?? OpenAI's New Data Source: 1 Million iPhones
Hello Tuners,
With Nvidia ending the last week at the very top of Market Value Charts, this week prevails for Apple, with them quickly reclaiming the throne with their keynote while prevailing over Microsoft. Well for an average Joe like me it meant that I finally got a calculator for my iPad, but let’s see what the week brought for developers around the globe.
From Apple’s intriguing partnership with OpenAI to Nvidia's mind-blowing financial milestone, this week has been a wild ride. Here are the top stories that were my focus of the week:
Apple Joins Hands with OpenAI for Apple Intelligence
With all eyes staring intently at Apple over the past few years ever since the resurgence of AI through GenAI, Apple has for the first time ever talked openly about AI in their latest keynote rather than softcoating it with terms such as neural networks and such, only though AI stands for Apple Intelligence.
WWDC June 2024, has introduced many Artificial Intelligence features and changes to Apple’s Operating Systems across all platforms, be it iOS, WatchOS, MacOS, or VisionOS. In a revolutionary tie-up between OpenAI and Apple, we will quickly see Siri stronger than ever with Apple’s own security guarantee with on-device processing or choosing to process AI requests on Apple’s secure servers.
3 Trillion Market Cap Leads to Stock Split for Nvidia
In an attempt to make their shares more affordable and maintain stock liquidity, Nvidia began trading on a 10-1 Stock Split earlier this Monday (Meaning if you had 5 shares of Nvidia on Friday, you own 50 shares now valued at the same amount). Earlier last week, Nvidia in a historic move had dethroned the tech giant Apple as the second most valuable company in market share with their insane valuation of 3 Trillion USD.
This rise in valuation and the following decision of Stock-Split comes after the power-hungry struggle between tech giants such as Microsoft, Meta, and OpenAI trying to secure Nvidia machines for themselves in the age of Generative AI.
AI Bots Join Election Season in Wyoming
In the age of more and more students and working professionals turning towards AI for their work, it's about time a politician did! In a more humorous move in AI over the week, Victor Miller, a fellow election candidate running for the Mayor of Cheyenne, Wyoming, with a rather peculiar Campaign Promise, that an all-knowing (not really) rendition of GPT-4, VIC will be making all the decisions once elected.
Miller has stated that the model rather perplexed him saying quote: “I realized that this entity is way smarter than me, and more importantly, way better than some of the outward-facing public servants I see”. Recounting the model’s data extraction and text generation abilities as “the bot has better ideas” and “a better grasp at law”. One seems to wonder is that all that makes a good leader?
Microsoft’s Copilot X PC Launch gets a Hiccup after Feature Announcement
Following the announcement of the unpopular feature “Windows Recall” on May 20, Microsoft has decided to take a back step on the feature launch. Windows Recall is a part of multiple features Microsoft is planning to launch in their new Snapdragon-powered “AI PC”. This feature lets users retrace their steps by an AI-powered search across various screenshots taken through a user’s journey on their systems, which turned out to be not so secure, especially in the hands of people who have access to your devices.
Although the term “AI PC” has lost its magic in the 6 months, with every company having its own rendition of the term (Check Apple’s New AI Announcement), Microsoft’s launch was special due to the introduction of a PC that is not only running base-level cloud-based Copilot, but rather running on-device AI operations on an OS level. Such buzzword-laden keynotes are a result of the ever-balancing act companies are facing rushing to integrate AI into their products.
LLM Of The Week
Tsinghua University's OpenBMB strikes again this week with their new and updated MiniCPM-Llama3-V-2_5. This updated Multimodal LLM stretches the horizons with its outstanding performance which is comparable to proprietary models such as GPT-4o and GPT-4V-0409. The model has been shown to process images of any aspect ratio and up to 1.8 Million Parameters.
By building upon SigLip-400M and Llama3-8B-Instruct, the model stands surpassing other Llama-3-based MMLLMs with just 8B parameters. Leveraging the newest RLAIF-V Method, the model exhibits a rather trustworthy behaviour as far as Multi-Modal LLMs go achieving 10.3% Hallucination rate on Object HalBench.
Weekly Research Spotlight ??
Scalable MatMul-free Language Modeling
A major dominating factor behind the Billions of dollars we see being pooled into training legacy models comes from Matrix Multiplication operations done while training the model. This research introduces a MatMul-free LLM, which does so by replacing the operations in dense layers with ternary accumulations using weights constricted to {-1,0,+1}.
By using a mixture of element-wise product and a Bit-Linear layer optimising the RMSNorm Operations, the research sees the training speeds improve by 25.6% and memory consumption by 61%. One rather amazing outcome of this is the processing of billion-parameter scale models at 13W beyond the human-readable throughput making the model already good enough for daily use.
Best Prompt of the Week ??
“A futuristic cityscape under a night sky with a glowing galaxy. Tall, illuminated skyscrapers dominate the skyline, while a massive, intricately detailed structure with purple foliage spans across a deep valley. The scene has vibrant purple and blue hues, featuring neon lights and advanced architecture. A serene, otherworldly ambiance is created by the celestial bodies and the lush, glowing plants adorning the structures. The entire landscape has a blend of high-tech and natural elements, exuding a sci-fi, cyberpunk aesthetic --stylize 250”
Todays Goal: Try new things ??
Acting as a Storyteller
Prompt: I want you to act as a financial planner. You will create a monthly budget based on income and expenses, and strategically work on how to reduce monthly spending without compromising the client's lifestyle. You will choose a target client profile, develop key strategies and action plans, select the tools and resources for financial management, and decide on any additional activities needed to ensure financial well-being. My first suggestion request is: "I need help creating a financial plan for a junior-level corporate employee."
This Weeks Must-Read Gem ??
That concludes this edition of the newsletter. Thanks for reading, and we can't wait to share more with you in our next one!
Follow our LinkedIn & Twitter for more updates and insights. Until then, stay curious, and we'll see you soon.
We want to give a huge shoutout to our contributors for their tireless efforts.
Content support from: Aryan Kargwal
Lots of ?? from,
Tune AI Team