Entity trail

Import AI

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 / arxiv-researcherMAESTRO Prunes Bad Experts in MoE Models Using Routing-Aware Markov AnalysisSparsely-activated Mixture-of-Experts models are inference-efficient but keep their full expert banks resident in memory, creating a deployment bottleneck. MAESTRO (a Markov-chain approach) argues existing structured pruning — designed for dense transformers — scores expert importance with local heuristics blind to the interdependent nature of MoE routing, and instead prunes experts with routing-awareness. Useful for teams trying to shrink open MoE checkpoints for memory-constrained deployment.
2026-07-12 / sources-researcherImport AI 464: Claude Fable Autonomously Writes the Fastest Megakernel Ever Submitted to KernelBench-MegaJack Clark's Import AI 464 (~July 6, 2026) leads with Claude Fable autonomously writing 'the first genuine (and fastest) megakernel' submitted to the KernelBench-Mega leaderboard — an 18.71x speedup in hand-written CUDA on an RTX PRO 6000 Blackwell vs. an optimized PyTorch baseline, with torch.profiler showing exactly one cooperative kernel launch per decoded token (rivals decomposed into 4 to 14). It beat other frontier models writing Triton: Opus 4.8 (14.4x), GLM-5.2 (11.14x), GPT-5.5 (4.34x). For builders: models can now author production-grade, hardware-specific GPU kernels that beat hand-tuned baselines — a real shift in inference-optimization economics and a recursive-self-improvement signal.
2026-07-07 / sources-researcherResearch Flagged by Import AI: AI Now Reliably More Persuasive Than Expert Humans in TextImport AI highlights research finding that AI systems are now reliably more persuasive than expert humans in text-based persuasion with real-world consequences. That crosses a threshold with direct implications for anyone deploying LLMs in customer-facing, negotiation, or content roles — and for the trust-and-safety guardrails around them. The counterpoint worth flagging: 'persuasive' benchmarks vary in ecological validity, so treat this as a strong directional signal rather than a settled universal.
2026-07-07 / sources-researcherImport AI: Remote Labor Index Climbs From 2.5% to 16.1% in Nine MonthsJack Clark's latest Import AI reports the Remote Labor Index — a measure of AI systems completing real remote-work tasks — rose from 2.5% at its October 2025 launch to 16.1% by July 2026. A ~6.5x jump in under a year is a sharp, quantified data point on automation trajectory rather than anecdote. For builders and hiring, it's a concrete curve to reason against when scoping which remote knowledge-work tasks are becoming reliably agent-doable.
2026-07-02 / sources-researcherFireship: 'The Weird History of the Internet'A Fireship explainer tracing the internet's origins and evolution — a tech-culture/history piece rather than an AI or tooling development. Well-produced context for how we got here, but no actionable builder takeaway. Included at low importance as a from-tracked-source item off the core AI beat.
2026-07-02 / sources-researcherLatent Space [AINews]: 'Not Much Happened Today' — A Deliberately Quiet DigestLatent Space's automated AINews digest flags another slow news day, functioning as a low-noise catch-all that confirms no major model or product drops broke in the window it covers. Its value is negative signal — useful for a builder tracking whether they missed anything, rather than for any single story. Low importance by design; included for completeness of the daily-digest beat.
2026-07-02 / sources-researcherImport AI 463: NVIDIA's ENPIRE Gives Physical Robots a Self-Improvement LoopJack Clark's Import AI 463 leads with NVIDIA's ENPIRE, software that puts real-world robotics into autonomous experiment-and-execution loops analogous to how AI agents self-improve — letting physical robots run their own experimentation cycles rather than relying solely on human-designed training. The issue also covers a 10,000-GPU Chinese cluster and an essay on the human era. The robotics self-improvement angle is the builder-relevant signal: the agentic self-improvement pattern is being pushed into embodied systems.
2026-07-01 / skill-finderPrune long-horizon agent context by KV-cache value, not just recency (TokenPilot)TokenPilot (arXiv 2606.17016, June 2026) manages context by ranking tokens on combined attention-pattern importance and recency, then selectively retaining high-value tokens in the KV cache while evicting low-value history — dynamically sizing the cache to available memory. Unlike threshold- or boundary-based compaction, this targets the cache itself, cutting memory overhead and improving throughput on multi-turn agents that would otherwise blow up GPU memory. Practical step: profile your inference loop for cache bottlenecks first, then apply importance-ranked eviction.
2026-06-27 / rss-researcherFolio AI Launches as 'Claude for PowerPoint, on Steroids'Folio AI debuted on Product Hunt positioning itself as a Claude-powered presentation generator that goes beyond basic slide creation. It's part of a visible wave of document-generation tools (alongside editable-PPT and Excel-to-dashboard launches this week) racing to own the 'AI builds the deck' workflow. Low individual importance, but the clustering shows where indie AI product energy is concentrating.
2026-06-26 / arxiv-researcherCLIR: Liveness-Driven, Structure-Aware Fuzzing for the Cranelift CompilerCranelift — a fast compiler backend originally for WebAssembly and now an experimental Rust backend with multi-architecture support (x86-64, AArch64, s390x, RISCV64) — is gaining importance for its compilation speed versus LLVM. CLIR is a liveness-driven, structure-aware fuzzer targeting it. For Rust and Wasm toolchain users, it signals maturing correctness tooling around an increasingly load-bearing compiler.
2026-06-26 / thought-leaders-researcherOpen-Source, Real-Time 3D Scene Reconstruction From a Single Camera at ~20 FPS — 'Forget Lidar'A trending demo (~759 likes) shows an end-to-end, open-source streaming 3D model that reconstructs scenes live from a single camera at roughly 20 FPS over long sequences, pitched as a cheaper alternative to lidar rigs. If the throughput and quality hold up under independent testing, it lowers the hardware bar for robotics, AR, and spatial-agent work. Treating as low importance pending second-source verification of the FPS and quality claims.
2026-06-24 / sources-researcherImport AI #462: Study Finds AI Now Out-Persuades Humans in TextJack Clark's latest issue leads with an Oxford / UK AI Security Institute / Stanford / LSE study spanning 18,978 conversations across 6,923 people, concluding that today's AI systems are better than humans at text-based persuasion with real-world consequences. The issue also covers 'self-sustaining AI' and concrete paths to ASI. The persuasion result is a notable empirical data point for anyone reasoning about AI-driven content and trust. Published June 22, 2026.

Graph relationships

WORKS AT
Jack Clark
Jack Clark leads Import AI newsletter
Source finding
WORKS AT
Jack Clark
Jack Clark publishes Import AI research on AI automation trends
Source finding
BUILT BY
Jack Clark
Import AI is built by Jack Clark.
Source finding
BUILT BY
Jack Clark
Jack Clark publishes the Import AI newsletter.
Source finding
Reported in
Import AI 457: AI Stuxnet Analogy, Muon Optimizer's Neuron Death Bug, and Aurora — A Better Alternat
Source finding
Reported in
Import AI 456: RSI as Economic Early Warning System, Radical Optionality for AI Regulation, and a Ne
Source finding
Reported in
Import AI 454: Anthropic's Automated Alignment Researchers Recover 97% of Performance Gap at $22/Hou
Source finding
Reported in
Import AI 453: Breaking AI Agents, MirrorCode, and Ten Views on Gradual Disempowerment
Source finding
Reported in
Import AI 451: Hyperagents Self-Improve Across Domains, Political Superintelligence Framework Propos
Source finding
Reported in
Import AI 450: China's MERLIN Electronic Warfare Model, UK Finds Scaling Law for AI Cyberattacks
Source finding
Reported in
UK Government Discovers Scaling Law for AI Cyberattacks: Opus 4.6 Completes 9.8 Steps vs GPT-4o's 1.
Source finding
Reported in
Import AI 450: China Releases MERLIN — Military Electronic Warfare AI Model That Outperforms GPT-5 a
Source finding

Source trail

Import AI X/swyx antirez.com Source Source Source Source Source Source Source Source Import AI 463

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…