It's frustrating that it's been more than 20 years since Katrina and we haven't done too much about this. Relocation seems like a multi-decade effort. > Coastal Louisiana faces sea level rise of around 10 to 23 feet, according to the analysis published in May in the journal Nature Sustainability. The impacts will be bleak: around 75% of its remaining wetlands are set to be lost and its shoreline could retreat inland by up to 62 miles, the scientists found. https://lite.cnn.com/2026/05/25/climate/new-orleans-sea-l...

Really excited to see this! The problem is easy to understand, but wickedly difficult to solve. > The result is also notable for how it was found. The proof came from a new general-purpose reasoning model, rather than from a system trained specifically for mathematics, scaffolded to search through proof strategies, or targeted at the unit distance problem in particular. As part of a broader effort to test whether advanced models can contribute to frontier research, we evaluated it on a collection of Erdős problems. In this case, it produced a proof resolving the open problem. https://openai.com/index/model-disproves-discrete-geometr...

Really impressed with with Antirez is doing over at https://github.com/antirez/ds4. My main gripe is that I'm running a 128GB Strix Halo machine with ROCm, and that's not a primary target for DS4. I think the endgame will be llama.cpp fixing this issue (https://github.com/ggml-org/llama.cpp/issues/22319 requests model support for DS4), and then running that via lemonade's build of llama.cpp (https://github.com/lemonade-sdk/llamacpp-rocm/releases) with an Unsloth quant of DeepSeek v4 Flash (like https://huggingface.co/unsloth/DeepSeek-V4-Flash). Just waiting for progress on llama.cpp, since everything else should be ready.

One interesting application of LLMs is to have them read through privacy policy and terms of use for you so you can look for anything particularly hazardous to your considered use cases. I did this for opencode.ai and discovered that their Zen product provides better guarantees about privacy than their Go product, so I'm switching to Zen, even though it's a little bit more expensive.

The real insight over the past 2-3 days has been a combination of: 1. Local models like Qwen 3.6 and Deepseek v4 Flash (even in a 2-bit quant!) are quite useful, and understand integrations like tool calls well. 2. A harness like OpenCode is an immense step forward in keeping local models on-task. This site was crafted almost entirely locally with Qwen 3.6 35B-A3B. 3. It's becoming clear that adding on additional skills and integrations will pay huge dividends in getting it to operate autonomously for hours. I think a project like my good friend's https://dex.rip for cross-session memory and tracking will take this even farther.

Working with some local models (Qwen 3.6 35B-A3B and Qwen 3.6 27B UD, both in Q_6K) as well as some cloud-hosted models (Deepseek v4 Flash hosted through OpenCode.ai) to build a simple feed page, can be hosted through CGI or as a server. This is the first deployment - fairly happy with how quickly it came together without any frontier model contribution!