brendanashworth

This link gives 404 Page not found, perhaps https://amplitude.com/blog/amplitude-and-statsig-partnership is a better URL [1]

[1] https://news.ycombinator.com/item?id=48028790

Out of the box, Opus/GPT-5.4 uses checks like `tsc --noEmit` or `python -m py_compile ...`, so compile checks on a Swift/Objective-C app probably get you pretty far. Also just setting up a Dockerfile maybe with swift:5.10-focal would give the agent the right tools to verify its own work.

Sandboxing isn't about having the perfect devex environment, most dev sandboxes (looking at codex cloud) don't actually have full verification available. Sometimes Github actions CICD is all you can get!

Is there a reason people don't use SHAP [1] to interpret language models more often? The in-context attribution of outputs seems very similar.

[1] https://shap.readthedocs.io/en/latest/

If choosing a Bayesian approach in a clinical trial can reduce the number of recruited subjects, I would imagine the pharma industry is strongly incentivized to adopt it.

Would a wearable model like this gain in predictive power by adding FHIR/EHR inputs? CoMET [1] has great AUC even without daily/lifestyle metrics like steps, awakenings at night, or vo2max estimates (which are present in say apple health). It would be interesting to know how much you would gain from having a holistic model that sees both inputs.

[1] https://arxiv.org/pdf/2508.12104v1

How does this compare to obstore? [1]

[1] https://developmentseed.org/obstore/latest/

Cursor [1] has excellent codebase RAG, and great recall of third party documentation.

Use "instruction @web @Flask" to search the internet and Flask docs before answering, for example.

[1] https://www.cursor.com/

Property boundaries aren't delimited by GPS coordinates, but rather a surveyor's instructions on how to locate the boundary from a set of known landmarks. In most US states, the Public Land Survey System [1] gives "meridians" and "baselines" for each region serving as the regional center. Radiating out, every 6 miles, is a new "township," with its corners serving as the local landmark for property boundaries.

[1] https://en.wikipedia.org/wiki/Public_Land_Survey_System

I am confused by the benchmarks you provided.

(1) The "baseline" comparison was to PyPDF + Naive RAG. For the LlamaParse evaluation, you appear to have used a different RAG pipeline, called "recursive retrieval." Why not use the same pipeline to demonstrate the improvement from LlamaParse? Can you share the code to your evaluation for LlamaParse?

(2) I ran the benchmark for the PyPDF + Naive RAG solution, directly copying the code on the linked LlamaIndex repo [1]

I got very different numbers: mean_correctness_score 3.941 mean_relevancy_score 0.826 mean_faithfulness_score 0.980

You reported: mean_correctness_score 3.874 mean_relevancy_score 0.844 mean_faithfulness_score 0.667

Notably, the faithfulness score I measured for the baseline solution was actually higher than that reported for your proprietary LlamaParse based solution.

[1] https://github.com/run-llama/llama-hub/tree/main/llama_hub/l...