[Dev Catch Up #30] - Ramp up with these Cursor resources, Reflection 70B, 3000x inference speed and much more!
[Dev Catch Up #30] - Ramp up with these Cursor resources, Reflection 70B, 3000x inference speed and much more!
Bringing devs up to speed on the latest dev news from the trends including, a bunch of exciting developments and articles.
Welcome to the 30th edition of DevShorts, Dev Catch Up!
We write about developer stories and open source, partly from work and experience interacting with people all over the globe.
Before starting today’s issue, we are proud to announce that DevShorts have crossed 1500 subscribers and we will celebrate that through a swag drop!!🎉🎊
We are giving away 2 JBL Wave Beam earbuds 🥳
To participate in the giveaway, head over to our Twitter Page and follow the instructions.
Some recent issues from Dev Catch up:
LLM training at Meta, iOS 18 beta release, AI security with PCC
Federated Language Models, Speculative Decoding API, AI in Figma
Join 1500+ developers to hear stories from Open source and technology.
Must Read
Cursor AI brought a revolution in the world of efficient code writing with its AI-driven approach that lets you prompt an edit, ask a question, accept an autocomplete, and chat with your codebase. To further smoothen the process, there is cursor.directory, a library containing a collection of prompts providing the best configuration rules that helps the Cursor editor get a better understanding of the developer’s intentions and provide the above intelligent features. Check out Cursor.directory and try out Cursor AI from here.
A developer of any tier will have familiarity with the JavaScript language and developing applications with this language and its frameworks have the strongest suit because of its low entry barrier. Developing an application with the help of one of the most popular backend framework, NodeJs and later deploying it in AWS Lambda with help of OpenTofu and GitHub actions might seem complicated but is made easy by Meysam Azad, where he takes you on a tour through the steps in his detailed-explainable tutorial.
The AI boom is real and it is shaping the future of tech in a rapid manner. In one of our past issues we have talked about AI inference and have discussed how better inference improves the model in terms of making predictions, classifying data, deriving new insights from unseen data, etc. Latent Space has released a new podcast where they have talked about how achieving 3000x faster, cheaper, and better AI inference is not a dream but a reality with hardware improvements, quantization, and synthetic data distillations.
Now, we will head over to some of the news and articles that will be at a place of interest for developers and the tech community out there.
Good to know
GitHub is the one of the most popular places to host your code and millions of developers use it on a daily basis. It is the home of open-source developer projects and many big organizations use the platform to host their code privately. In GitHub, you can make changes in your code repository by pushing the new code into that repository and after pushing the code, a lot of interesting things take place that can be coined as push processing. GitHub recently improved it as there was something wrong in the process. The process involves an enormous job called the RepositoryPushJob which contains the push processing logic and due to its massive size and complexity, it leads to many problems. Learn more on how GitHub solved this problem and thereby improving the push processing from this article published by the GitHub Engineering team, where they put a deep explanation of the issue along with the solution.
Choosing the right database for your SaaS product is highly essential, given that it will have an impactful decision on the data processing performance of the application, lowering operational costs, reliability to handle the data load, etc. Both Neon and Supabase provide database solutions focusing on serverless postgres and managed postgres respectively. Here is an in-depth article covering configuration setup, architecture, data import options, real-time processing capabilities, etc., about the two platforms that will help developers choose the right platform for their project needs.
Open-Source technologies are developer’s paradise and what better way to promote them than spreading out the word for more developers to get involved with that tool or the project. Today we are focusing on Sparrow. It is a tool that allows efficient data extraction and processing from various documents and images. It can extract data from various unstructured data sources like forms, invoices, receipts, etc. With a pluggable architecture, you can easily integrate it with your system and run data extraction pipelines using other tools and frameworks. Sparrow also has agents with which you can build independent LLM agents and use API to invoke them from your system. Learn more about Sparrow from its official GitHub repository here and leave a star to support the project.
Lastly, we will take a look at some of the trending scoops that hold a special mention for the community.
Notable FYIs
Leveraging customer data with the help of LLMs can improve the customer experience, automate internal processes, create new contents, and access valuable information. Retrieval Augmented Generation or RAG, Supervised Fine Tuning or SFT, Prompt engineering, etc., are some of the methods that bring data to foundational models. Learn more about these methods from this article from Google Cloud, where they discussed the information with which you can get started with.
Matt Shumer recently announced Reflection 70B and claimed it to be the world’s top open-source LLM. The model is reportedly beating GPT-4o on every benchmark tested and holds its top among closed-source models like Claude 3.5 Sonnet. The model is built on Reflection-tuning technique that enables LLM to recognize and fix mistakes before answering. The demo of this model can be found here and the model weights at huggingface.
Preparing unstructured data for Generative AI usage is a tiring job. It involves a variety of document types and requires many different python packages making the job long and tedious. This article focuses on the challenges of preparing unstructured data for Generative AI and also discusses how Tonic Textual solves these problems with an easy streamlined solution.
Smoke test in Site Reliability Engineering is a quick preliminary test that ensures the proper functioning of critical components as per expectations after any deployment or changes. Learn how Incident.io is testing their products with the help of smoke tests from this article published by their engineering team.
Continuing with new LLMs, DeepSeek V2.5 is now officially launched and is available in web and API. It combines the strengths of DeepSeek-V2-0628 and DeepSeek-Coder-V2-0724 and helps in enhancing writing, following instructions, and human preference alignment. This tweet from DeepSeek will help you have a better understanding of the model.
That’s it from us with this edition. We hope you are going away with a ton of new information. Lastly, share this newsletter with your colleagues and pals if you find it valuable and a subscription to the newsletter will be awesome if you are reading for the first time.