The Archive

Open-source research agent built on Claude Code outperforms OpenAI and NVIDIA systems in deep research benchmarking.

Intentional errors can be useful.

Jack Clark (Anthropic co-founder) estimates 30% probability AI research automation by end-2027, 60%+ by end-2028, citing rapid progress from coding to ML systems research.

Reddit discussion on em dashes as an unintended fingerprint of AI-generated content, creating social pressure to avoid natural writing patterns.

White House exploring pre-release vetting requirements for AI models, raising policy questions for open-weights distribution.

Appfigures finds visual model launches generate 6.5x more downloads — but most don’t convert that spike into revenue.

The retracted study on ChatGPT in education was already cited hundreds of times.

Reddit user describes subjective experience with Claude Opus 4.7 behavior and pattern-matching cognition; anecdotal observation without technical evidence.

Found some Open Source Claude skills from last 15 days. Some of them are pretty decent to use, personally liked the npm downloads one. Check out: **- brand-alchemy:** A brand strategy and naming skill that interrogates your thoughts for branding first, then applies phonosemantics, category design frameworks, and auto-checks domain availability across any TLD. **- npm-downloads-to-leads:** Give it a list of npm packages. It pulls 12 weeks of download data, scores each one by growth velocity, maps maintainers to GitHub and X, and gives you a ranked lead brief who built it, how to reach the...

Reddit discussion questioning Codex's current competitive position and download trends; lacks substantive analysis or new information.

SpecKV adapts speculative decoding's speculation length dynamically based on target model compression, improving LLM inference throughput.

Unsupervised ML framework detects structural anomalies in European regional socio-economic statistics using Eurostat NUTS2 data.

Simon Willison demonstrates TRE regex engine's resistance to ReDoS attacks via experimental Python binding, comparing resilience against standard library.

Review of multi-fidelity surrogate modeling techniques for composite materials prediction combining low and high-fidelity simulation data.

SHAP-based framework decomposes RL algorithm and hyperparameter contributions to generalization gaps in robotic control tasks.

I usually have two or more Claude Code sessions open at once. One in the backend repo, one in the frontend. Half the time I'd be in the frontend asking "wait, what shape did the user object end up as?", then alt-tab, ask the backend session, copy the answer, alt-tab back, paste. The other Claude was right there. It already knew. I was the bottleneck. So I wrote a plugin called Relay. In the frontend window I just say: ▎ask the backend session what the user object looks like The backend session sees the question between turns, answers it, and the reply pops up in my frontend session as a n...

Knowledge distillation from LLMs to compact open-source models for cross-language code clone detection without black-box inference costs.

Reddit user reports degraded performance in Claude Opus 4.7 compared to 4.6, speculating smaller base model or optimization tradeoffs.

Layer-wise peeling framework monitors transformer training dynamics by locally optimizing each layer against intermediate representations.

Reddit user reports ChatGPT extended thinking feature enabled by default; likely user-facing feature discussion without technical depth.

Pattern-based AI-assisted methodology for rapid sensor-driven application development using Pegasus workflows on FABRIC testbed.

Second-order retraction-free optimization method on Stiefel manifolds via Newton-Schulz iteration with quadratic convergence.

Reddit post recounts Sam Altman interview on talent retention at OpenAI during Meta's AI hiring competition.

PLACE: closed-form persistent-homology pipeline for point cloud and graph classification with margin-based guarantees and per-prediction certificates.

VideoNet benchmark with 1,000 domain-specific actions revives action recognition evaluation for vision-language models.

HAAS framework enables adaptive task allocation between humans and AI systems in software engineering and manufacturing contexts.

JACTUS unifies parameter-efficient fine-tuning and model compression into single joint optimization framework.

Statistical approach improves Monte Carlo estimation of Shapley values and semivalues for model explainability.

User shares prompt injection technique to reduce em dash usage in Claude via system preferences.