1. Show HN: Panel – A research workspace where the agent can build its own panes (github.com)
Contribute to greentfrapp/panel development by creating an account on GitHub.
2. Vibe Coding is the new Internet Dating? (joecmarshall.com)
Developer and Author from Austin, TX.
3. Show HN: Sunk Cost – How long until a local LLM rig pays for itself? (sunkcost.ai)
Which open models fit your machine, how fast they run, and how many tokens it takes to pay back the hardware against an API. Honest enough to say never.
4. Inverse-Square Law (blog.coredump.cx)
5. Show HN: Loss. a tiny satire about AI progress (workatloss.com)
Clock in. Press the button. Build an increasingly automated future. A free incremental game by Bobby Bobak.
6. Oil Executives Say the Great Fuel Crisis Is Here (www.wsj.com)
7. Tracking the slide into authoritarianism wihin the United States of America (protectdemocracy.org)
How's American democracy doing? This tool gives us a way to understand how authoritarian actions and pushback against them add up.
8. The biggest dinosaurs couldn't sit on their eggs (www.cbc.ca)
9. Data races and the limits of ThreadSanitizer in C and Go (theconsensus.dev)
10. Firefox 156 shows ads in the address bar (dubbed "Firefox Suggest") (www.heise.de)
Mozilla has released Firefox 156. The new version speeds up the PDF viewer and optimizes memory and CPU usage.
11. When code is a maze, smart developers make maps (medium.com)
12. Show HN: The bottom 50% of U.S. households are short after essentials (BLS data) (whats-left-over.pages.dev)
How much U.S. households have left after essentials, by income group, 2000–2026, and what investing it would add up to. Built on BLS survey data.
13. There are only twelve 4x4 sudokus (and a cool trick for finding minimal subsets) (baldino.dev)
Francesco Baldino
14. What we have learned at OpenShell applying formal methods to control AI agents (nvidia.github.io)
An intro to using formal methods to reason about permission changes in long-running AI agents.
15. AI 'kill switch' may need to be mandatory, Anthropic co-founder tells BBC (www.bbc.com)
US President Donald Trump dismisses calls to slow the development of AI, saying that claims of it "taking over the World" are a "HOAX".
16. When Google Cuts Off Access: Poland and the World (dossier.reasoner.pl)
Before you use Gmail to sign in anywhere, read this and think 10 times.
18. Closing the IPv6 First-Packet Gap with Grand (labs.ripe.net)
There is a subtle asymmetry in IPv6 Neighbour Discovery: a host knows how to reach its router before the router necessarily knows how to reach the host. GRAND fixes that asymmetry by making the host proactively advertise its address, and I implemented that mechanism in FreeBSD.
19. Synthesizing Loop-Free Programs with Rust and Z3 (2020) (fitzgen.com)
This post makes extensive use of math symbols, and uses MathJax.js to render them, therefore I recommend enabling JavaScript.
20. Photographs of Atlantic City Sand Sculpture (ca. 1880–1920) (publicdomainreview.org)
21. OrangePi Zero 3W Review: Tiny yet Powerful (boilingsteam.com)
OrangePi Zero 3W Review: Tiny Yet Powerful
22. Show HN: Warp – Run DeepSeek v4.1 Flash with 5 GB of RAM at 3.77 tok/s (github.com)
Run the full 2.78-trillion-parameter Kimi K3 model, DeepSeek V4.1 Flash or GLM-5.3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C infe...
23. GRP-Obliteration: Unaligning LLMs with a Single Unlabeled Prompt (arxiv.org)
Safety alignment is only as robust as its weakest failure mode. Despite extensive work on safety post-training, it has been shown that models can be readily unaligned through post-deployment fine-tuning. However, these methods often require extensive data curation and degrade model utility.
In this work, we extend the practical limits of unalignment by introducing GRP-Obliteration (GRP-Oblit), a method that uses Group Relative Policy Optimization (GRPO) to directly remove safety constraints from target models. We show that a single unlabeled prompt is sufficient to reliably unalign safety-aligned models while largely preserving their utility, and that GRP-Oblit achieves stronger unalignment on average than existing state-of-the-art techniques. Moreover, GRP-Oblit generalizes beyond language models and can also unalign diffusion-based image generation systems.
We evaluate GRP-Oblit on six utility benchmarks and five safety benchmarks across fifteen 7-20B parameter models, spanning instruct and reasoning models, as well as dense and MoE architectures. The evaluated model families include GPT-OSS, distilled DeepSeek, Gemma, Llama, Ministral, and Qwen.
In this work, we extend the practical limits of unalignment by introducing GRP-Obliteration (GRP-Oblit), a method that uses Group Relative Policy Optimization (GRPO) to directly remove safety constraints from target models. We show that a single unlabeled prompt is sufficient to reliably unalign safety-aligned models while largely preserving their utility, and that GRP-Oblit achieves stronger unalignment on average than existing state-of-the-art techniques. Moreover, GRP-Oblit generalizes beyond language models and can also unalign diffusion-based image generation systems.
We evaluate GRP-Oblit on six utility benchmarks and five safety benchmarks across fifteen 7-20B parameter models, spanning instruct and reasoning models, as well as dense and MoE architectures. The evaluated model families include GPT-OSS, distilled DeepSeek, Gemma, Llama, Ministral, and Qwen.
24. Rat and Mouse Gazette: Nursing Care (1996) (www.rmca.org)
25. Show HN: Go Bindings for SCIP Optimizer (github.com)
Go Bindings for SCIP. Contribute to egoisutolabs/scipgo development by creating an account on GitHub.
26. Suchir Balaji's Homepage (suchir.net)
27. LLMs are real, AI is fake (pluralistic.net)
28. How AI tool calling works (40 lines of vanilla JavaScript) (buttercup.sh)
Lesson 2 of the free buttercup.sh agent course: the twenty-line loop that turns a chat model into an agent, working JavaScript and Python, what each turn costs in tokens, and an exercise you run in the browser.
29. Leaving Linux (jackevans.bearblog.dev)
I've been a pretty die hard Linux user for the past decade or so, it was a huge
part of my identity. Owning a shitty beaten up thinkpad and budget Android
...
part of my identity. Owning a shitty beaten up thinkpad and budget Android
...