Is there a relevant statement similar to the form of Moore\u2019s Law?", "5": "2026-08-23T07:07:43.694972"} {"0": 2, "1": "hackernews", "2": "https://github.com/JessyMorissette/CrossingBench", "3": "Show HN: CrossingBench \u2013 Modeling when data movement dominates compute energy", "4": "Show HN: CrossingBench \u2013 Modeling when data movement dominates compute energy. I built a small reproducible microbenchmark exploring when system energy becomes dominated by boundary crossings rather than intra-domain compute.
The model decomposes total energy into:\nC = C_intra + \u03a3 V_b \u00b7 c_b
Includes:
CLI sweeps
Elasticity metric (\u03b5) as a dominance indicator
CSV outputs
Working draft paper
*DOI
Looking for critique, counter-examples, or prior related work I may have missed.", "5": "2026-08-23T07:07:43.695160"} {"0": 3, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=45582973", "3": "Ask HN: Where can I track training cost trend for AI models?", "4": "Ask HN: Where can I track training cost trend for AI models?. I'm curious how the cost of training AI models (compute, energy, data, etc) has changed over time.
Are there any public resources or datasets tracking training costs for open-weight models (I'm guessing this data is hard to get for closed models, but happy to be proved wrong.)
I'm especially interested in understanding which architectural changes (e.g., attention variants, parameter sharing, mixture-of-experts) have led to major cost optimizations, and NOT just from the companies behind these models, but from anyone who has trained or replicated them.", "5": "2026-08-23T07:07:43.695182"} {"0": 4, "1": "hackernews", "2": "https://databridge.gitbook.io/databridge-docs/api-reference/endpoints/cache", "3": "Show HN: I Made an API for Persistent KV-Caching (Cache Augmented Generation)", "4": "Show HN: I Made an API for Persistent KV-Caching (Cache Augmented Generation). Hi HN!
I wanted to demonstrate an easy to use API for Cache Augment Generation. For any open source LLM available on Llama cpp, we can store the KV-cache and model state after it has processed a large corpus of documents, and then load that state in every time we query the document.
This leads to a drastic reduction in latency as well as compute/energy used by the model.
This demo is part of a larger system that I'm building called DataBridge[0] - with a focus on implementing new and useful techniques for knowledge retrieval - allowing developers to use the latest research in production.
I'd love to hear your feedback on DataBridge, and the CAG feature. If you have papers or particular techniques you'd like to see implemented, I'd love to hear about it :)
[0] https://github.com/databridge-org/databridge-core/", "5": "2026-08-23T07:07:43.695193"} {"0": 5, "1": "hackernews", "2": "https://news.ycombinator.com/item?id=34549724", "3": "Classical ML Still Relevant?", "4": "Classical ML Still Relevant?. ** Pls bear with me if its been already discussed or not related **
With all the proliferation of DL and LLM along with near unlimited compute, energy and bandwidth do we still need classical ML approach for solving the problems? Is DL / NN going to take over everything?", "5": "2026-08-23T07:07:43.695203"}