Lo and behold, the Cupertino colossus unveiled Apple Intelligence at the illustrious WWDC 2024 gala, thus joining the titanic AI skirmish against the likes of Google, OpenAI, and Microsoft. In a bid to infuse fresh AI innovations into iOS 18 and macOS Sequoia, Apple has meticulously crafted its own foundational AI models, catering to both on-device and cloud processing endeavors.
Upon scrutinizing Apple’s petite on-device model, trained on a trifling 3 billion parameters, one discovers that the grandiose server-class models have found solace in Apple’s very own Private Cloud Compute. Alas, while the on-device model excels in most tasks, when faced with Herculean challenges, the baton is passed to Apple’s towering server models. Interestingly, Apple has ushered in ChatGPT to grace the realms of iPhone, iPad, and Mac with its presence.
A gallant leap into uncharted waters, Apple has birthed its very own LLM (Large Language Model), thus piquing curiosity regarding its performance in comparison to the stellar creations of OpenAI, Google, and Microsoft.
Behold the saga of Apple’s AI models! Two distinct beings emerge from Apple’s crucible: a modest on-device model, wielding 3 billion parameters, and a grandiose server model ensconced within the folds of Apple’s cloud infrastructure. The lingering mystery of the server model’s parameter size enshrouds it in an air of enigma.
To power its on-device endeavors, Apple has harnessed the ingenuity of LoRA (Low-Rank Adaptation) adapters, ushering in an era brimming with precise modules tailored to specific tasks. These adapters, akin to whispers from ancient tomes, serve to elevate accuracy and efficiency, harmonizing with the vast, untamed plains of large uncompressed models.
Apple avows to have imbued its AI models with the essence of licensed data and domain-specific datasets, lending them an aura of refinement and prowess. Delving deeper, Apple’s web crawler, AppleBot, has gallivanted across the digital landscape to gather publicly available data, ensnaring it within the grasp of Apple’s AI algorithms.
Traversing the realms of performance, Apple’s on-device AI model (3B) grapples with Microsoft’s Phi-3-mini model (3.8B), Google’s Gemma-1.1-2B and 1.1-7B behemoths, and Mistral’s 7B titan. In tests of email and notification summarization, Apple’s on-device model emerges triumphant in the face of Phi-3-mini.
In a captivating evaluation tempered by human judgment, Apple’s on-device model emerges victorious, outshining Gemma 2B, Mistral 7B, Phi-3-mini, and Gemma 7B.
As the spotlight turns to the towering Apple server model, a dance of comparison ensues, with revelations of prowess eclipsing GPT-3.5 Turbo, Mixtral 8x22B, and DBRX Instruct. Alas, against the formidable foe that is GPT-4 Turbo, Apple’s server model bows out, yet not without leaving a mark of distinction in the annals of AI lore.
In the realm of instructions following, Apple’s on-device model waltzes with grace, holding its own against the shadow of GPT-4 Turbo, outshining Mixtral 8x22B, GPT-3.5 Turbo, and DBRX Instruct.
Amidst the realm of writing benchmarks, Apple’s on-device and server models stand tall, overshadowing their counterparts from rival factions.
Finally, in a test of safety and harmlessness, Apple’s AI models emerge as paragons of virtue, generating responses of the least harmful nature. A testament to Apple’s arduous efforts in guiding and honing their AI creations, ensuring they remain benevolent in the face of sensitive topics.
In the grand tapestry of AI elegance, Apple’s foundation models shine forth as beacons of promise and innovation. Despite their late arrival at the gala, Apple’s generative AI models stand proud, outshining the likes of Microsoft’s Phi-3 and Google’s Gemma 1.1 creations. The server model, too, basks in admiration, often eliciting responses superior to GPT-3.5 Turbo.
Eagerly we await the grand unveiling of Apple Intelligence on compatible devices, a moment of reckoning where we can pit Apple’s models against the titans of industry. What say you, dear readers, of Apple’s AI wonders? Share thy thoughts and musings in the hallowed sanctum of the comments below.
Support our work ❤️
If you enjoyed this article, consider leaving a tip to help us keep publishing great content.
