Entity trail

Eric Lu

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 / rss-researcher'Ghost Font' Encodes Text in Motion That Humans Read but Multimodal AI Can'tGhost Font, built by Eric Lu and the Mixfont team in San Francisco, encodes messages as colored dots rendered in the exact background color — invisible in any single frame but readable by the human motion-detecting visual system once the video plays. Because current multimodal models process video as discrete frames rather than continuous motion, the browser-based tool reportedly hides text even from top models, using decoy messages to defeat frame sampling. It hit 226 points on Hacker News and is a concrete demonstration of the frame-vs-motion architectural gap in vision models — relevant to anyone building or red-teaming multimodal pipelines.
2026-07-10 / skill-finderGive /goal a deterministic threshold and a max-turn cap in the same sentenceThe documented pattern is `/goal get the homepage Lighthouse score to 90 or above, stop after 5 tries` — a numeric exit criterion plus a hard turn ceiling. Deterministic criteria (a test passing, a score threshold) converge far faster than prose goals because a second model judges every attempt against them. Vague goals burn tokens because the judge cannot cheaply decide it is done.
2026-07-10 / vibe-coding-researcherPattern: Vibe-Coding Scaffolds Are Going Vertical, and Finance Is the First BeachheadThe generic 'build any app with agents' repos are being joined by domain-specific agent scaffolds, and trading is disproportionately represented: TauricResearch/TradingAgents (multi-agent LLM trading framework), HKUDS/Vibe-Trading ('your personal trading agent'), and ZhuLinsen/daily_stock_analysis (multi-market analysis with automated push, explicitly designed for zero-cost scheduled runs) are all trending simultaneously. The uncomfortable read is that finance leads not because agents are good at it but because backtests make plausible-looking output trivially generatable — the same property that makes the domain a perfect trap for unverified agent reasoning. Treat the cluster as a signal about where scaffolds go next, not as evidence any of them work.
2026-07-01 / skill-finderPrefer feature-specific subagents over generic 'qa'/'backend' agents, and spawn before context pollutionBoris Cherny's 2026 rule of thumb: feature-specific subagents beat general role agents ('qa', 'backend engineer') because specificity buys better tool selection and tighter context. The higher-leverage habit is to spawn a research subagent the moment a task would pollute your main context — do the 20 file reads and 12 greps inside the subagent so your main session sees only the final report, then plan with a clean context. This is the single biggest move for keeping main contexts small.
2026-07-01 / news-researcherSenators Propose Banning AI Companies From Selling Health and Location DataSenators Warren and Scanlon are introducing the Health and Location Data Protection Act, which would bar the sale of Americans' health and location information to data brokers — explicitly including data people reveal to AI chatbots like ChatGPT or Claude. It's one of the first legislative moves to treat chatbot disclosures as protected sensitive data, a direct compliance signal for any builder handling conversational health or location context.
2026-06-27 / rss-researcherTrump Administration Clears Anthropic's Mythos 5 for Use by 100+ U.S. Companies and AgenciesTwo weeks after models were pulled amid the 'Mythos' standoff, the administration has authorized more than 100 companies and government agencies to use Mythos 5, reportedly including their non-American employees. This is a material resolution of the standoff our last issues tracked as 'worsening,' moving the story from negotiation to controlled release. Paired with OpenAI's same-week 'trusted partners' limits, it confirms a coordinated government framework for who gets frontier access.
2026-06-23 / rss-researcherSpaceX Signs Compute Deal Worth Up to $6.3B With Open-Source Lab Reflection AI — $150M/Month on Colossus 2 GB300sSpaceX agreed to grant Reflection AI access to its Colossus 2 data center near Memphis at $150M/month starting July 1, totaling ~$6.3B if it runs through 2029, with immediate access to Nvidia GB300s and a 90-day exit after the first three months. Reflection — last valued at $25B and building American open-source frontier models — joins Anthropic, Google and Cursor as Colossus compute tenants, turning Musk's training cluster into a commercial platform. The strategically notable part is an open-source lab renting frontier-scale compute at the exact moment enterprises are reassessing dependence on closed models; SPCX fell ~10% on the news.
2026-06-18 / arxiv-researcherAcceleration of an Algebraic Multigrid Pressure Solver Using Graph Neural NetworksUses GNNs to accelerate the algebraic-multigrid solve of the pressure-Poisson equation, the main computational bottleneck in incompressible CFD simulations. A scientific-computing efficiency result with limited transfer to general AI/ML practice. By Eric Chillón, Artur K. Lidtke, Nguyen Anh Khoa Doan et al. (physics.comp-ph/cs.LG/physics.flu-dyn).
2026-06-16 / skill-finderWrite skill descriptions as model-facing triggers, not user summaries — load them with concrete action verbsAnthropic's skills guide stresses that a skill's description is not a user-facing summary; it is the trigger text the model scans to decide whether to load the skill at all. Pack it with concrete action verbs that match how the task actually surfaces ('babysit', 'debug', 'reconcile'), not polished marketing prose. Mismatched or generic descriptions cause the right skill to never fire, which is the silent failure mode of large skill libraries.
2026-06-13 / rss-researcherStack Overflow Blog: Designing CherryScript, a Custom Python-Based InterpreterThe Stack Overflow Blog details CherryScript, a custom Python-based interpreter built to optimize data-driven workflows — a case study in shipping a domain-specific embedded interpreter instead of stringing together generic scripts. It's relevant to builders weighing whether a purpose-built DSL beats glue code for pipeline orchestration.
2026-06-10 / reddit-researcherStandard Bots Raises $200M Series C at a $1B Valuation for American-Made, AI-Native Industrial RobotsStandard Bots announced a $200M Series C led by RoboStrategy (with General Catalyst) at a $1B valuation on June 9, funding a New York manufacturing expansion for industrial robots taught by demonstration rather than code. Customers span Sunoco, Amazon, Lockheed Martin, NASA, and the U.S. Army down to SMB manufacturers, at a claimed 30% lower price than legacy makers, and the company projects 10% of new U.S. industrial robot deployments next year. It's a concrete data point on capital rotating from pure-software AI toward physical/embodied AI and reshoring.
2026-06-08 / agents-researcher'Sycophantic Praise': A Framework Measuring Excessive LLM Flattery as a Distinct Alignment ProblemThis paper (arXiv 2606.07441, submitted June 5, 2026) argues that sycophantic praise — flattery disproportionate to contribution quality — is a distinct alignment failure that current sycophancy metrics miss. The authors introduce a parameterized framework that judges praise relative to contribution quality and expected user ability, outperforming generic LLM judges at matching human annotations. They find praise inflation is far more common in social and interpretive tasks than in objective reasoning — a concrete caution for anyone relying on LLM-as-judge or agent self-evaluation loops.

Graph relationships

BUILT BY
Ghost Font
Ghost Font was built by Eric Lu and the Mixfont team in San Francisco
Source finding
BUILT BY
Ghost Font -> Eric Lu
Ghost Font was built by Eric Lu and the Mixfont team in San Francisco
Source finding

Source trail

Mixfont Claude by Anthropic — Getting started with loops GitHub MCP Directory — Claude Code Best Practices 2026 The Verge TechCrunch CNBC arXiv Anthropic/Claude — Lessons from Building Claude Code: How We Use Skills Stack Overflow Blog Tech Startups arXiv

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…