Show HN: Clop – optimise screenshots and recordings automatically https://ift.tt/MH1vwWY

Show HN: Clop – optimise screenshots and recordings automatically https://ift.tt/tLsfu59 August 12, 2023 at 01:41AM

Show HN: Pykoi – a Python library for LLM data collection and fine tuning https://ift.tt/sxvwXdP

Show HN: Pykoi – a Python library for LLM data collection and fine tuning Hi HN, pykoi is an open-source python library for ML scientists. pykoi makes it easier to collect data for LLMs, to use that data for finetuning, and to compare models to each other (e.g. your model pre- and post- finetuning, or your model vs openai vs claude). The library comes from pain points we experienced in LLM development: 1. Collecting feedback data from users isn't as easy as it could be. (The current process usually involves sharing excel files of annotated responses back-and-forth, offering no insight into how users actually engage with your models). 2. RLHF remains complicated to carry out. By complicated , we mean requires a lot of steps, hundreds of configs, lengthy setups, etc. 3. Comparing models to each other as they're used (that is, independent from academic metrics) is full of friction. The current approach: spin up a model, ask questions, write them down. Repeat for other models then compare. At a high-level, we think that the active learning process should be closed-loop: data collection, fine tuning, and inference all feed from the same system. This library is our first step in that direction. The project is still very early but we hope that some if it is useful. Note, we're fully open-source, and actively adding features! Website: https://ift.tt/uiOqMAd GitHub: https://ift.tt/ydxDImS We would love your feedback! https://ift.tt/bdMzJUL August 11, 2023 at 10:42PM

Show HN: Covert – Rewrite of HashiCorp Vault Using Rust, SQLite and Litestream https://ift.tt/bqfv2Lp

Show HN: Covert – Rewrite of HashiCorp Vault Using Rust, SQLite and Litestream https://ift.tt/ZGes9dC August 11, 2023 at 09:16PM

Show HN: A DIY airgapped, e-ink TOTP device https://ift.tt/veRdGbD

Show HN: A DIY airgapped, e-ink TOTP device https://ift.tt/fqs2dJl August 11, 2023 at 10:20PM

Show HN: Tetris, but the blocks are ARM instructions that execute in the browser https://ift.tt/lOBezas

Show HN: Tetris, but the blocks are ARM instructions that execute in the browser OFRAK Tetris is a project I started at work about two weeks ago. It's a web-based game that works on desktop and mobile. I made it for my company to bring to events like DEF CON, and to promote our binary analysis and patching framework called OFRAK. In the game, 32-bit, little-endian ARM assembly instructions fall, and you can modify the operands before executing them on a CPU emulator. There are two segments mapped – one for instructions, and one for data (though both have read, write, and execute permissions). Your score is a four byte signed integer stored at the virtual address pointed to by the R12 register, and the goal is to use the instructions that fall to make the score value in memory as high as possible. When it's game over, you can download your game as an ELF to relive the glory in GDB on your favorite ARM device. The CPU emulator is a version of Unicorn ( https://ift.tt/KP8WJQf ) that has been cross-compiled to WebAssembly ( https://ift.tt/wPmfdLq ), so everything on the page runs in the browser without the need for any complicated infrastructure on the back end. Since I've only been working on this for a short period of time leading up to its debut at DEF CON, there are still many more features I'd eventually like to implement. These include adding support for other ISAs besides ARM, adding an instruction reference manual, and lots of little cleanups, bug fixes, and adjustments. My highest score is 509,644,979, but my average is about 131,378. I look forward to feedback, bug reports, feature requests, and strategy discussions! https://ift.tt/m7czlj2 August 11, 2023 at 05:26AM

Show HN: Openform – use Google Forms and Google Sheets as a simple database https://ift.tt/Rsynk73

Show HN: Openform – use Google Forms and Google Sheets as a simple database https://ift.tt/j8rUbWm August 11, 2023 at 07:10AM

Show HN: VectorFlow – Open-Source Vector Embedding Pipeline https://ift.tt/5E0FJSM

Show HN: VectorFlow – Open-Source Vector Embedding Pipeline Hey HN We've just released the initial version of our open-source vector embedding pipeline. It's designed to embed large volumes of data. While embedding a few documents for Q&A is straightforward, consistently ingesting gigabytes of unstructured data is a whole different ballgame. By using our API, you can embed raw data and store vectors in your vector database, sidestepping the complexities of cloud infrastructure. Now, in true YC spirit, we're launching before everything is polished. Our Github repo ( https://ift.tt/J9XixH3 ) is a work in progress, and we're eager to get it in front of the community early on. Check out our Github repo and let us know what you think. Your feedback, suggestions, and critiques will be highly appreciated as we continue to refine and develop. https://ift.tt/J9XixH3 August 10, 2023 at 11:05PM

Show HN: Fully isolated honeypot SSH server using thrussh https://ift.tt/7y0ZFOa

Show HN: Fully isolated honeypot SSH server using thrussh https://ift.tt/BIHNUOu August 11, 2023 at 02:50AM

Show HN: Computer Socialism https://ift.tt/6ok5j3Z

Show HN: Computer Socialism https://chernikov.xyz/computer-socialism August 10, 2023 at 03:43PM

Show HN: Easiest way to convert PDF tables to Excel https://ift.tt/uoTLPRZ

Show HN: Easiest way to convert PDF tables to Excel We built a Windows app to convert PDF tables to Excel. You just have to take a screenshot of the table. Why try another app when there's so many PDF to Excel converters? It is: - Faster - More convenient - More accurate than other converters (including Excel's Power Query). Feel free to test it with complex tables (like the ones found in financial documents). https://table2xl.com August 10, 2023 at 06:05AM

Show HN: Remindme – A tool to set short-term reminders from the terminal https://ift.tt/PSZ690K

Show HN: Remindme – A tool to set short-term reminders from the terminal Hello there! The idea behind this CLI tool is to use it for short-term reminders, e.g. to remind you (via OS-specific notifications) to do something in 2 hours or in 10 minutes. The app is supposed to be cross-platform compatible. Still, I have limited access to platforms, so it was tested only on MacOS, Ubuntu (via VirtualBox), Fedora (via VirtualBox) and Windows 10 Home. I'd appreciate any feedback on how it works on these and other OS-es. Also, I have never tested it outside my time zone (GMT+2), so it would be nice to hear from folks from other parts of the world. Any feedback, comments, and questions are greatly appreciated, as it's just a tool built for fun. Here is a short intro on how to use it: Once installed (via Homebrew, APT, YUM or binary), the usage is pretty simple: - to run the app: remindme start - to set a reminder in 5 minutes: remindme in --min 5 --about "Rewatch LOTR editors cut" Use `--sec` and `--hr` flags to set reminders in seconds or hours, respectively. - to set a reminder at a specific time: remindme at --time "22:20" --about "Replay Gothic video games" Use `--am` or `--pm` flags with 12-hour A.M./P.M. format time value if you from this way of setting time. - to see the list of incoming reminders: remindme list - to stop the app: remindme stop Please, check the GitHub repo README.md file to see the full list of available commands. Thanks, and have fun! https://ift.tt/GPqvkAa August 10, 2023 at 12:56AM

Show HN: Miteiru, a Japanese Immersion ElectronJS Video Player https://ift.tt/SUw9aEv

Show HN: Miteiru, a Japanese Immersion ElectronJS Video Player Hi people! It's been about 9 months and I think it's ready for people to try this. This is a Japanese learning desktop video player that is cross-platform, easy to setup, and open source. You might have come across some similar app, but this app + OpenAI's Whisper is a bomb... I'll be maintaining this app and tryna achieve all CJK supported video player. https://ift.tt/GjX9Y4L August 10, 2023 at 02:50AM

Show HN: Automatically convert your GPT-3.5 prompt to Llama 2 https://ift.tt/E62eHrh

Show HN: Automatically convert your GPT-3.5 prompt to Llama 2 Hey HN! I'm working on OpenPipe, an open source prompt workshop. I wanted to share a feature we recently released: prompt translations. Prompt translations allow you to quickly convert a prompt between GPT 3.5, Llama 2, and Claude 1/2 compatible formats. The common case would be if you’re using GPT 3.5 in production and are interested in evaluating a Claude or Llama 2 model for your use case. Here's a screen recording to show how it works in our UI: https://twitter.com/OpenPipeLab/status/1687875354311180288 We’ve found a lot of our users are interested in evaluating Claude or Llama 2, but weren’t sure what changes they need to make to their prompts to get the best performance out of those models. Prompt translations make that easier. A bit more background: OpenPipe is an open-source prompt studio that lets you test your LLM prompts against scenarios from your real workloads. We currently support GPT 3.5/4, Claude 1/2, and Llama 2. The full codebase (including prompt translations) is available at https://ift.tt/TRMp8sX. If you’d prefer a managed experience, you can also sign up for our hosted version at at https://openpipe.ai/. Happy to answer any questions! August 10, 2023 at 01:47AM

Show HN: Automatic commit messages with AI copilot https://ift.tt/YkKOJLo

Show HN: Automatic commit messages with AI copilot https://ift.tt/Mfz4xQ2 August 10, 2023 at 12:52AM

Show HN: Kvak.io – Online note-taking app https://ift.tt/YzS3mw2

Show HN: Kvak.io – Online note-taking app Web-based note taking app with real-time colab. No accounts, barely any UI. Visiting the homepage generates a new note with a unique URL. https://kvak.io/hn August 9, 2023 at 01:25PM

Show HN: The Population Project https://ift.tt/Nf8r7LT

Show HN: The Population Project Two years ago, I turned 50. After a successful career as an entrepreneur, a business angel and a novelist, I set out to start a philanthropic venture under the following constraints: - it had to be global. - it had to be beautiful (in my eyes, at least). - it had to be technology and stats driven. I decided I would try to list the full name and date of birth of all humans alive. While some may find the concept pointless, I immediately knew I had struck gold: - it was global and incredibly hard. - it had an almost artistic quality to it, like an ever-changing installation. - as a libertarian, I resent that states conduct censuses and then sit on the data. - One billion people in the world aren't officially registered. At least someone would acknowledge their existence. I created a non-profit called The Population Project. I would never make a dime off it, but at least my costs would be tax-deductible. I then started researching lists of names online. I quickly adopted two principles. First I would collect a minimal set of information : full name, birth date, and birth place. Second, I would only scrape public information, i.e. nothing behind a password. After a few months, I realized I needed help from more experienced developers. I chose to work on 4D, a platform I had used in the past to develop my company's information system. It was a tough choice: 4D is not a leading player in the back-end world, but I figured the growth of API tooling would make language choice less critical. The first iteration of our database was frustrating - way too slow to publish a website. I learned the power of incremental change, with each marginal improvement saving you a few percent of speed or space. I also got to implement concepts I had heard about but never implemented, such as mirroring, partitioning, or hash-indexing. Then I hired a team of six data processors in Madagascar who clean up and process the lists found online. Lots of Python and Excel macros in their day-to-day. I have instilled in them an obsession with quality. A bad record will sit in our base forever. After trying dozens of softwares, we've settled on Adobe Acrobat and Octoparse. The final piece was the website. I lucked out in finding a strong team in Romania. They build with Next.js and deploy on Vercel. I gave them Wikipedia as the model to aim for. We/they haven't been able to match Wikipedia's simplicity. Our pages are too heavy. But I find the site user-friendly, pleasing to the eye and reasonably fast. We can and we will do better. A word about privacy. Some people complain that because it publishes names and DOBs, the Population Project infringes on their privacy. We obviously don't see it that way. - All our info is public. That DOB you find on the site is probably in the voter list of your state, a list that anyone can request or plainfully download. - The info we publish is minimal. Basically, we say that you exist. No one will find anything about your race, religion, sexual preferences, job or income. - We have adopted Wikipedia's privacy policy. We do not record your IP, unless you create or edit a record. - We're using Matomo for our Analytics. Great stuff. It's not free but they do not use your data like GA. Why am I telling you all this? From the beginning, I've envisioned a three-step process: 1) Build the database and populate it with millions of Western profiles. 2) Launch the site, where anybody can create or edit records and share them with their family. 3) When we've reached critical mass (1B records?), start making deals with NGOs and governments, and venture into other alphabets. We have just completed step 1. Step 2 is daunting as hell. I have grown a business but I have never grown a website. While I am ready to spend a bit of money on PR or SEO, I am not delusional: to reach the level of success we have in mind, we need this thing to go (somewhat) viral. How do you do that? https://ift.tt/CHoIdA3 August 7, 2023 at 09:16PM

Show HN: All-you-need OSS lib for modeling and working with ASTs in Python https://ift.tt/cYzBjIW

Show HN: All-you-need OSS lib for modeling and working with ASTs in Python https://ift.tt/6FML85i August 9, 2023 at 03:19AM

Show HN: Subscription-Based Alternative to Gumroad https://ift.tt/IKZaF8y

Show HN: Subscription-Based Alternative to Gumroad https://turbolink.io August 9, 2023 at 03:14AM

Show HN: Easyful – A Free Gumroad Alternative https://ift.tt/5NZPwqa

Show HN: Easyful – A Free Gumroad Alternative Hi HN, If you’re selling templates or digital assets online, platforms like Gumroad have a ton of amazing features . . . but they’re also expensive. It’s not uncommon to be paying 10%, 20% or even 30% of your revenue just to host and deliver some digital content to customers. Instead, we think most creators should own their own Stripe account and use a lightweight fulfillment layer to send customers their orders. So we built Easyful, a platform built on Stripe to email your content to customers when they buy it. And it’s free! We’ve been using Easyful ourselves for a few months now. Try it out and let us know what you think! https://ift.tt/g3pzMNK August 9, 2023 at 01:06AM

Show HN: Ggml.js – Serverless AI Inference on Browser with Web Assembly https://ift.tt/LXIGmuT

Show HN: Ggml.js – Serverless AI Inference on Browser with Web Assembly What is ggml.js? Run any ggml ported ML models directly on your web browser with ggml.js This project provides JavaScript bindings on ggml models, so that you can embed GGML models to your web apps to build serverless AI inferencing. Currently the framework supports following model types: - Dolly v2 - GPT2 - GPT J - GPT NEO X - MPT - Replit - StarCoder You can find more details about the framework: https://ift.tt/0GcMNeK For live demo examples, you can visit here: https://ift.tt/57cOKZD Source Code: https://ift.tt/5ezmIZo Please do provide any feedback/questions/suggestions in the comments. I'll be happy to take it up :) https://ift.tt/0GcMNeK August 9, 2023 at 12:50AM