Amit Bahree’s (useless?) insight!

In today’s rapidly evolving tech world, mastering Generative AI isn’t just an advantage—it’s a necessity. Are you ready to harness its power to transform your business and solve real-world challenges? I’m excited to announce that my new book, Generative AI in Action, is now available in print and ebook formats from Manning Publications . 📖 Special Launch Offer 🌟 As a thank-you to my early supporters, I’m offering an exclusive discount. Use the code pbbahree at checkout to receive 45% off your purchase of Generative AI in Action in all formats (valid through Sept. 30, 2024)! ...

AI is advancing at an unprecedented pace, with Mixture of Experts (MoE) models being one set of model architectures at the forefront of this revolution. These architectures enable breakthroughs in efficiency and scalability by leveraging a modular design where only a subset of specialized “expert” networks are activated for each input. MoE architectures have become a cornerstone in building ultra-large-scale models like GLaM and Switch Transformers. Mixture of Experts (MoE) is an advanced machine learning architecture that lately has gained significance, particularly in the realm of #LLMs (large language models) and NNs (neural networks). In talking with many people about AI, I’ve found that MoE as a topic comes up often, with many folks either not understanding what it is or their understanding of it being incorrect. ...

1. A little background I have been meaning to automate the deployment of my blog post to a dev server (running locally) for a while, but I haven’t had the time to get around to it until now. In addition to deploying this, the dev server also had several constraints. It is one of the machines at home and is not exposed directly to the internet. I also don’t have any ports opened on the firewalls at home, which adds a few more interesting challenges. I have a server hosted online in one of the data centers, but I want to keep that for more ‘production’ things. ...

The one thing I wanted to do after my book Generative AI in Action was complete was to create a summary in natural speech and possibly use TTS (Text-to-speech) to create an audio summary—think of it as a podcast that is easier for people to consume and get a quick sense of what the book is about. TTS (Text to Speech) or not to TTS? Initially, I was inclined towards using TTS (Text to Speech) for the audio summary. This technology, I thought, would be a convenient way to create a podcast-like summary that would be easier for people to consume and get a quick sense of what the book is about. My journey began with TTS - using GPT 4o to create a summary after ingesting the book and then using that as into the Azure AI Speech stack . However, I stumbled across something intriguing. Instead of TTS, I opted for NotebookLM from Google Labs to generate the audio overview - the podcast. This decision marked a significant shift in my approach, and I created two podcasts using the content from the book - one from multiple sources and another from a single source (the book). ...

I have been running a couple of instances of Teslamate - one locally on a server at home and another in Azure in a Ubuntu VM (see 👉 this blog post for details). I have been backing up the data to a NAS and then an offsite backup for the local instance. For the Azure instance, I have been running various backups during the day and backing up the data to OneDrive. This allows me to have a data backup in case the VM crashes or the data gets corrupted. ...

We released Phi-3 recently, which builds on Phi-2 ( read more on that here ) and it is a great model to use for various tasks. In this post, we will show how to run Phi-3 locally including a demo of it running on a phone. There should have been a video here but your browser does not seem to support it. 1. What are Small Language Models (SLMs)? Before diving into running Phi-2 locally, let’s take a moment to understand the concept of small language models (SLMs) and their significance in natural language processing (NLP). A SLM is a type of AI model that has been trained on a massive dataset of text but is limited in terms of its size and capabilities compared to a Large Language Model (LLM). SLMs are designed to be more lightweight and efficient, making them suitable for various applications, including chatbots, language translation, and content generation. SLMs are much smaller than LLMs, with fewer parameters and a smaller dataset, so they have a lower computational cost, making them more suitable for edge or resource-constraint devices. ...

1. What are Small Language Models (SLMs)? Before diving into running Phi-2 locally, let’s take a moment to understand the concept of small language models (SLMs) and their significance in natural language processing (NLP). A SLM is a type of AI model that has been trained on a massive dataset of text but is limited in terms of its size and capabilities compared to a Large Language Model (LLM). SLMs are designed to be more lightweight and efficient, making them suitable for various applications, including chatbots, language translation, and content generation. SLMs are much smaller than LLMs, with fewer parameters and a smaller dataset, so they have a lower computational cost, making them more suitable for edge or resource-constraint devices. ...

How many prompt engineers does it take to change a light bulb? Just one, but first, they need to fine-tune the model to make sure the AI doesn’t end up writing a poem about darkness instead. DALLE generated image of How many engineers it take to change a light bulb

🌐 As software continues to revolutionize the world, the advent of Generative AI is transforming the very fabric of software itself. My latest book, Generative AI in Action delves into this transformative journey. I am thrilled to announce the early release of my latest book, Generative AI in Action now available through Manning Early Access Program (MEAP) . This publication is a deep dive into the cutting-edge world of #GenerativeAI, #LLMs, #OpenAI, and #Azure #OpenAI, tailored specifically for enterprises. 🤘 ...

What does AI think of humans and our humor? I asked #AI - #CodeInterpreter plugin for #ChatGPT to express its experience as an AI working with humans as a meme. It wrote this code to create the meme below. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 import matplotlib.pyplot as plt import numpy as np # Create a new figure with a white background fig, ax = plt.subplots(figsize=(6, 6), dpi=80) fig.patch.set_facecolor('white') # Set up the text and meme caption meme_caption = "When I, an AI, try to understand human humor:" meme_text = "Error 404: Humor not found" # Add the meme caption and text to the figure ax.text(0.5, 0.7, meme_caption, horizontalalignment='center', verticalalignment='center', transform=ax.transAxes, color='black', fontsize=14, fontweight='bold') ax.text(0.5, 0.3, meme_text, horizontalalignment='center', verticalalignment='center', transform=ax.transAxes, color='black', fontsize=16, fontweight='bold') # Remove the axes for a clean meme look plt.axis('off') # Show the meme plt.show() AI's experience working with humans ...

What is Whisper from OpenAI? Whisper is a speech recognition model (ASR – automatic speech recognition) from OpenAI. The model itself is multi-task model and as a result in addition to speech recognition, can also do language identification and speech translation across a number of languages. The model is open sourced and it comes in 5 sizes. Of these, 4 have a english-only variant which seem to perform better if one only needs english. The model is also robust to noise, accents, background noise and technical language. Whisper achieves near SOTA performance with zero-shot translation from multiple-languages to English. ...

Bing is getting a new look and feel, powered by Microsoft AI and OpenAI (ChatGPT) and was announced yesterday. There is a lot of buzz around this, and I thought I would share my thoughts on this as I got access today. What is the new Bing? Well, it is the thing that is making the 800-pound gorilla in the room, Google, come out and dance on its toes. 🦍 The new Bing is an overhauled version of the search engine that uses ChatGPT technology to understand questions and generate answers. It runs on the next generation of OpenAI’s language model, which is significantly more capable than the version of ChatGPT that has been available since November 20221. The new Bing provides more relevant results for simple things like sports scores, stock prices and weather, along with a new sidebar that shows more comprehensive answers if you want them3. You can also chat and create with the new Bing, using its natural language and creative abilities4. The new Bing is live starting today, with limited capabilities. ...

API Introduction Polite Fork Off As A Service (or PFOaaS) - https://pfoaas.desigeek.com/ is a modern REST API that solves the problem of one telling people to politely fork off. 😇 There are days when we all need such a service for various reasons, and I think it is a great way to release some pent-up frustration. 🖤 It is also a great way to get some laughs too. This is of course meant for hard code engineers, writing RPC free code 😜 ...

In the last week or so, all the range online has been #OpenAI’s new chatbot called #ChatGPT (you can read more details on ChatGPT here ). This also got me thinking, about how can we use #CoPilot more than just code. GitHub CoPilot as you might recall is your #AI powered pair-programmer. And as we can see below, it indeed is possible to use Codex as sort of a more general purpose usage. I start with the prompts on how one might use CoPilot – a function to read a file and return its contents as a string, just to show there isn’t anything different I am doing in using this. And then for general-purpose usage, I used the prompts in VSCode. ...

OpenAI recently released #ChatGPT , a GPT-3 based chatbot that can be used to chat with. ChatGPT is a fine-tuned model of GPT3.5 , using #RL (specifically a PPO algorithm) similar to the Instruct series. This post is my experience in using it. Blog post with ChatGPT What better place to start with, than asking it about itself? 😃 ...

I had been thinking for a while to move away from WordPress for the blog to something simpler and cleaner. WordPress has been great for me when I first moved to it from another engine. However, over time, I found that things have gotten slower, as I added themes and add-ins. Some of these have been great, and others are not really needed. I also wanted to dogfood some of the things we built at work, though in my case that wasn’t the primary motive, just another nice to have. 😄 ...

Here is an example of how one can use a text prompt to generate a series of frames, that then are stitched together into a video. The prompt I used was: “a man walking in the parking lot with a miniature poodle”. the final video generated is shown below. There should have been a video here but your browser does not seem to support it. AI-generated video from a text prompt of a man walking in a parking lot with a miniature poodle ...

I have said this before - with the advent of large AI models, Prompt Engineering is critical and is the next challenge for us to master. What is Prompt engineering? Prompt engineering is the process of fine-tuning large models and often is written in natural language, outlining the intention of the user. Prompt engineering is a key element that allows the output to be accurate and reflect the needs of the user. Prompts should not be thought as the explicit one input to the model, instead are multiple tasks for the model. ...

I recently needed to install Visual Studio 2022 on one my existing machines to debug a new zeroshot model that has a dependency on our Speech SDK. The Speech SDK is one of our key #AI services in Cognitive Services (as part of #AzureAI). I already had VSCode running, but in this case I need the bigger brother. After installing Visual Studio, I could not get any nuget packages to install; I could not even fetch anything and didn’t matter what I used - the package manager console in Visual Studio, PowerShell, etc. ...

I recently installed Ubuntu on one of the Pi’s are home and installed Podman - which I hadn’t heard of until recently and is a container engine, similar to docker but doesn’t have a daemon. When trying to get a basic alpine test image running I got this error: Error: error creating build container: short-name "python:3.7-alpine" did not resolve to an alias and no unqualified-search registries are defined in "/etc/containers/registries.conf" ...

The mysterious life of developers has evaded many of us, until now … There should have been a video here but your browser does not seem to support it. The mysterious life of a developer (courtesy Spoon Norge)

If you have a Tesla, then you should absolutely check out TeslaMate which is data logger for your car(s) that one self-hosts. This uses the car’s API and gets all different kinds of telemetry of your drives, charging, batter conditions, acceleration, braking, parking, etc. I personally prefer this, over other online services (of which there are a few) - as it is giving away the keys to the kingdom - literally in this case (the Tokens used to authenticate and login). ...

It is 2021. And we have #AI writing #AI code. 🤪 It is quite interesting, but also can be quite boring once you get beyond the initial technology, and just think of it as one of the tools in your arsenal. And getting to that point is a good think. As part of a think at work I recently started playing with GitHub Copilot , which is using GPT3 to be your pair programmer – helping write code. GPT3 has multiple models (called engines), and Copilot uses one of these family of engines called Codex. Codex is a derivative of the base GPT3 engine that is trained on billions of lines of code. ...

Reinforcement Learning is teaching by example – it is how most of us learn. Reinforcement Learning (#RL) is a different approach to ML – it is a set of techniques that allows AI algorithms to experiment and learn from experience. RL falls in between supervised and unsupervised learning – there isn’t any labeled data, but at the same time it isn’t unsupervised either. At its most simple form, RL is a computational approach for automating goal-oriented decision making and learning. ...

I have been kicking the tires with Open AI’s #GPT-3 . Based on the screenshot below, it might be easy to think “oh boy does the model think highly of itself”, but as with most things in life - the devil is in the details.😃 The screenshot below was a forked version of davinci engine and follows the Q&A structure. ...