TOGETHER WITH HYPERMODE.

Good morning, human brains, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

• 50% faster AI = 100% dumber humans 🤖 🤤

  OpenAI unveiled four key updates at DevDay 2024.

• Windows Search = slightly less useless 🙄 📱

  Microsoft announced changes to Copilot and Windows 11.

• Joe Rogan, Schmo Rogan 🎙️ 🔥

  Choose your new favorite podcast.

  New here? Subscribe!😎

Peep today's ‘What Would You Do?’ at the bottom. 👇

MAIN COURSE

OpenAI’s DevDay updates 🤖 🤤

On Tuesday, OpenAI unveiled four key updates at DevDay 2024. OpenAI claims the goal is to make AI more accessible and cost-effective.

1. Realtime API:

• Six new AI voices for you to integrate into apps

• Enables real-time, “lifelike” conversations

• Costs $18 per hour

• Ideal for creating virtual assistants, customer service bots, or interactive storytelling apps

2. Vision fine-tuning API:

• Improves GPT-4o's visual understanding with as few as 100 images

• Enables you to create specialized visual AI for advanced product search in e-commerce, object detection for autonomous vehicles, precise medical image analysis, and more.

3. Prompt Caching:

• Reuses input tokens from previous prompts

• Can reduce processing times by up to 50%

• Particularly useful for code editing, multi-turn conversation bots, any app with repetitive AI queries.

4. Model Distillation:

• Allows developers to create cheaper, specialized models based on GPT-4o or o1-preview outputs

• Includes tools for automatic dataset generation and performance testing

• Offers free training tokens until October 31 (2 million daily for GPT-4o mini, 1 million for GPT-4o)

Why do I care?

These tools could reduce costs and development time for AI-powered applications, and potentially lead to more innovative and accessible AI products across various industries.

What about the last DevDay?

What a coincidence. Last year, we covered OpenAI DevDay 2023. OpenAI introduced GPT-4 Turbo, lower prices, customizable GPTs, a new Assistant API, and more.

Click here for more details.

SPONSORED BY HYPERMODE

Hypermode is how you build intelligent APIs.

Hypermode makes building easy with native access to your favorite libraries, models, and tools. Ship faster without getting bogged down by backend complexity, Built by the team who brought you Vercel, Astronomer, Sentry, and VS Code.

SIDE SALAD

Windows Search = slightly less useless 🙄

On Tuesday, Microsoft announced changes to Copilot and Windows 11 at its New York City event. The updates focus on integrating AI more deeply into the user experience.

What's new with Copilot?

• Redesigned interface across mobile, web, and Windows

• Copilot Vision: Ability to understand what you're looking at

• Natural voice conversation mode

• Virtual news presenter that can read headlines

Any Windows 11 updates?

• Phone Link status in Start menu shows notifications and phone battery life

• AI-powered Windows Search with Click to Do feature

• Paint and Photos get Generative Fill and Erase features

When will these be available?

The Windows 11 2024 update is rolling out now. Some features are for Copilot Plus PCs coming in November.

What's the goal of these changes?

To create a more personalized, AI-driven computing experience across devices.

Any cool AI features?

• Search for photos using text descriptions

• Remove objects from images in Paint (like Google's Magic Eraser)

• Add AI-generated elements to images in Paint

Click here for more info.

YOUR DAILY MUNCH

Think Piece 🧠

Negative time? Physicists observed photons seemingly exiting a material before entering it, suggesting evidence of “negative time” in quantum physics.

Startup News 💰

Microsoft is relaunching its controversial Recall program. It tracks and stores your computer activity to create a searchable digital history.

Anthropic hires another OpenAI cofounder. Durk Kingma announced was joining the team on 𝕏 (formerly Twitter).

Research 👨‍🔬

DiaSynth — generate realistic synthetic dialogues by simulating personas and conversation settings.

TPI-LLM — a system that enables large language models (up to 70 billion parameters) to run efficiently on low-resource edge devices like laptops and smartphones.

Emu3 — a multimodal AI model that tokenizes text, images, and videos into discrete tokens and trains a single transformer using next-token prediction.

FURRY FRIENDS

Respond to this email with your pet’s name and pic for a chance to be featured here tomorrow!

MEMES FOR DESSERT

AI ART SHOW

WHAT WOULD YOU DO?

Joe Rogan, Schmo Rogan… 🎙️

What podcast are you watching? 👇

Ideas? Comments? Complaints?

Respond to this email or hit me up on 𝕏.

Until next time 🤖😋🧠