- AI Tangle
- Posts
- ☕️ Meta's AI Chips to Superpower AI Models
☕️ Meta's AI Chips to Superpower AI Models
The first 48 hours of the week have been news-packed, from Meta's next-generation AI hardware to boost the already-powerful modern AI workload GPUs, to Apple's line of AI chips to give its Mac a much-needed boost. Meanwhile, Microsoft looks to up its presence in Japan with billions in AI and cloud infrastructure investments, Nvidia and Georgia Tech gives students a hands-on course with an AI supercomputer, and OpenAI releases a silent update for its GPT-4 Turbo with Vision. Join us at AI Tangle as we untangle this week's happenings in AI.
THE BIG AI STORY
In an attempt to catch up with the likes of Microsoft and Google, Meta recently debuted its latest next-generation AI accelerator hardware. Dubbed the Meta Training and Inference Accelerator (or MTIA for short), it is the successor of the company's last year's MTIA v1. Though Meta says the next-generation MITA is already live in 16 of its data centers, the company's statement of vague "3x gains" could be something to read between the lines.
What makes it special anyway?
Meta states that the next-generation MTIA was built on a 5nm process compared to its older sibling's 7nm, which has numerous advantages, from less heat generation to faster computing. Although its power consumption is much higher, the next-generation MTIA makes up for it with 128MB of internal memory compared to its predecessors 64MB, doubling it, and much higher average clock speeds, from 800Mhz to 1.35Ghz, to be complementary to GPU-based AI workloads.
COLLABORATING WITH HUBSPOT
AI Powered Productivity: Transform Your Day
Revolutionize your workday with the power of ChatGPT! Dive into HubSpot’s guide to discover how AI can elevate your productivity and creativity. Learn to automate tasks, enhance decision-making, and foster innovation through the capabilities of ChatGPT.
Download the offer for actionable insights and tips on harnessing AI for a more efficient, impactful professional lifestyle.
Efficiency Unlocked: Enhance productivity with daily ChatGPT integrations
Everyday Solutions: ChatGPT for streamlined email and project management
Practical Insights: Guidelines for effective ChatGPT application in work
Ethical Guidelines: Best practices for responsible ChatGPT use
Future Ready: Skill enhancement and learning with AI tools
6 QUICK HITS
Microsoft-backed OpenAI made an X/Twitter post about its latest update to GPT-4 Turbo with Vision, announcing that the model is now "generally available" through OpenAI's application programming interface (API). One major part of the update is OpenAI allowing developers to use the JSON structure and function calling to handle Vision requests, aiming to streamline workflows. This update comes amid a drought of OpenAI's major GPT releases (such as GPT-4.5 or GPT-5), so it's welcome news to many.
According to a report from Bloomberg's Mark Gurman, Apple is prepping itself to refresh its entire line of Macs with its M4 line of AI-focused processors. These processors come as a family of three: The entry-level M4 dubbed Donan, the middle-of-the-pack Brava, and the flagship Hidra. Bloomberg reports that Apple will most likely unveil more at its Worldwide Developers Conference (WWDC) on the 10th of June, as the company aims to catch up on the "AI PC" trend.
Nvidia and the Georgian Institute of Technology announced this Wednesday that the two designed the first AI supercomputer for education. The Georgian Tech supercomputer runs on 20 of Nvidia's HGX H100 systems totaling 160 of Nvidia's H100 GPUs. So far it has seen use in a single class at the institute this semester - foundations of machine learning. However, the institute plans to scale up its usage eventually, with the institute's undergraduates and graduates having access to it by spring 2025.
American multinational tech company Microsoft announced that it would be putting forth a plan to invest $2.9 billion into its cloud and artificial intelligence infrastructure in Japan over two years, following in the tracks of Google, Amazon, and others making investment plans in foreign countries. The funds will additionally go towards equipping approximately three million people in AI and building its Microsoft Research Asia lab in the Japanese capital, Tokyo.
Rolling out to its U.S. creators first, Quora-owned AI chatbot platform Poe recently introduced a revenue model allowing bot creators to set a per-message price, enabling them to pocket revenue when users interact with their bots, following Quora's revenue-sharing program that launched in October of 2023. Additionally, Poe has increased functionality to show its bot developers a more detailed analytics dashboard to give creators a better perspective on their revenue.
AI-powered patent analytics platform Patlytics secured $4.5 million in seed funding, led by Google's AI-focused Gradient Ventures. Co-founded by Paul Lee and Arthur Jen, Patlytics aims to streamline patent workflows for enterprises and law firms. Its unique approach includes end drafts and extensive chart solutions, mixing software solutions with some hands-on involvement. With its SOC-2-approved product already serving a prestigious clientele, Patlytics plans to invest in product development and expand its engineering team.
4 AI TOOLS
UpRizz - Amplify your Instagram presence with UpRizz, a go-to AI-powered platform to reach a wide audience and increase engagement, be it for yourself or your brand.
Odaptos - Uncover potential user experience problems with Odaptos, an AI-powered analytics problem on a mission to enhance user experience and turn feedback into usable insights.
Dokkio - Find, organize, and understand your online files - Dokkio gives you and your team a single place where files from everyone's cloud accounts come together for effective team collaboration.
Repurpose.io - Repurpose.io is an innovative AI-enhanced software platform that helps make content creation and marketing easier than ever, from repurposing existing content to automating content distribution.
AI READ & WATCH
Safety Evaluations Can't Keep Up with AI (4-min read)
The speed at which artificial intelligence is progressing is old news, but many experts and executives don't see it as a good thing as traditional safety and accuracy methods can't keep up with the rapid development, and what could be done about it.
Udio, Infinite Attention, and GPT's Silent Update (14-min watch)
AI Explained, a YouTube channel dedicated to in-depth reviews of emerging tech and AI, goes over the interesting past few days in AI - Music-making AI Udio, GPT's silent update, and what Google researchers dubbed "infinite attention" for models.