[Dev Catch Up #27] - Meta unveils Llama 3.1, Open AI's GPT-4o mini, Search-GPT, LLMs to generate Terraform code, 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 27th 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:
LLM training at Meta, iOS 18 beta release, AI security with PCC
Federated Language Models, Speculative Decoding API, AI in Figma
Join 1000+ developers to hear stories from Open source and technology.
Must Read
Nowadays, new Large Language Models are sprouting out every now and then with the aim of building surpassing one another in terms of capabilities and sustainability. In this race, big tech companies have a major chunk and they are using all of their resources in research and development of better models. Recently, Meta released their most capable and powerful open-source model Llama 3.1. It is the largest ever open-source model and it has outperformed Anthrophic’s Claude Sonnet 3.5 and OpenAI’s GPT-4o in multiple benchmarks. Meta has been using this model in their Llama assisted AI assistant in Whatsapp. Significantly smaller than Llama 3 released earlier this year, this model extends the context length to 128k and has 405 billion parameters trained with 16000 of Nvidia’s H100 GPUs. Learn more about the most capable model from this article published by the Meta team, where they have discussed extensively on the model features along with its competency in benchmarks.
AI is making its way into the search ecosystem with its expansion of capabilities. This is first witnessed with the introduction of the summary of search results feature released by Microsoft in their Bing browser, which is powered by their Copilot assistant. This feature is later replicated by Google and Apple which thumps the entry of AI in search capabilities on the web. Now, OpenAI has made its entry into this exclusive club with the introduction of SearchGPT. It is an AI-powered search engine with real-time access to information across the internet. This service is powered by OpenAI’s powerful and capable GPT family of models and is currently in the prototype phase. OpenAI is currently testing it out with developers with a plan to roll it out permanently in the future. This will pose a major competition to tech giants like Google and Microsoft, who enjoy a major share of the search market. The Verge has published this article that tells all about the recent developments of the product and you can join the waitlist from OpenAI’s official product page here.
Managing and provisioning of software infrastructure resources got a complete makeover with the introduction of Infrastructure as Code or IaC. This revolution has a blistering effect in increasing the reproducibility and scalability of the resources. Treating infrastructure as code has increased the efficiency and reliability of infrastructure management, as teams can now use the same practice and tools for software development to version, test, and deploy infrastructure. Like IaC, AI has brought an overall revolution in the productivity and development of modern applications. With AI-assisted tools used for different developmental workflows, LLMs have performed with top-notch capabilities and produced amazing results. The same LLMs can be used to generate IaC configurations with the use of Terraform, as Terraform is considered to be the industry standard for implementing IaC. The folks at Terrateam have penned down a wonderful article in which you will witness example workflows showcasing the use of LLMs to generate IaC configurations using Terraform.
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
OpenAI brought a jolt to the “Race to Build the most powerful LLM” with the introduction of its latest model GPT-4o. The “o” in the model name stands for omni and the model has improved audio, video, and text capabilities. It also has the ability to handle multiple languages at improved speed and quality. But the problem for a lot of developers is the expensive cost that comes along with this intelligent model. To tackle that, OpenAI recently launched the GPT-4o mini model that comes with better performance than that of GPT 3.5 turbo along with a huge cost efficiency. It is by far the most cost-efficient model of the company and the model has performed significantly well in multiple benchmarks. Learn more about this model from the article published by the OpenAI team, where they have talked about the model features in detail along with its pricing.
The growth of Artificial Intelligence is phenomenal in recent times and the development of Retrieval Augmented Generation (RAG) type AI application has assisted the evolving of applications using Large Language Models. RAG uses the strength of vector databases and LLMs to provide high quality results. This AI technique is largely used in developing customized AI applications that includes chatbots, recommendation systems, and personalized tools. While selecting the right LLM is crucial for developing the perfect RAG application, the cost factor of the LLMs provide a significant hindrance in the ambition. Hence, it is better to look at open-source LLMs and build RAG applications on top of them. The Newstack has recently published this well-detailed article in which it discusses the development of a cloud-hosted RAG application using an open-source LLM.
Today’s issue will also cover one of the best open-source projects that have generated quite the storm in the open-source world. It’s a no-brainer that this week, we will celebrate by picking one of them among the many. The open-source tool for this week is Patchwork. It automates the developmental gruntwork like PR reviews, bug fixings, documentation building, and security patching with the help of a self-hosted CLI and Large Language Models. The tool also features customizable LLM prompts that are optimized to do developmental chores such as code generation, library updates, issue analysis, and vulnerability remediations. Patchflows of developmental work with the tool involves LLM-assisted automations that combine steps and prompts. Steps is an important feature that shows the reusability of performing atomic actions like creating a PR or calling an LLM. Check out Patchwork from its GitHub page here and leave a star if you like it.
Lastly, we will take a look at some of the trending scoops that hold a special mention for the community.
Notable FYIs
Artificial Intelligence is taking technical research to great heights and the evolution of large language models are happening over time. Models are being trained using different methodologies aiming for the betterment of their performances. This podcast from Latent Space talks about the use of synthetic data in RHLF methodology for training LLMs along with the vision of Llama 4 model, scheduled to be released in the future, can open up the path towards open source AGI.
The Software Development LifeCycle (SDLC) contains a lot of frameworks that outline the development of a software in a semantic manner. Here are eight different types of frameworks presented in the form of this short article published by ByteByteGo that discusses their workflow in layman terms along with an interactive diagram showing the differences in the workflow.
We have covered OTel Community Days 2024 in our previous issue and the sessions in them are magnificent. Collecting information about the performance and behavior of software application by means of Profiling is an important aspect of Observability and this practice gets better with implementation of OpenTelemetry. Here is a 14 min YouTube video presented by Braydon Kains of Google in one of the sessions of the OTel community day 2024 where he demonstrated the power of Profiling by applying it to the OpenTelementry collector.
Microservices is one of the most popular system architectures in the market and if you are in the technical field or are familiar with system design, you might be familiar with its name or with its battle of dominance against Monolith services. Learn about the 9 essential components of microservices architecture in production from this short article published by ByteByteGo, where they explained each component in layman terms.
Interested in joining an event where the brilliance of AI will meet the complex technicalities of cloud-native engineering? Then, you must have a look at the Cloud-native + Kubernetes AI day 2024 North America. It is one of the co-located events of the KubeCon and the Cloud-native Con NA 2024. To attend this, you will need an all access pass and get one from the official event website 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.