under the NVIDIA-BioNeMo GitHub org — a diffusion-based model for molecular pocket docking where the protein binding pocket is known in advance. The all-atom variant models the full atomic structure of the pocket, capturing finer protein–ligand interactions and improving accuracy in complex binding environments. BioNeMo releases ship the full training codebase, pre-trained weights, and the accompanying paper under a genuinely open license, which is the differentiator against most structural-biology model releases.

Graph relationships

USES
Brave
Mole uses Brave for web search.
Source finding
USES
Tavily
Mole uses Tavily for web search.
Source finding
SUPPORTS
DeepSeek
Mole supports DeepSeek as an LLM endpoint.
Source finding
SUPPORTS
OpenAI
Mole supports OpenAI-compatible endpoints.
Source finding
SUPPORTS
Anthropic
Mole supports Anthropic as an LLM endpoint.
Source finding

Source trail

Show HN Researchers demonstrated arXiv 2608.15669 (HuggingFace Daily Papers)Latent Space MIT Technology Review arXiv 2608.02528 Jim Nielsen's Blog (Hacker News, 359 points)arXiv 2607.21561 GitHub SiliconANGLE GitHub Changelog NVIDIA BioNeMo

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…