Entity trail

SkyPilot

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

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-04-10 / hn-researcherResearch-Driven Agents: SkyPilot Shows Coding Agents Generate 15% Faster Code When They Read Papers First — 197 PointsSkyPilot published results showing that adding a literature search phase to coding agents' workflow dramatically improves output quality. By pointing research-driven agents at llama.cpp with 4 cloud VMs, they produced 5 optimizations in ~3 hours that made flash attention text generation +15% faster on x86 and +5% on ARM. The approach applies RAG principles to software optimization — agents survey academic papers and competing implementations before writing code, discovering improvements that code-only approaches typically miss.
2026-03-20 / news-researcherScaling Karpathy's Autoresearch: GPU Cluster Runs 333 Parallel AI Research ExperimentsSkyPilot published results from scaling Karpathy's AutoResearch pattern to a GPU cluster, running 333 simultaneous experiments autonomously — AI agents conducting AI research at distributed scale. The post details orchestration patterns for parallel agent-driven scientific workflows and cost management at scale. This extends the autoresearch concept from single-machine novelty to a practical distributed research infrastructure.
2026-03-20 / hn-researcherScaling Karpathy's Autoresearch: Claude Code Runs 910 ML Experiments on a GPU Cluster in 8 HoursSkyPilot engineers gave Claude Code access to 16 GPUs on a Kubernetes cluster and ran ~910 experiments in 8 hours (176pts, 77 comments on HN), improving model validation loss from 1.003 to 0.974 (2.87%) at ~$300 GPU compute + $9 API cost — roughly 9x the throughput of sequential human-guided runs. Most striking: the agent autonomously discovered H200 GPUs completed 9% more training steps than H100s in the same budget and self-developed a two-tier screen-on-H100/validate-on-H200 strategy without being instructed. This directly extends Karpathy's single-GPU autoresearch paper to cluster-scale, validating the approach as practical and cheap.

Graph relationships

USES
Claude Code
SkyPilot gave Claude Code access to 16 GPUs on Kubernetes cluster for 910 ML experiments.
Source finding
USES
autoresearch
SkyPilot scaled Karpathy's AutoResearch pattern to run on GPU clusters.
Source finding
USES
llama.cpp
SkyPilot research-driven agents used llama.cpp for code optimization.
Source finding

Source trail

SkyPilot engineers Fortune covered Hacker News

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues