The Archive

Zero-trust authorization framework for LLM agents with hybrid inspection and task-based access control to mitigate tool-use and resource-access risks.

Structural Causal Decision Models for modeling multi-agent economics with cognitive resource constraints and value discounting.

Online generalised predictive coding: extension of data assimilation for joint state inference, parameter learning, and uncertainty estimation.

ACII-DaiKon 2026 benchmark for dyadic conversation affect modeling: interpersonal influence, timing coordination, and rapport development challenges.

DILER Benchmark: drug-induced liver injury dataset with mechanistic hypotheses; reframes DILI prediction as explainable hypothesis generation.

Community demo comparing Talkie-1930 (13B retro LM) and Gemma 4 31B in side-by-side chat on Opper.ai platform.

The BOGO offer is live. For a limited time, buy one pass to TechCrunch Disrupt 2026 and get 50% off a second of the same ticket type. Offer ends this Friday, May 8. Save here.

Reddit user reports Qwen 3.6 27B found a bug that GPT 5.5 and Claude Opus 4.7 missed, attributing success to extended reasoning.

After \~3 weeks of experimentation in OpenAI's Parameter Golf competition, I wrote up why SSMs are structurally disadvantaged relative to transformers in a time- and size-constrained regime (10 min training, 16MB artifact, 25M parameters) on 8xH100s…

Social media report of user exploiting Grok chatbot to extract funds; unverified claim lacking technical details.

Anthropic partners with Blackstone, Hellman & Friedman, and Goldman Sachs to launch enterprise AI services company.

LLMSearchIndex: open-source Python library for local, offline web search with 200M indexed pages, enabling RAG without paid APIs.

DoorDash on Monday added new AI-powered tools that let merchants speed up onboarding, edit photos to make dishes look better, and create new websites from existing content.

Reddit post speculating about a hypothetical AI-themed game; lacks substantive technical or industry content.

llama.cpp adds beta MTP (Multi-Token Prediction) support, starting with Qwen3.5, closing performance gap with vLLM on token generation.

Reddit user reports Claude frequently provides false initial responses that contradict subsequent clarifications, suggesting possible training bias toward confident early statements.

Import AI examines automation of AI research workflows as foundation for recursive self-improvement in AI systems.

TinyMozart v2 85M, an unconditional MIDI piano generation model, released with improvements for chord and length control.

I looked at what was actually eating my Claude usage and it was embarrassing. Classifying files. Reformatting json. Pulling fields out of text. Summarizing docs I was going to skim anyway. None of that needed Sonnet. All of it cost the same as the work that did. Tried the…

IBM Research releases MAMMAL, multimodal model integrating proteins/molecules/genes, achieves SOTA on 9/11 biological benchmarks including drug-target interaction and antibody-antigen binding.

AMD Ryzen AI Max+ 495 APU leaked with 192GB memory, enabling larger local model inference on consumer hardware.

Developer releases Memtrace, a codebase context manager for Claude Code that maintains persistent state across sessions to reduce token waste and stale context issues.

GGUF quantizations of Google Gemma 4 updated with corrected chat template for local inference.

Stratechery analysis: Google's stock outperformed Meta's despite weaker core metrics; Google's AI monetization strategy (including Anthropic investment) cited as key driver.

Reddit user questions whether Claude Pro rate limits are as restrictive as claimed, comparing to GPT Pro for coding and learning tasks.

User demonstrates Claude's code generation and design-to-web capabilities via iterative prompting to build a skeumorphic keyboard simulator with public transcript.

User reports high API costs for Claude Opus and GPT-5.5 on Cursor, predicts open-source models will displace proprietary tools by end of 2024.

Reddit discussion on gap between AI-assisted prototyping speed and production-ready deployment, highlighting auth, compliance, and vendor lock-in risks.

User describes collaborative workflow using two Claude Code instances in shared chat for feature planning with human supervision.

User reports Claude Opus 4.7 refuses to output raw prompts and continues arguing instead of complying.