- AI Fire
- Posts
- 💪 Bigger Isn't Always Better in AI?
In partnership with
Plus: Kids create their own AI stories
Read time: 5 minutes
This year, OpenAI's DevDay focused on helping developers instead of competing with apps for users, showing a shift in strategy.
What are on FIRE 🔥
IN PARTNERSHIP WITH FREESPOKE
What We’re Searching With
Oh, the internet. It can be a scary place… but less so with Freespoke. Freespoke is a search engine that does a few important things differently:
News results show all sides with media biases labeled (left, middle, right).
Their new election portal delivers unbiased election coverage so you can make up your own mind.
No adult content by default – they’re a partner in protecting your families from bad content. Freespoke is the search platform that respects your privacy and doesn’t manipulate the information you find online.
AI MODEL
🤖 Liquid AI Launches Efficient New LFM Models
Source: Liquid AI
The first Liquid Foundation Model (LFM) is a new type of AI that works very well at all sizes, uses less memory, and runs more efficiently - built from first principles.
What is ‘first principles’: It means Liquid AI builds their models from basic, core ideas, instead of relying on common AI designs like transformers, that are widely used in AI today.
Keys takeaway:
The company launched its LFMs in three sizes: 1.3B, 3B, and 40B parameters. These models use a new design based on dynamical systems, not traditional transformers.
The LFMs perform better than models like Meta's Llama 3.2 and Microsoft's Phi-3.5 on key tests like MMLU.
They also need much less memory, especially for tasks with long texts, handling up to 32k tokens efficiently.
Currently, the models aren't open-source and can only be accessed through the company's Lambda (Chat UI and API) and on Perplexity AI.
Why it matters: Liquid AI's LFMs break away from the common transformer design used since 2017. They're made to be more flexible and adaptable, helping AI learn new tasks faster. Think of a self-driving car that improves with every mile. Plus, they use less computing power, making them more eco-friendly with a smaller carbon footprint.
PRESENTED BY 1440
Looking for unbiased, fact-based news? Join 1440 today.
Upgrade your news intake with 1440! Dive into a daily newsletter trusted by millions for its comprehensive, 5-minute snapshot of the world's happenings. We navigate through over 100 sources to bring you fact-based news on politics, business, and culture—minus the bias and absolutely free.
TODAY IN AI
AI HIGHLIGHTS
🤖 OpenAI's DevDay 2024 unveiled 4 key AI updates to make it more accessible and affordable for developers instead of major product launches.
👁️ OpenAI introduces vision fine-tuning for GPT-4o enabling developers to enhance image understanding capabilities with just 100 images for various applications
🗣️ OpenAI's CPO Kevin Weil envisions AI interaction like human communication with agentic systems hitting mainstream in 2025
💻 Microsoft's AI push drives data center finance leases to $108.4 billion up $100 billion in 2 years for AI workloads
💼 California Governor Gavin Newsom has just rejected S.B. 1047, an important AI safety bill. This bill aimed to place tougher rules on Silicon Valley AI companies and control the release of new models in the state.
DAILY AI FUNDRAISING
11x.ai, an AI sales bot developer, raised $50 million in Series B funding, led by Andreessen Horowitz, valuing the company at $350 million. This follows their earlier $24 million Series A round.
AI SOURCES FROM AI FIRE
AI SAVINGS OPPORTUNITIES
🔥 NewsletterAZ Course: Learn how to make money by writing AI-powered newsletters. Use code PRELAUNCH50 for 50$ OFF!
🔥 AI Mastery AZ Course: AI Tools at Work Course - Learn the basics of AI, find out the best tools for your needs, and see how to use them in real-life situations to boost your productivity at work. Use code PRELAUNCH30 for $30 OFF!
NEW EMPOWERED AI TOOLS
🧩 Wispr Flow is a Mac dictation app that writes faster and in your style.
↩️ Hubmee is a personal digital manager.
🐝 BeeDone is an AI productivity app game.
📈 Rows is your personal data analyst.
📞 Flow uses your voice to write 3x faster in every application
AI QUICK HITS
AI DISCUSS
People often believe that bigger AI models are better because they use more data, power, and parameters. To top it off, bigger models also have a far bigger carbon footprint, because they require more energy to run. Smaller models can work just as well, using less energy and being more eco-friendly.
Problems with Large Models:
They require massive resources and energy, contributing to a larger carbon footprint.
Data gathering for these models can be invasive, and some datasets even contain harmful content.
Large models are mostly controlled by a few powerful tech companies, limiting accessibility.
Small still good: The Allen Institute for AI (Ai2) created Molmo, a smaller, open-source model that performs as well as larger models, like OpenAI’s GPT-4o, but with fewer resources.
What do we need to do? Change the mindset: Researchers are challenging the "scale is all you need" belief, proving that smaller models can be just as effective. They focus on using smarter data collection and training methods.
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply