Apple's new M4 Chips

Good Morning! Welcome to another edition of Lightway.

Here’s everything you need to know to start your week:

→ Apple’s new M4 chips
→ OpenAI could launch new Google-competitor today
→ Microsoft developing MAI-1 AI model to compete with GPT-4
→ Google’s Med-Gemini AI outperforms doctors, and more Tech & AI news.

Reading time in 5 minutes.

Highlights of the week

Apple introduces new M4 chips

Apple introduced the next-generation M4 chip, designed to enhance the iPad Pro's performance, particularly in AI-related tasks.

• Performance Upgrade: The M4 chip comes with a new neural engine, achieving 38 trillion operations per second, which is 60 times faster than the A11 chip's NPU.

• Efficiency and Power: Despite its power, the M4 requires only a quarter of the power compared to the latest PC chips.

• 3 nm chips: Utilizing 3nm technology, the M4 includes a 10-core CPU split between performance and efficiency cores and a 10-core GPU.

• Dynamic caching: The GPU supports new capabilities like dynamic caching and ray tracing, and the chip enhances display engine speed significantly.

This development is a part of Apple's strategy to maintain leadership in high-performance, energy-efficient computing for mobile devices. (Read More)

Microsoft developing MAI-1 to compete with GPT-4

Microsoft's CTO, Kevin Scott, confirmed on LinkedIn the ongoing development of a powerful AI model called MAI-1, aiming to redefine tech giant collaborations and model capabilities.

• MAI-1: Microsoft is developing its own large-scale AI model, MAI-1, set to rival OpenAI's GPT-4 and Google's Gemini.

• 500 billion parameters: MAI-1 is built on technology and training data from the startup Inflection, boasting around 500 billion parameters, surpassing many existing models.

• Dual Strategy: Microsoft isn't just stopping at big models; they’re also working on smaller, efficient models for everyday devices.

Microsoft might show a preview of MAI-1 the Build at developer conference later this month. (read more)

Google’s Medical AI outperforms Doctors

Google's Med-Gemini AI model set a new performance record for 10 out of the 14 standard benchmarks for AI models in Medical care, beating GPT-4 and even some humans. The new AI models -

• Multimodal AI Model: Med-Gemini is designed to process diverse data types—text, images, videos, and audio—enabling it to support doctors in complex clinical situations.

• Advanced Diagnostics: The model has been trained to aid in tasks such as detecting early signs of colon cancer, diagnosing autism from eye images, and ensuring complete cancerous tissue removal during surgeries.

• High Accuracy: In tests, Med-Gemini achieved a 91.1% accuracy rate on the US Medical License Exam benchmarks, surpassing previous models and even some human benchmarks.

• Web Search Integration: Med-Gemini can search the web to pull in additional information when faced with uncertain medical scenarios, enhancing its diagnostic capabilities.

Med-Gemini and similar AI models could significantly change how medical professionals access and utilize information, potentially leading to better patient outcomes. (read more)

 Tech Roundup

→ SpaceX unveils new Extravehicular activity (EVA) spacesuits for commercial astronaut spacewalk (link)

→ OpenAI to launch a new tool capable of detecting AI-generated images from Dall-e. (link)

→ Apple could partner with OpenAI to integrate ChatGPT into the iPhone's upcoming iOS 18 (link)

→ OpenAI could unveil a new Google-Competitor ‘SearchGPT’ today, along with new GPT-4 features (link)

→ Waymo, a self-driving company owned by Alphabet, hit 50,000 paid trips every week across Phoenix, San Francisco, and Los Angeles. (link)

→ Apple issues rare apology over ‘crushing’ iPad Pro ad that sparked outrage (link)

→ ElevenLabs unveils a new model to create songs from user prompts, with voice cloning and synthesis. (link)

→ Mistral AI reportedly seeking to raise $600M at $6B valuation (link)

→ Bumble founder says your dating 'AI concierge' will soon date hundreds of other people's 'concierges' for you (link)

 Feedback