[Dev Catch Up #23] - Apple Intelligence, AI Agents with LLMs, Smart Paste from Google, and 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 23rd edition of DevShorts, Dev Catch Up!
I write about developer stories and open source, partly from my work and experience interacting with people all over the globe.
Some recent issues from Dev Catch up:
Join 1000+ developers to hear stories from Open source and technology.
Must Read
Apple recently concluded their annual developer conference and finally, they officially entered into the AI race with the introduction of Apple Intelligence. It announced major AI updates and updates based on its operating systems at the developer keynote. Apple will incorporate generative ai models at the core of their products which will enable new capabilities across Apple’s native applications like generating images or summarizing texts. Apple Intelligence will be free in Apple IPhone pros, ipads, and macs with an M1 and later chips. The conference also includes announcements from Apple on customization of iOS 18, RCS support in iPhone, a Password app to keep track of logins, and a new calculator app for iPadOS 18. Check out all the announcements and learn about the features from this article published by The Verge where they covered the whole conference in a nutshell and discussed all the key changes coming to Apple’s products and softwares.
AI is the future and as enterprises and organizations are already using it in their features and products, the focus on the technology’s security is essential to map a secure future with the technology. But security mechanisms are implemented in Large Language Models that are fine-tuned for safety and instruction following. It creates a stream that moderates the model's ability to refuse requests. While it focuses security in mind, preventing the model from representing this direction makes the model lose its ability to refuse requests. Artificial adding to the direction can cause the model to even refuse harmless requests. Hence, comes the concept of Abliteration. It is the technique that uncensor any large language model without retraining. It removes the model’s built-in refusal mechanism effectively and allows it to respond to all types of prompts. This article from Hugging Face discusses the concept of abliteration and gives a walkthrough to the readers on ways to implement it.
With the evolution of AI, Large Language Models or LLMs are becoming powerful and in the quest to make them powerful, arises the new breed of software commonly known as agents that enhances the power and capabilities of the LLMs. An AI agent is an autonomous software entity that leverages the language processing capabilities of LLMs to perform a wide range of tasks that are beyond text generation and comprehension. Agents attempt to overcome limitations that are faced by LLMs and rely heavily on the LLM to perform reasoning while augmenting an LLM’s functionality to add new capabilities. To learn more about an AI agent and how it adds capability to the overall functionality of an LLM, refer to this article published by the Newstack, where they dived deep into the issue.
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
Generative AI has been the hottest technology trend in recent years and it has spread its influence from big techs to startups. Implementing Large Language Models or LLMs along with Generative AI is very common nowadays and hence, the resiliency of this technology has come under scanner. Managing resiliency of this technology is hugely unexplored because it is hard to manage if the vendor providing the LLM API comes across a service disruption. To resolve this issue, developers using LLM in their application can switch between vendors using the fallback feature. Learn more about this feature from this article published by Portkey, where they have explained this feature in detail and have also explained how you can create a resilient generative AI application that can switch from GPT-4o to Gemini Flash using Portkey’s fallback feature.
AI has brought a whole new revolution in the field of developer productivity and several big tech companies are using AI to increase the developer productivity of their employees. Their AI-assisted tools help developers with code writing, reviewing, and other things. One such tool is Smart Paste from Google that predicts the next state of a code environment. It uses generative AI to create context-aware adjustment to pasted codes. During code revisions, this application streamlines the copy-pasting process. Internally while testing with their developers, they found that 6.9% of all code-pastes have utilized Smart Paste and 42.5% of the suggestions from the feature were accepted. This blog from the engineering team of Google discusses the Smart Paste application, its features and how the team developed it over time.
Celebrating new open-source tools and projects in our issues is a given and this one will be no exception to that. This time we are focusing on a conversational AI JavaScript library that provides a UI for large language models. The tool named NLUX makes it easy for integration of powerful large language models into web applications. It features react components and hooks, LLM adapters, streaming LLM output, and custom renderers. With highly customizable and zero dependencies, it can build high quality AI chat interfaces within minutes by applying a few lines of codes. Check out NLUX from its official GitHub repository here and leave a star to support the developers.
Lastly, we will take a look at some of the trending scoops that hold a special mention for the community.
Notable FYIs
The emergence of AI and its effect on building effective software applications is visible with the emergence of Retrieval Augmented Generation or RAG. It has revolutionized the way of interacting with data and has offered exceptional performances in similarity searches. But it faces a bunch of shortcomings when handling complex tasks like time-based queries or relational database queries. To resolve this problem, the use of SQL vector databases is essential because it can manage various query types essentially and handle complex relational queries. Learn more about enhancing a RAG application with SQL vector database from this article published by the Newstack, where they have built an application from scratch using the same.
While rummaging through the database management system, you might have come across sharding. Commonly known as data sharding, it is a database architecture pattern that is used to distribute a local dataset across multiple databases or servers. Here is a small article from ByteByteGo that explains the top 4 data sharding algorithms with an interactive diagram.
AI is slowly setting a lion’s share in the modern day tech world and training LLMs are a crucial part of that journey. Model training across large contexts help in perfect utilization of the model in getting accurate responses. Learn more about training an LLM across a million context window from this podcast hosted by Latent Space along with Mark Huang of Gradient.AI, where they have discussed scaling of Llama 3 across a million context window, the effective usage of GPT 4 for creation of synthetic data to help context extension finetunes, the difference between different Gradient.AI models, and more.
SQL or Structured Query Language is the most effective language for inspection of relational databases and it is a domain specific language for different platforms across the internet. Cross-referencing different aspects of the platforms can be achieved with joins that can be done with SQL-based pipelines resulting in focusing on specific application items with applying different commands. Learn more about SQL based pipelines from this article published by the Newstack, where they have given a deep dive on the technology by analyzing an open-source tool for the same.
If you have a knack of obtaining great learnings and incidents from the world of Kubernetes and Cloud-native by attending conferences, then KubeVirt Summit 2024 will be the go-to conference for you. Register to this conference from the official CNCF website to learn more Kubernetes and Cloud-native applications from the eyes of the experts with reference to live projects and demos.
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.