⚡️ David vs Goliaths

Good Morning. In the race to build the most powerful AI, Microsoft's Research Lab has introduced a fresh twist with Phi-2, a surprisingly potent language model with a fraction of the parameters of its competitors. Let’s dive right in.

Today’s Highlights:

• UK's legal system embraces AI

• Nvidia's strategic move in AI chip diplomacy

• Snapchat+ introduces a creative AI twist

DEEP DIVE

Microsoft's Phi-2 Outperforms Bigger Models

Small in size but not in intelligence, Microsoft Research debuts Phi-2, a text-to-text small language model AI, packing a punch far above its weight class in the AI arena.

With only 2.7 billion parameters, Phi-2 flexes its computational muscles to outshine its bulkier counterparts like Meta's Llama 2-7B and even Google's latest Gemini Nano 2, showing that efficient design can rival sheer size.

Microsoft trained this Transformer-based model on “textbook-quality” data, including synthetic datasets tailored for teaching common sense, general knowledge, and even coding. In just 14 days using 96 A100 GPUs, they scaled knowledge from their prior Phi-1.5 model and integrated it into Phi-2.

How does Phi-2 stack up against the Goliaths?

• Although dwarfed by other models boasting billions more in parameters, Phi-2 demonstrates less "toxicity" and bias in its outputs, says Microsoft

• This lean machine also took a sly jab at Google's Gemini Ultra, tackling complex physics problems—and catching mistakes just like its larger adversary.

While performance is stellar, Phi-2 does come with bindings. For research only, it remains strictly off-limits for commercial product development under its current licensing - a potential drawback for enterprises itching to harness its capabilities.

This achievement beckons reconsidering the "bigger-is-better" paradigm in AI development, showcasing how a well-crafted smaller model can deliver top-tier performance.

Keep reading here.

PUNCHLINES

Robes Meet Robots: The UK now permits judges to use ChatGPT in legal rulings.

Snapchat 2.0: Snapchat+ subscribers can now create and send AI-generated images based on text prompts.

Essential Capital: Startup Essential AI reportedly emerges from stealth with $56.5M in funding.

From Chatbot to Chatbrat? Users complain ChatGPT ChatGPT is getting lazy and sassy.

TLDR

Meta AI enhances Ray-Ban smart glasses: Meta's new multimodal AI for Ray-Ban smart glasses rolls out in a limited US test phase, offering object identification and language translation through voice commands. Zuckerberg showcases the assistant by demonstrating fashion advice and translation features.

US negotiates AI chip sales with Nvidia: The US government collaborates with Nvidia to restrict China's access to high-end AI chips while allowing sales of less powerful GPUs. Commerce Secretary Raimondo insists Nvidia follows new regulations to prevent China from advancing its AI military capabilities, with Nvidia aiming to develop compliant products for the global market.

Mind-reading AI translates thoughts into text: University of Technology Sydney pioneers DeWave, an AI capable of decoding brainwaves into words with 40% accuracy. This non-invasive EEG-based technology, still in the experimental stage, demonstrates the potential for thought-to-text translation.

Output introduces AI music magic with Pack Generator: Output's new AI tool, Pack Generator, lets users spawn 30-piece, royalty-free sample packs from simple text prompts. Catering to all major DAWs, it crafts unique sounds from Output’s own library. Available in beta, it foretells a broader Co-Producer AI suite launch in 2024.

TRENDING TOOLS

📊 ClientZen: Leverage AI to transform feedback into strategic business decisions

💼 BoldDesk: Accessible customer service software solution at a competitive price

🔥 Krater.ai: Your versatile AI toolkit for diverse digital tasks

🎵 Xound: Enhance your audio content with premium sound production

📑 Claude for Sheets: Integrate Anthropic's Claude AI into your Google Sheets

That’s all for today—if you have any questions or something interesting to share, please reply to this email. We’d love to hear from you!

P.S. If you want to sign up for the Supercharged newsletter or share it with a friend, you can find us here.