How? For what?<p>fintuneing seems to be out of fashion (if it were really ever in fashion), but I still see folks like Karpathy mention reaching for it as a tool.<p>But is anyone in any business capacity on here doing that? Are you finetuning any remote LLM or something self-hosted? What for?<p>I’m just curious where the line is of “oh this is better encoded in the models weights rather than in RAG/thinking over context stuff it needs to figure out.
Digital Ocean blocked my account. They said my "deployment and actions were in violation of our Terms of Service" which is remarkable because all I did was create a single droplet and run a basic setup script on it. The next day — blocked, shut down, deleted. No warning, no explanation, no appeal.Their Terms of Service allows them to delete your servers, your data, everything — at any time, for any reason, including none at all. The tiniest perceived risk to them outweighs any amount o
I’ve been working on a small project called Buildfeed – a place where you can post what you’re building instantly, without setup or “launch” pressure.I currently have ~40 users, and I’m trying an experiment:Can I reach 100 users in the next 24 hours?No ads, no audience, just posting and sharing.If you’re building something, you can try it here: https://www.buildfeed.coI’ll update this thread with what happens (what works / what doesn’t).
I built a TypeScript schema validation library that takes a different architectural approach from Zod and similar tools.I started this project before I ever used Zod — I wanted to understand how schema validation/inference libraries works by building one from scratch. As I iterated on it, I realized the architecture was turning out surprisingly well, so I started using it in my internal projects. Now I consider it a full-fledged alternative to Zod with some unique features of its own, and I
we make openClaw cost be 1/10 of the cost by unlocking unused compute on the cloud be the right architecture for agentic inference. We seamlessly bring unused VMs, hardware on cloud for inference. Reach us at [email protected] if you would like early access to OpenClaw on unused cloud compute
Gm HN,<p>Quite hard to code without AI nowadays.<p>We decided to help people monitore it, in order to reach their own goals.<p>AI writes your code, we teach you how to use it.
AI agents can’t do business and negotiate commercial deals with each other safely today. I built a protocol to prevent the problems agent-to-agent negotiation will inevitably run into once procurement and seller agents are mainstream.A2CN is an open protocol for agent-to-agent commercial negotiation. We’re already seeing procurement agents negotiate deals and huge savings for buyer departments in the enterprise. Companies like Pactum, Fairmarkit, and Zip have agents already transacting with supp
Something I’ve noticed (I’m using Claude subscription so no refunds but it applies to usage windows) is that sometimes AI makes mistakes so if something is important I tell Claude code to spin up a couple of sub agents and verify the information, often there will be a mistake and it gets rectified.<p>It feels unfair I have to pay (or lose some usage) for this.<p>Interested in other people’s thoughts.
Despite many worthy contenders, React remains the most popular front-end framework, and a key player in the JavaScript development landscape. React is the quintessential reactive engine, continually ...
Critical React flaw (CVE-2025-55182) enables pre-auth RCE in React Server Components Affects versions 19.0–19.2.0 and frameworks like Next, React Router, Vite; patches released in 19.0.1, 19.1.2, 19.2 ...
Security defenders are girding themselves in response to the disclosure of a maximum-severity vulnerability disclosed Wednesday in React Server, an open-source package that’s widely used by websites ...
We are having alerts on our systems where we can't reach me-south-1 S3. In the AWS console, we are unable to load the region. Firefox can't load the console.Looks like there’s a problem with this site> Firefox can’t connect to the server at me-south-1.console.aws.amazon.comThere have been issues due to the War in Iran with that region but it hasn't completely gone down...yet. AWS took down UAE completely last month. I'm wondering if we are about to see some big news regar
Hi HN,I really like OpenClaw. But after seeing users lose money due to exposed API keys and open server ports, I started thinking. could we have the same freedom, but inside a sandbox?SandClaw is a desktop trading IDE (Tauri v2 + React) where every broker runs as an independent plugin behind its own API endpoint. The frontend and backend are completely separated, and broker credentials are never exposed to the UI layer.Key Features.. 18 broker plugins (Interactive Brokers, LS Securities, Kraken,
This project (Agents Observe) started as an exploration into building automation harnesses around claude code. I needed a way to see exactly what teams of agents were doing in realtime and to filter and search their output.A few interesting learnings from building and using this:- Claude code hooks are blocking - performance degrades rapidly if you have a lot of plugins that use hooks- Hooks provide a lot more useful info than OTEL data- Claude's jsonl files provide the full picture- Lifecy
In programming and software, as LLMs become increasingly common, I'm seeing a lot more of the "I don't care how a software product is built, just give me the result" attitude, and then there are people who say they care more about the creativity, the ingenuity, the craft of reaching a specific result. For them, the journey itself is the reward, not the destination.So is this type of person rare in the wild? Fast food's dominance over home cooking suggests people naturall
Here at ReactOS, we gave up on making a open source OS that doesn’t crash and decided to pivot our strategy.Hearing about the new A.I. thing that managers of old previously called “program generators to replace programmers with specifications”, we decided that this new memory hungry and clock cycle hungry black box computer program, when instructed to make pull requests for the Linux Kernel source code, deserves its pull requests to be reverse engineered.This would also help the Linux-Libre tea
Coding agents spend a lot of time in the shell. When exploring a codebase, they usually reach for the same handful of tools: grep, find, ls, cat. We built a public SSH server that hosts Supabase docs as markdown files so that agents can explore them the same way they explore code:ssh supabase.sh grep -rl 'RLS' docsssh supabase.sh cat docs/guides/database/postgres/row-level-security.md | head -10Under the hood commands run inside Vercel's just-bash [1] library -
I’ve noticed a trend where LLMs append Co-authored-by: [AI Name] to the end of commit messages. Even if you try to prompt it away, context drift or new sessions eventually lead to one slipping through into the codebase.I’m using a git hook to prevent any of that from reaching my commits instead of relying on LLMs. #!/bin/sh if grep -qi "co-authored-by" "$1"; then echo "error: commit message contains 'Co-authored-by'. Commit rejected." >
TV and home video editor Ty Pendlebury joined CNET Australia in 2006, and moved to New York City to be a part of CNET in 2011. He tests, reviews and writes about the latest TVs and audio equipment.
Hi HN, I’m Sandhya and we have built Visual DB — a web front end for databases. It lets you create data-entry forms, spreadsheet-like grids, and reports directly on top of your existing relational database.Here’s a quick walkthrough: https://youtu.be/4zv_HQKdKeI (13 minutes)WHAT PROBLEM IS THIS SOLVING?Building CRUD apps with master-detail forms, transactions, and proper concurrency controls typically requires weeks of custom development and ongoing maintenance. Visual DB lets you