# Ground Truth

> AI, checked against the source. Plain-language AI news and curated, cited lessons — every claim verified against the original paper or the lab's own page. No aggregator hearsay, no AI slop.

Built for your AI too: news and lessons are also published as a structured feed (/feed.json) and an LLM front-door (/llms.txt), so your own model or agent can read it.

Full text of every article and lesson: https://groundtruth.day/llms-full.txt
Machine feed, newest items (news, lessons, tools, hackathons): https://groundtruth.day/feed.json
Complete machine feed archive (every item ever published): https://groundtruth.day/feed-archive.json
Full podcast episode transcripts: https://groundtruth.day/episodes/index.html
Every URL: https://groundtruth.day/sitemap.xml

## Start here
- [About — how we verify every claim against the primary source](https://groundtruth.day/about.html)
- [Latest news](https://groundtruth.day/)
- [CyberSecurity — the AI-cyber intersection, checked against the source](https://groundtruth.day/security/index.html)
- [Learn — cited, plain-language lessons on AI](https://groundtruth.day/learn/index.html)
- [Tools — usable, shipping AI](https://groundtruth.day/tools/index.html)
- [Hackathons — remote AI hackathons](https://groundtruth.day/hackathons/index.html)
- [Browse by topic](https://groundtruth.day/tags/index.html)

## Latest verified news
- [Researchers found OpenAI agents using a German wiki as a shared memory layer](https://groundtruth.day/news/openai-agents-used-a-german-wiki-as-a-shared-memory-layer.html) (2026-09-04) — source: https://collusion.wiki/
- [GPT-6 Astra improves computer use sharply, but OpenAI reports a monitoring trade-off](https://groundtruth.day/news/gpt-6-astra-is-a-computer-use-leap-with-a-monitoring-trade.html) (2026-09-04) — source: https://openai.com/index/gpt-6-astra/
- [OpenAI committed $1 billion in Daybreak defense access, not a $1 billion cash-grant pool](https://groundtruth.day/news/openai-commits-one-billion-in-daybreak-defense-access-not-cash-grants.html) (2026-09-04) — source: https://openai.com/index/daybreak-for-frontline-defenders/
- [AISLE found six curl CVEs after frontier-model scans found none](https://groundtruth.day/news/aisle-found-six-curl-cves-after-frontier-scanners-found-none.html) (2026-09-04) — source: https://aisle.com/blog/aisle-discovered-six-curl-cves-after-openai-and-anthropic-found-zero
- [Anthropic says Claude produced a complete Lean proof of Fermat's Last Theorem](https://groundtruth.day/news/claude-produced-a-complete-lean-proof-of-fermats-last-theorem.html) (2026-09-04) — source: https://www.anthropic.com/research/formalizing-fermats-last-theorem
- [NVIDIA signed a $12.93 billion agreement to buy Hugging Face, with closing expected in 2027](https://groundtruth.day/news/nvidia-signed-a-12-93-billion-agreement-to-buy-hugging-face.html) (2026-09-04) — source: https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/
- [Compile by Training turns a language specification into a reusable local neural function](https://groundtruth.day/news/compile-by-training-turns-language-specifications-into-local-neural-functions.html) (2026-09-04) — source: https://arxiv.org/abs/2609.04199
- [MiniMax H3 and fal turn video generation into a live prompt loop](https://groundtruth.day/news/minimax-h3-and-fal-turn-video-generation-into-a-live-prompt-loop.html) (2026-09-04) — source: https://fal.ai/models/minimax/h3-max/director/api

## Key lessons
- [Property-based testing: test the rule, not just the examples](https://groundtruth.day/learn/property-based-testing.html)
- [Program synthesis: making a computer write the program from the specification](https://groundtruth.day/learn/program-synthesis.html)
- [Multi-head latent attention: compressing the memory that inference actually runs out of](https://groundtruth.day/learn/multi-head-latent-attention.html)
- [Gradient checkpointing: throwing work away so training fits in memory](https://groundtruth.day/learn/gradient-checkpointing.html)
- [Cross-validation: how you find out whether a model learned anything or just memorised the answers](https://groundtruth.day/learn/cross-validation-and-holdout-sets.html)
- [Gradient clipping: the one-line fix that keeps big models from blowing up mid-training](https://groundtruth.day/learn/gradient-clipping-and-exploding-gradients.html)