- WKND AI
- Posts
- Unlocking the Secrets of LLMs: Who's Leading the Pack?
Unlocking the Secrets of LLMs: Who's Leading the Pack?
OpenAI vs Google vs Meta vs The World
Josh Huilar
October 08, 2023
Hello my AI Weekend Warriors,
If you're like me, you've probably heard of ChatGPT. But did you know that's just the tip of the iceberg?
Why You Should Care: LLMs are not just a tech buzzword; they're the engines driving everything from your voice assistants to healthcare diagnostics. And guess what? The big tech giants are not just competing; they're also collaborating in ways you wouldn't believe!
Big Tech's Intriguing Dance: Microsoft is cozying up with OpenAI and Meta, while Amazon is throwing big bucks at Anthropic—a startup founded by ex-OpenAI folks. It's like a high-stakes game of chess, and we're here to give you a front-row seat.
What's Inside: We're diving deep into the LLM universe—from Microsoft's Bing AI to Google's Bard and beyond. Ever wondered what the CIA is cooking up in the AI department? We've got that covered too!
Today’s newsletter features:
📰 AI Viral News
🤿 AI Deep Dive
🛠️ AI Tools Of The Week
🎨 AI Images Of The Week
AI Viral News
📷 Remember Those Awkward Yearbook Photos?
Step aside, FaceApp! The internet is abuzz with a new sensation: the #aiYearbook Challenge. Powered by the EPIK AI Photo Editor app, this trend lets you relive your '90s high school glory days—or experience them for the first time.
All you need are 8-12 selfies, and voila! You get 60 AI-generated yearbook photos, complete with grunge-era flannel and butterfly clips.
🤩 Celebrity Seal Of Approval
From Hollywood A-listers to TikTok influencers, celebrities are fueling this trend's virality. Their '90s yearbook photos are flooding social media feeds, making us all nostalgic for a time when dial-up internet was a thing.
🔒 Data Privacy: The Elephant In The Room
As we chuckle at our AI-generated frosted tips, it's crucial to pause and consider data privacy. The app's parent company, SNOW Corporation, claims not to store personal data. But in the age of data breaches, how secure is your selfie, really?
🤔 The Celebrity Paradox
Interestingly, some celebs endorsing this app have previously threatened legal action against AI companies for using their likeness. It's a stark contrast that highlights the complex relationship between public figures and AI ethics.
How It Works 🤳
Download the EPIK AI Photo Editor app.
Upload 8-12 selfies.
Pay $4 to get your photos in 24 hours, or $6 in less than 2 hours.
Get 60 different '90s yearbook photos, complete with iconic hairstyles and outfits!
I even gave it a try myself. Can you guess which picture is my real high school year book picture?
AI Deep Dive
Large Language Models (LLMs) are advanced AI systems trained on massive datasets to understand and generate human-like text. Utilizing deep learning techniques, particularly transformer neural networks, LLMs excel in a variety of Natural Language Processing (NLP) tasks such as language translation, text classification, and sentiment analysis.
They bridge the gap between human communication and machine understanding, finding applications beyond tech, in fields like healthcare and science. However, their deployment raises ethical concerns, including biases and potential misuse, making responsible development crucial.
Check out the main LLMs below:
Microsoft Bing AI
Microsoft’s partnership with OpenAI
OpenAI ChatGPT
Microsoft’s partnership with Meta
Meta LLaMA 2
Amazon Alexa AI
Amazon’s partnership with Anthropic
Anthropic Claude (former employees of OpenAI)
Google BERT
Google BARD
Google PaLM 2
Google Gemini
Apple Siri 2.0
HuggingFace Falcon
Alibaba Qwen-VL
Baidu Ernie Bot
The CIA’s Top Secret Bot
Meet Bing Chat, Microsoft's revamped search engine featuring an AI chatbot that competes with ChatGPT…even though it uses ChatGPT’s LLM. Integrated with Bing's AI-powered search, it offers detailed, human-like responses and up-to-date information. It even supports creative tasks like writing and coding, and can generate images from text.
Microsoft's $13 billion investment in OpenAI could yield over $30 billion in new annual revenue, largely from Azure cloud services. The partnership makes Microsoft the exclusive provider of computing power for OpenAI's research and products. Despite the financial complexity and regulatory hurdles, the deal positions Microsoft as a major player in the AI landscape.
ChatGPT, OpenAI's language model, is developed using public internet data, licensed third-party info, and user-provided data. It learns by reading vast amounts of text, predicting the next likely word in a sentence. OpenAI ensures compliance with privacy laws and focuses on responsible data use.
Microsoft and Meta have joined forces to offer Meta's LLaMa 2 language model on Microsoft's Azure cloud platform. While not exclusive to Azure, the partnership designates Microsoft as the preferred cloud partner for LLaMa 2. The collaboration aims to democratize AI technology and expand the range of AI tools available to Azure users.
Meta's LLaMa 2 is a versatile language model with a scale ranging from 7B to 70B parameters. Optimized for dialogue and programming tasks, it outperforms most open-source models in benchmarks. Built on Google's transformer architecture, it incorporates advanced features like RMSNorm and SwiGLU. A must-know for AI professionals.
Meta faces internal and external pressures to accelerate the launch of LLaMa 3. With Falcon 180B setting new standards and OpenAI focusing on multimodal capabilities, Meta risks falling behind. The delay in LLaMa 3's development could impact the open-source community, as many smaller initiatives rely on LLaMa-based models.
Amazon's Alexa LLM is a game-changer in the commercial large language model arena. Designed for multimodal interactions, it processes voice, video, and text. It's extendable, connected to thousands of APIs, and offers a tailored user experience. A bold move, Alexa will even have opinions, making interactions more engaging.
Amazon invests up to $4 billion in AI startup Anthropic, intensifying its rivalry with tech giants like Microsoft and Google. The deal includes an initial $1.25 billion for a minority stake and an option to increase the investment. Anthropic gains a strategic partner in Amazon, which will provide compute power and cloud services, bolstering its ambitious AI projects.
Claude 2 by Anthropic aims to be a safer alternative to existing large language models. Designed with unique safety guardrails, it's already making waves as a potential "ChatGPT killer." The model powers Claude, an AI chatbot integrated into various products, focusing on AI safety without compromising functionality.
Google's BERT, standing for Bidirectional Encoder Representations from Transformers, revolutionized search algorithms by understanding context in queries. Developed before Bard, it's not a chatbot but a silent workhorse that refines Google Search results. Trained on Wikipedia and BookCorpus, it's a cornerstone in Google's AI ecosystem.
Google Bard, powered by Google's advanced LLM PaLM 2, elevates chatbot performance. Initially built on LaMDA, it now outpaces rivals like ChatGPT by supporting 40+ languages. A strategic move by Google to use its own LLMs, Bard is a noteworthy player in the multilingual AI chatbot arena.
Google's PaLM 2 is a leap in language models, trained on 100+ languages and boasting advanced domain knowledge. It's part of Google's Pathways AI architecture, aiming for multi-modal AI capabilities. From powering Google Docs features to potentially going offline on Android phones, PaLM 2 is Google's answer to GPT-4.
Google Gemini, unveiled at Google I/O 2023, aims to be a multimodal AI powerhouse. Designed for seamless API integrations, it's not just a single large language model but a group of them. Set to rival OpenAI, it will integrate into Google's existing products and be available via Google Cloud.
Apple's Siri 2.0 aims to reclaim its voice assistant throne with cutting-edge generative AI. Investing billions and recruiting top-tier engineers, Apple focuses on large-language models that rival OpenAI's. The challenge? Fitting these complex models into iPhones while maintaining user privacy. A game-changer for Siri and a must-watch for AI enthusiasts.
Meet Falcon 180B, Hugging Face's latest language model with a whopping 180 billion parameters. Trained on 3.5 trillion tokens, it's setting new standards in natural language tasks.
China is rapidly advancing in generative AI with relaxed regulations and active participation from tech giants like Alibaba’s “Qwen-VL” and Baidu’s “Ernie Bot”. The country aims to set global AI standards and is already home to half of all China-developed AI models.
The CIA is developing a ChatGPT-like AI tool to enhance intelligence access. Aimed at outpacing China in AI, this tool will serve multiple U.S. agencies and include a chat feature for real-time queries.
AI Tools Of The Week
Whether you're a professional or an AI enthusiast, these tools promise to increase your productivity 10x.
🛠️ Send us your favorite AI tools to be featured next week.
EPIK (Android/iOS): Take part in the viral “aiYearbook” challenge with EPIK. An app to edit, retouch and customize images.
Hummingbird: An AI personal assistant for macOS. Lightweight, convenient, and customizable.
AI Images Of The Week
Whether crafted by our team on Midjourney or submitted by readers like you, these are sure to inspire.
🎨 Send us your favorite AI images to be featured next week.
Image generated by Midjourney. Try out the prompt below.
intricate mood board, graffiti-inspired streetwear line by Urban Outfitters, bold saturated neon colors, thick lines and shading, photo knolling, lifestyle photos of models wearing streetwear fashion, product shots --c {0, 5, 10, 15}
Image generated by Midjourney. Try out the prompt below.
Candid asymmetric composition, medium shot, Star Wars, black stormtroopers in technologically advanced armor, myth punk, intricate detail, gorilla warfare, cinematic still --ar 16:9 --s 350 --c 20 --v 5.1Prompts courtesy of Rory Flynn on LinkedIn:
How'd you like this newsletter?Your feedback helps us make cooler emails for you! |
To support this newsletter, tell a friend to subscribe here!
