- Everyday AI
- Posts
- The big secret about small language models 🤫
The big secret about small language models 🤫
🤖 LLMs vs SLMs, OpenAI challenges Google in web search, Stability's new AI model, and more!
Everyday AI
February 15, 2024
Outsmart The Future
Today in Everyday AI
6 minute read
🎙 Daily Podcast Episode: Everyone seems to be making the push from large language models to small language models. But what are they? Here’s what you should know. Give it a listen.
🕵️♂️ Fresh Finds: Turning your leads into AI summaries, Lambda hits $1.5B, and how a historian uses AI. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: OpenAI challenges Google in web search, Google’s Goose AI coding tool, and Stability AI’s new AI model. For that and more, read on for Byte Sized News.
🚀 AI In 5: we stumbled upon a custom GPT that’s perfect for verifying emails for cold outreach. See it here
🧠 Learn & Leveraging AI: Confused about small language models? Here’s what to know and how they differ from LLMs. Keep reading for that!
↩️ Don’t miss out: Did you miss our last newsletter? We talked about AI and music creation, State-Backed Hackers using OpenAI, and ChatGPT getting a new memory. Check it here!
Small Language Models: What they are and do we need them? 🤔
It seems like we just started to understand large language models.
LLMs are extremely powerful and can do just about anything.
Buuuuut there’s a new kid on the block.
Enter small language models.
All the talk now is about small language models. So what are they and how do they compare to LLMs?
We explain small language models and their future.
Join the conversation and ask Jordan questions on small language models here.
Also on the pod today:
• Advantages and Usage of Small Language Models ✅
• Comparison of Small and Large Language Models 🥊
• Future of Small Language Models 💭
It’ll be worth your 37 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Friday, February 16th at 7:30 am CST ⬇️
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – Intent turns your leads into AI summaries, AI Community Widget gives you a ChatGPT-powered search widget, and RDMC AI is our own AI digital marketing agency.
Trending in AI – Shares of smaller AI firms are starting to surge after NVIDIA announced its stakes in them.
Money in AI - Lambda, a cloud computing startup, has hit $1.5B in valuation.
AI in Society - Here’s how the US government is planning to use AI for enhanced security.
LLMs – API company Kong has launched an open-source AI gateway, making multi-LLM app building easier.
Future of Work – This historian is using AI to sort through his work.
1. OpenAI Takes on Google in Web Search 🤯
In a bold move, OpenAI is reportedly developing a web search product to rival Google, with possible backing from Microsoft's Bing. This strategic development not only escalates the competition between tech giants but also signals a potential shift in the search engine landscape. Could this new player shake up the industry and redefine how we navigate the digital realm?
2. Google’s Goose: AI-Powered Coding Assistant 🪿
Google's new coding assistant, Goose, powered by the advanced AI model Gemini, is set to revolutionize coding processes at the tech giant. Google is launching this tool internally for its employees. By integrating Gemini's AI capabilities, Google aims to accelerate product development and streamline engineering tasks like never before.
3. Stability AI's New Model 'Stable Cascade' 🖼
Stability AI's latest creation, 'Stable Cascade,' is shaking things up in the world of image generation, promising faster speeds and higher quality outputs. This new model, not just one but three models in one, is changing the game with its innovative Würstchen architecture, compressing requests for quicker and more efficient processing.
4. Apple's Keyframer AI Tool Revolutionizes Animation Creation 🍎
Apple's Keyframer AI tool, powered by OpenAI's GPT4 model, allows users to generate CSS code to animate 2D images based on text prompts. This innovative technology simplifies the animation process without requiring coding skills, making it a game-changer in the creative industry. Professional motion designers see it as a valuable addition to their toolbox.
5. Microsoft's 3.2 Billion Euro Bet on Germany's AI Scene 🇩🇪
Microsoft is going all in on Germany with a jaw-dropping 3.2 billion euros! Microsoft plans to double AI capacity and amp up training programs, solidifying Germany's spot as an AI powerhouse in Europe. With tech giants like TSMC and Intel also eyeing Germany, the competition is heating up!
BounceBan Custom GPT Review
Need to send an important email, but not sure if the email is correct?
Or, maybe you're doing some cold outreach and don't wanna shell out some big bucks for a paid email verification service.
Well, we stumbled on a custom GPT that's ACTUALLY useful.
Check out today's AI in 5.
Or check out this related video:
🦾How You Can Leverage:
Now that we’re all JUST getting used to large language models, let’s throw a monkey wrench in our GenAI education.
(Monkey wrench incoming.)
Small language models deserve our attention.
Here’s the reality — LLMs are amazing.
They can do just about anything.
Their big neural network brains hold about the history of humankind.
Their wizardry is actually dumbfoundingly powerful.
But hold on a sec, what if I told you that small language models, with their fewer parameters and more specific task focus, could be the unsung heroes of AI?
You heard that right, shorties.
These tiny tech titans can’t be overlooked.
These compact yet efficient models are making waves in the world of AI applications, offering speed, efficiency, and a balance between performance and resource usage that's hard to ignore.
Let's dive into the world of small language models and explore their potential in shaping the future of AI technology.
Get your monkey wrench in hand and let’s get to work. 🛠️
1 – LLMs: Power with a Price 🧑🎨️
While Large Language Models (LLMs) are powerful, they come with significant drawbacks. They can be tricky for beginners to navigate and are costlier to implement on a large scale.
Like, crazy expensive. 😬
And just because LLMs can do everything, doesn’t mean you should use them for everything.
Like, pretty sure a Swiss Army Knife has a screwdriver, but we’re gonna grab a drill for assembling that 67-piece Ikea desk with no directions.
Worse yet?
The copy-and-paste Prompt Plague. When these 19-year-old Billy Boys share their ‘game changing’ prompts and try to sell you stuff, you think that a large language model will turn those prompts into gold.
But, LLMs don’t work like that. Because they know EVERYTHING, you have to be a bit patient and go through prompt engineering basics to improve your results.
Try this:
Speaking of that, Tera shouted out our free Prime, Prompt, Polish course on today’s show.
So, reply PPP to this email, and we’ll send you registration for our freshly updated course if you wanna stay in the Large leagues of LLMs.
2 – SLMs are more secure and speedy 🎯
Fast and secure?
Small models got our attention, shorties!
These small models offer enhanced security and speed by operating locally on your device. Plus, you have the freedom to download them for seamless use.
Try this:
From Microsoft’s Phi-2 to Meta’s Code Llama and Gemini Nano, small models are getting BIG traction.
Here’s a good fact-based recap from Microsoft on the rise of small models.
3 – The future may be small 🧍
See a trend recently in language models?
In the past few months, we’ve seen Tech Titans like Microsoft (Phi-2), NVIDIA (Chat With RTX), Meta (Llama 2 and Code Llama) and Google (Gemini Nano) all go big on small models.
Trend or the future? 🤔
We’d say probably the latter.
With Google Ultra Nano on the Samsung S24 phone, NVIDIA's Chat With RTX, Microsoft's AI laptops, and the upcoming Apple announcements (like Edge AI or on-device), it's clear that the spotlight is shifting towards compact and efficient Small Language Models.
Try this:
Want to get your feet wet in the SLM space?
Here’s a brief guide:
1. Hugging Face — probably the best place to download open source small language models, like Mixtral 8x7B, Meta's Llama 2, Microsoft's Phi-2 and more. Here’s the trending models you can download locally.
2. Hardware — To run these models on your local device, you’re gonna need some horsepower.
(*Insert horse sound here*)
Mac — If you’re on a Mac, you’ll want to be on a 2023 or later upgraded model with at least an M2 chip.
PC — Options here vary wildly, depending on your budget. But, the more RAM and powerful GPU, the better. Think, at least 32GB of RAM.
For best results, you should be looking at PCs released in the past year that have AT LEAST an Intel Core i7 chip or an equivalent.
To use the popular new Chat with RTX from NVIDIA, you’ll need a PC with a GeForce RTX 30 or 40 Series GPU.
Microsoft — The Microsoft Surface Laptop Studio 2 is a great option and it should be announcing new PCs this spring.
⌚
Numbers to watch
96%
LinkedIn researchers estimated that 96% of a software engineer’s current skills could eventually be replicated by AI.
Now This …
What do you think?
Vote to see live results
Are you excited for small language models? |
Reply