Rex "garbage in" Douglass Ph.D.
@RexDouglass
Applied Scientist in Industry. Previously Microsoft, UCSD, Princeton (PhD) Follow me for recreational methods trash talk.
New Link Aggregator/News Digest 👉 github.com/rexdouglass/Re… I'm launching Rex Reviews, a new, curated digest of the latest in AI and ML. Topics include:🧠 NLP & Agents 👁️ Machine Vision 🔗 Causal Inference 🛠️ General AI/ML Tooling #AI #MachineLearning #NLP #ComputerVision…

Here is Nietzsche talking about attention, parsimonious sampling, and how hallucination is the default mode of human perception in 1886.
Is publication bias “as Old as Science Itself?” pharmacytimes.com/view/publicati…
suddenly remembering that time Fitbit data leaked and revealed all these American dudes jogging around deep inside of Syria
Rule Number One: Do not take a selfie to sign up for a gossip app while working on the flight line at a secret military base.
Chain-of-Thought Is Not Explainability @FazlBarez, Tung-Yu Wu, Iván Arcuschin (@IvanArcus), Michael Lan, Vincent Wang (@vinnylarouge), Noah Siegel (@noahysiegel), Nicolas Collignon (@nccollignon), Clement Neo (@_clementneo), Isabelle Lee (@wordscompute), Alasdair Paren,…
What’s the purpose of a theorem? (Hint: It’s not what you think.) statmodeling.stat.columbia.edu/2025/07/26/wha…
To me, the most surprising Claude 4 result was that they couldn't evaluate alignment, as alignment faking transcripts leaked into the training data. Please protect your datasets! It's not just benchmarks Fortunately, there's a great new tool to easily & securely share datasets!
I figured out how to get Gemini Image Segmentation to draw little portraits for me with its bounding boxes. I'll explain how down-thread. Here's SpongeBob. (For whatever reason, it likes to draw these portraits sideways.)
did u know you can use the new Gemini image segmentation feature in… a lot of different ways
Safety tips for devs & vibe coders diving into Gemini CLI: - The CLI will ask for permissions. Read it. Review it. - gemini --sandbox is your safe playground (macOS Seatbelt, Docker and Podman support) - gemini --checkpointing = instant "undo" button with /restore - Back up,…
Study finds "retractions due to data problems have increased significantly" since 2000. tandfonline.com/doi/full/10.10…
A Simulated Reconstruction and Reidentification Attack on the 2010 U.S. Census: Full Technical Report. arxiv.org/abs/2312.11283
The two faces of academic social science statmodeling.stat.columbia.edu/2025/07/25/the…
For anyone doing their first NeurIPS rebuttal, some advice (full post below) Note: 50% of NeurIPS spotlights were rejected by an independent review committee in an RCT. *Before* LLMs! Don't take the RNG too seriously. But don't be too defensive to engage with real feedback
Can “screenshotting a page and feeding it to the LLM” can replace existing document processing solutions? I saw some discussion on this on HN, and was inspired to write a blog post on this. TL;DR you absolutely should use LLMs for doc processing, but you need a lot of tuning to…
LLM APIs aren't enough for production document parsing - here's what you need instead. While frontier models like GPT-4.1, Claude Sonnet 4.0, and Gemini 2.5 Pro have made traditional OCR obsolete, screenshot-only parsing still has critical gaps for enterprise use: 📊 Accuracy…
I take it back my advice did not go far enough. Being employee #2 and childhood friends is not enough. If you are a founding engineer, you need to be legally married to the CEO.
How To Make Sure You Makr It on The Evac Boat When Your Founders Are Acquihired tldr: make it so that not acquiring you is materially damaging to the acquihire! 1) Learn the full history of the product direction. Not just things that have worked now but all the things that were…