AWS reinvent 2023

On 22nd November 2023, Generative AI was the focus at AWS's biggest annual cloud event, as well as news, keynotes, and innovation talks.

Details are available at https://aws.amazon.com/blogs/aws/top-announcements-of-aws-reinvent-2023/ 

AWS ECS idempotency

Amazon Elastic Container Service (Amazon ECS) now supports idempotency for task launches, allowing you to safely retry task launches without side effects. This feature helps ensure that timeouts or connection errors do not result in the launch of more instances than you originally intended, saving you time and money.

Idempotent operations allow requests to be retried with no additional side effects. You can now ensure that at most the desired number of tasks are launched as part of a RunTask API request by adding the following input to the request —client-token abcd. 

With an idempotent request, after requested tasks are successfully launched, subsequent retries using the same client token “abcd” within the same Amazon ECS cluster will not launch any additional tasks. When you use the AWS SDK or the AWS Management Console, a client token is automatically generated and added to RunTask requests (and used in any subsequent retries) if you do not pass one explicitly. 

Amazon ECS now supports idempotency for task launches on AWS Fargate, Amazon EC2 instances, and Amazon ECS Anywhere in all AWS regions. To get started with adding idempotency to RunTask API requests, see the ECS documentation at https://docs.aws.amazon.com/AmazonECS/latest/APIReference/ECS_Idempotency.html

AWS MSK replicator

Last week, AWS introduced MSK Replicator, a new capability of Amazon MSK that makes it easier to reliably set up cross-Region and same-Region replication between MSK clusters, scaling automatically to handle your workload.

Amazon Managed Streaming for Apache Kafka (Amazon MSK) provides a fully managed and highly available Apache Kafka service simplifying the way you process streaming data. When using Apache Kafka, a common architectural pattern is to replicate data from one cluster to another.

Cross-cluster replication is often used to implement business continuity and disaster recovery plans and increase application resilience across AWS Regions. Another use case, when building multi-Region applications, is to have copies of streaming data in multiple geographies stored closer to end consumers for lower latency access. It is possible to aggregate data from multiple clusters into one centralized cluster for analytics.

To address these needs, it is required to write custom code or install and manage open-source tools like MirrorMaker 2.0, available as part of Apache Kafka starting with version 2.4. However, these tools can be complex and time-consuming to set up for reliable replication, and require continuous monitoring and scaling.

Ref: https://www.infoq.com/news/2023/10/amazon-msk-replicator/ 

Amazon RDS Multi AZ

Amazon Relational Database Service (Amazon RDS) PostgreSQL Multi-AZ deployments with two readable standbys now supports major version upgrades. 

Starting this week, RDS PostgreSQL is possible to upgrade multi-AZ deployments with two readable standbys from major version 13.4 and above and 14.5 and above to 15.4 with just a few clicks on the AWS Management Console.

PostgreSQL 15.4 offers several performance and feature benefits that include SQL standard "MERGE" command for conditional SQL queries, performance improvements for both in-memory and disk-based sorting, and support for two-phase commit and row/column filtering for logical replication. 

PostgreSQL 15 release also adds support for new extension pg_walinspect, and server-side compression with Gzip, LZ4, or Zstandard (zstd) using pg_basebackup. 

Amazon Bedrock

Pinecone is now available (private preview) as a Knowledge Base for Amazon Bedrock, a fully managed service from Amazon Web Services (AWS) for building GenAI applications.

With this release, we can quickly and effortlessly integrate your enterprise data when building search and GenAI applications. This workflow is called Retrieval Augmented Generation (RAG), and with Pinecone, it aids in providing relevant, accurate, and fast responses from search or GenAI applications to end users. 

With Pinecone as a Knowledge Base for Amazon Bedrock, GenAI applications are ready to build with:

• Highly performant: Speed through data in milliseconds. Leverage metadata filters and sparse-dense index support for top-notch relevance, ensuring quick, accurate, and grounded results across diverse search tasks.
• Cost effective at scale: Start for free on the starter plan and seamlessly scale usage with transparent usage-based pricing. Add or remove resources to meet your desired capacity and performance, upwards of billions of embeddings.
• Enterprise ready: Launch, use, and scale your AI solution without needing to maintain infrastructure, monitor services, or troubleshoot algorithms. Pinecone meets the security and operational requirements of enterprises.

ImageBlind

Meta engineering team released a paper with the model named ImageBlind, which can bind 6 modalities. It outperforms prior specialist models trained for a particular modality.

It joins data from:

• - Images
• - Text
• - Audio
• - Depth
• - Thermal
• - IMU data

It uses large-scale vision-language models and extends its zero-shot capabilities to new modalities just by using their natural pairing with images, such as video audio and image-depth data, to learn a single joint embedding space.

The paper shows that not all combinations of paired data are required to train a joint embedding, but only image-paired data is sufficient to bind the modalities together.

Ref: https://lnkd.in/gXNndk63

Harvard AI/ML free course

 

6 free online courses by Harvard University, in ML, AI, and Data Science.

◼️ 𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐭𝐨 𝐀𝐫𝐭𝐢𝐟𝐢𝐜𝐢𝐚𝐥 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 𝐰𝐢𝐭𝐡 𝐏𝐲𝐭𝐡𝐨𝐧

Link: https://lnkd.in/gygaeAcY

◼️ 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞: 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠

Link: https://lnkd.in/gUNVYgGB

◼️ 𝐇𝐢𝐠𝐡-𝐝𝐢𝐦𝐞𝐧𝐬𝐢𝐨𝐧𝐚𝐥 𝐝𝐚𝐭𝐚 𝐚𝐧𝐚𝐥𝐲𝐬𝐢𝐬

Link: https://lnkd.in/gv9RV9Zc

◼️ 𝐒𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐬 𝐚𝐧𝐝 𝐑

Link: https://lnkd.in/gUY3jd8v

◼️ 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐟𝐨𝐫 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐏𝐫𝐨𝐟𝐞𝐬𝐬𝐢𝐨𝐧𝐚𝐥𝐬

Link: https://lnkd.in/g8gQ6N-H

◼️ 𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐭𝐨 𝐏𝐫𝐨𝐠𝐫𝐚𝐦𝐦𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐏𝐲𝐭𝐡𝐨𝐧

Link: https://lnkd.in/gAdyf6xR

From Introductory to Intermediate, great for beginners.

LangChain experimentation

Last month, LangChain experimentation was published at my company's blog at https://developer.trimblemaps.com/engineering-blog/experimenting-with-langchain-ai/

FACTOOL

This week, read an interesting research paper on "FACTOOL: Factuality Detection in Generative AI - A Tool Augmented Framework for Multi-Task and Multi-Domain."

FACTOOL is designed as a task and domain agnostic framework for detecting factual errors of text, which are generated by large language models like ChatGPT.

As part of this paper, the source code is released for FACTOOL associated with ChatGPT plugin interface at https://github.com/GAIR-NLP/factool

Happy AI coding!

Apple GPT

Today, it's a big day for Apple AI space.

Apple is developing artificial intelligence tools to challenge OpenAI, Google and others, according to a new report from Bloomberg’s Mark Gurman. The tech giant has created a chatbot that some engineers are internally referring to as “Apple GPT.” 

The report says Apple has built its own framework, codenamed “Ajax” to create large language models, which are AI-based systems that power offerings like Open AI’s ChatGPT and Google’s Bard. 

Apple has yet to determine a strategy for releasing the technology to consumers but is reportedly aiming to make a significant AI-related announcement next year.

Vector db usage

As generative Artificial Intelligence is surging in recent times, vector databases are in high demands in this era.  It is represented in the industry newsletter: TheAiEdge.io

The purpose of vector db is to index the data with vectors that relate to that data.  Hierarchical Navigable Small World (HNSW) is one of the most efficient ways to build indexes for vector databases.  It builds a similarity graph and traverse that graph to find the nodes, which are the closest to a query vector.

Counterfactual variants

Another interesting research paper on "Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks." 

Ref: https://arxiv.org/abs/2307.02477

It briefs the various language models to show impressive performance on a wide variety of tasks.  It covers transferrable reasoning skills by introducing counterfactual variants of familiar tasks. Some of use cases like drawing & music tasks, are attached.

It has been observed a consistent and substantial performance degradation on these counterfactual tasks, across LMs (GPT-4, GPT-3.5, Claude, PaLM-2). 

LLMs Survey

This week, read an interesting research paper on LLM - "A Survey of Large Language Models"

It reviewed the recent progress of Large Language Models (LLMs) with size larger than 10B.

This research covers 4 important aspects of LLMs

1. Pre-training
2. Adaptation tuning
3. Utilization
4. Evaluation

More details are available at

Paper ref: https://paperswithcode.com/paper/a-survey-of-large-language-models

Github code ref: https://github.com/RUCAIBox/LLMSurvey

AI Landscape

Artificial Intelligence is an emerging technology across the world. 

Recently, I came across Top-50 platform/frameworks from Twitter feed.  They are foundational aspects of Generative AI application landscape.  

Being software engineer, it is essential to be aware of these disturptive technologies.

LangChain

In a nutshell, LangChain connects the industry popular Large Language Models such as OpenAI and HuggingFace Hub, to few external sources like external files, Database, Google, Wikipedia and Wolfram. 

It provides abstractions (chains and agents) and tools (prompt templates, memory, document loaders, output parsers) 

The core objective of Langchain, is to orchestrate LLM workflow/pipeline by interfacing between text input and output. 

Ref: https://python.langchain.com/ 

Google Bard

In the current era of Artificial Intelligence (AI), Bard is Google's experimental, conversational, AI chat service. 

It is meant to function similarly to ChatGPT, with the biggest difference being that Google's service will pull its information from the web. Like most AI chatbots, Bard can code, answer math problems, and help with your writing needs.

Bard is currently available in 3 languages and over 180 countries and territories. It will gradually expand to more countries and territories in a way that is consistent with local regulations and AI principles.

Github CoPilot

CoPilot is a powerful term in recent days at Microsoft Windows 11 and GitHub world.

GitHub Copilot is an AI pair programmer that offers autocomplete-style suggestions as you code. You can receive suggestions from GitHub Copilot either by starting to write the code you want to use, or by writing a natural language comment describing what you want the code to do. 

GitHub Copilot analyzes the context in the file you are editing, as well as related files, and offers suggestions from within your text editor. 

GitHub Copilot is powered by OpenAI Codex, a new AI system created by OpenAI.

Microsoft Fabric

 

Last week, Microsoft CEO Satya announced Microsoft Fabric product during their annual tech conference - Build 2023!

Microsoft Fabric is a unified platform that can meet your organization's data and analytics needs. It is about bringing the data into AI era. In an essence, Fabric is a new analytics platform powered by AI.

It reshapes how everyone accesses, manages, and acts on data and insights by connecting every data source and analytics service together—on a single, AI-powered platform.

Video link: https://www.youtube.com/watch?v=X_c7gLfJz_Q 

Technology used to evolve in years, now in days. Quite challenging to catch up for an engineer!

ionic Capacitor

 

ionic Capacitor architecture is divided into 4 layers, namely

1. App layer - Shared code is built using one of the popular User Interface (UI) frameworks like Angular, React, etc.  
2. Ionic layer - This abstract layer consists of the common UI controls, which can be leveraged in application development.
3. Capacitor layer - It wraps the web application in WebView, which can display web apps inside the native app. It is an interface to inject the application into the native code of the underlying platform.
4. Platform layer - Capacitor’s shared code is portable to any platforms i.e. windows, web, android and iOS.

Windows Copilot

Yday, Microsoft announced in annual Build conference - "Windows Copilot ", which is first PC platform with AI assistance for customers.

Preview at https://youtu.be/FCfwc-NNo30 

xMaps GPT

xMaps is building a location intelligence platform with focus on data, algorithms, and artificial intelligence products.

Singapore based firm is building B2B SaaS platform, with data marketplace, API platform, and GPT-powered location engine to connect both in a dashboard as attached.

AI Era

 

As, I'm excited to the technology aspect of AI era, one of the inspiring Indian leader Mr. Narayana Moorthy (founder of Infosys) gave a different dimension to this platform during the recent interview with CNBC.

https://www.cnbc.com/video/2023/04/20/chatgpt-is-not-going-to-replace-anybody-n-r-narayana-murthy.html 

Few impressive quotes, by him:

• Can't beat the human mind
• Assistance, rather than replacement
• Human mind is the most powerful imagination.
• Typewriter vs Word processor - efficiency

Google Flutter

 

There are multiple frameworks available in the industry to develop cross platform applications.  During the recent business need, got an opportunity to work in 2 popular frameworks namely Flutter and Capacitor.  

Flutter is built by Google with the first release in May 2017.  Capacitor got released during the initial days of covid pandemic i.e., April 2020.  

Flutter architecture is divided into 3 layers namely Embedder, Engine and Framework. 

1. The base bottom layer “Embedder” generates the execution instructions for the specific device/operating system.  It instructs the underlying device to perform all types of processing.  On running Flutter app in any android device, Embedder generates the runtime instructions for android.
2. The middle “Engine” handles the core operations like network requests, input, output and rendering. It is completely written in C++
3. Top “Framework” layer is the foundation element of application development, which is achieved using DART language.  Framework has all the essential UI development elements like material parts, widgets, animations, gesture detectors, etc.

  

AI .NET

This week blog is quite interesting, which is in alignment with the industry trend "Artificial Intelligence."

As you know, OpenAI supports node & python in native way.  

It's the maiden attempt to build ChatGPT using .NET and so seeking your usage feedback.

• Step 1: Extract zip in your local folder (https://drive.google.com/file/d/11xyuX8ZMhj8pGlVKQsfaenXu9Z9mizdw/view?usp=sharing) 
• Step 2: Double click on FirstApp.exe
• Step 3: Enjoy your question/answer session with AI .NET engine

Why is Cross platform

Historically, platform & mobile software developments were isolated with different technology frameworks.  As an example, desktop software is built using Windows form, web apps are using web form. In terms of mobile app development, iOS is using objective C/swift whereas android uses java.  

Typically, a developer will need to learn and master each platform’s specific development language and software frameworks.  It drives the development/business need to allow one shared language that could be used across multiple platforms and mobile devices.  

As noted above, Microsoft’s .NET unified platform roadmap/strategy is one of the best testimonies, which was highlighted in their recent annual tech conference.  

In this alignment, Micorsoft’s Xamarin is uplifted to Multi-platform App UI (MAUI), which builds native, cross-platform desktop and mobile apps all in one framework.

What is Cross platform

In the last 5 decades, IT computing has evolved dramatically from Mainframe (COBOL) to Artificial Intelligence/Machine Language (OpenAI).

It's no longer just about installing hardware or software; beyond this ecosystem named "Mobile devices".

Wikipedia defines mobile devices as "a computer small enough to hold and operate in the hand, which typically have a flat LCD or OLED screen, a touchscreen interface, and digital or physical buttons"

In the last two decades, the world has seen an explosion of mobile devices like phones, tablets, and now wearables. It generates great opportunities for mobile hardware and most importantly software systems.

Cross platform refers to a single solution which fits across various platforms like windows, web and mobile. Cross mobile refers to multiple mobile devices like android, iOS.

Amazon Rekognition

Amazon Rekognition Content Moderation automates and streamlines your image and video moderation workflows using machine learning (ML), without requiring ML experience. 

It's a deep learning-based feature that can detect inappropriate, unwanted, or offensive images and videos, making it easier to find and remove such content at scale. 

Last week, Amazon Rekognition content moderation comes with an improved model for image moderation that significantly reduces false positive rates for e-commerce, social media and online communities' content, without reduction in detection rates for truly unsafe content. 

Lower false positive rates mean faster approvals for user uploaded content leading to a better end-user experience. Lower false positive rate also implies lower volumes of flagged images to be reviewed further, leading to a better experience for human moderators and more cost savings.