[Dev Catch Up #25] - Federated Language Models, Speculative Decoding API, AI in Figma, 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 25th 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
AI has undergone massive developments with the introduction of generative AI. Generative AI has brought a revolution with significant developments with the rise of Small Language Models or SMLs and Large Language Models or LLMs. The former runs on small devices while the latter runs in the cloud with better context lengths, multimodality, and complex reasonings. Keeping these trends in mind, came the idea of Federated Language Models or FLMs. It is an idea that takes advantage of SMLs and LLMs and enables enterprises to adhere to confidentiality, privacy, and security. Federated Language Models deal with two language models - one running at the edge, while the other one running in the cloud. An SLM running on the edge is primarily used for generation, while the LLM that runs on the cloud is leveraged for mapping the prompt into a set of tools and the associated parameters. Learn more about Federated Language Models from this article published by The NewStack, where more details on the technology along with its architecture, usage, and implementation have been discussed in detail.
With artificial intelligence on the rise, developers are keen on creating products and tools using different techniques based on AI. One of such techniques is speculative decoding. It is an advanced technique used in artificial intelligence to enhance the efficiency and performance of text generation models. Speculative decoding in terms of LLM inference enables parallelization of the token generation which enables users to speculate a large number of tokens in parallel and consume them without deviating from the provided content. Learn more about speculative decoding from this article published by Fireworks.ai, where they explained in detail about the technology and sheds light on how Cursor, one of their clients enabled by Fireworks inference stack, built Fast Apply and achieved 1000 tokens per second using the Speculative Decoding API with low frequency.
AI is spreading its wings in the current tech world and organizations are not shying away from pumping resources to build products and features on top of this euphoric technology. Recently, Figma announced major changes and a big redesign with AI. With a major UI redesign, new generative AI tools, and a built-in slide show functionality, Figma is all set to start the new AI beginning. The new generative AI tools allow users to quickly get started with a design, while the major UI revamp involves inclusion of a toolbar, rounded corners, and new icons. The developers also brought Figma Slides, a feature similar to Google Slides, that lets users build and share presentations within the app. It has interactive features for audience members and allows designers to present app-design prototypes directly from the deck. Learn more about all the big developments in Figma from this article written by The Verge, where they have given a detailed in-depth explanation of the new features and developments.
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 along with LLMs has done a complete transformation of the Software Development Lifecycle (SDLC). It is reshaping developer productivity and has been redefining SDLC with code generation and other features. Machine learning has already started gaining leaps with code-quality improvements, observability into code, build-time issue resolution, etc. With the rise of generative AI and models like GPT and others, we can see the developments in the field of code interpretation, improved code generation, enhanced problem solving capabilities, and much more. This article from Middleware dives deep into the discussion with the impact that generative AI has produced within developer productivity.
Speaking of developer productivity brings attention to a new product release of one of the most popular and favourite developer developer productivity tools in the market. Anyone associated with development and development-related activities have used or heard about Notion. Notion has introduced a new tool named Notion Sites that turns Notion pages into publicly-facing websites. It has the ability to set custom favicons, gives you the ability to build navigation bars with links. Also, it involves the ability to build breadcrumb navigation. Notion Sites can be easily published under custom domains and has basic SEO features with a Google Analytics integration. Learn more about Notion Sites from this article published by TechCrunch where they talked about the tool in detail.
Open-Source projects, tools, and technologies contribute largely in the software development ecosystem. They act as a stepping stone for young developers or someone who is getting started in the software engineering arena. One of such open-source projects created by Dan Richelson is Dorkly. It is a feature flag system for LaunchDarkly’s open-source SDKs. With it, developers can implement feature flagging consistently across dozens of languages. It is designed to fit into an existing workflow and aims to be a simple feature flagging system without the cognitive load of yet another tool. Have a look at Dorkly from its GitHub page here and leave a star to support it.
Lastly, we will take a look at some of the trending scoops that hold a special mention for the community.
Notable FYIs
Advancements in the field of AI stands strong with the incoming developments happening every week. Training large models on the largest clusters is not an easy task and this podcast from the Latent Space sheds light on the process of training a model greater than 70 billion parameters on 10,000 H100 clusters. It also discusses the challenges faced during the process and the learnings learned in the meantime.
Optimizing log management is essential for organizations to improve the processes and tools used to collect, store, analyze, and manage log data. This enhances the system performance, security, and troubleshooting efficiency. Here is a YouTube video demonstrating the talk of Eugene Tolbakov on Optimizing log management with AWS OpenSearch at the ChaseUK Conference.
As AI is getting stronger, the ability to generate video, audio, and images from simple text prompts that are indistinguishable from human-generated content is gaining on a rapid scale. But amidst all these, there is a huge gap between model capabilities and systems that are effective in practice which is because of HCI. Learn how HCI is the bottleneck in a world with strong AI from this short article written by Josh Pollock.
If you are familiar with Database Management Systems, you might have heard about locks. Locks are mechanisms that prevent concurrent access to data to ensure data integrity and consistency. Here is a short article from ByteByteGo covering the 9 common types of locks used in databases.
Prometheus is one of the best tools available in the market for monitoring and observability. Are you a Prometheus expert and want to share your valuable opinion on the tool and technology? PromCon EU 2024 is the perfect conference for you to attend. CFP is still open for this conference and you can submit your talk from the official PromCon site here. Alternatively, book your tickets as an attendee from here.
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.