OpenAI, Anthropic, and even Google are the leading AI companies at the moment, but they are having a hard time improving their AI models. They are only greeted with dissipating returns on their massive investments.
A recent report explores those challenges by those companies as they work harder to advance AI just as Apple Intelligence features are highlighted by the privacy-first approach.
OpenAI's Orion and Google's Gemini Face Hurdle
Bloomberg initially reported that the company's latest model code named "Oron" failed to meet OpenAI's own benchmarks, especially in terms of coding. Much like GPT-4, Orion is nothing compared to other leaps in performance and even in more complex tasks.
Google also faces similar issues with its new Gemini model, and Anthropic has even delayed the launch of the much-talked-about Claude 3.5 Opus. According to experts, the real problem seems to be the increasing difficulty of finding "new, untapped sources of high-quality, human-generated data" that would train these models.
The other irritation to the nerves of these firms is the excessively prohibitive development and operating costs of the new AI models alongside existing ones.
The Myth of Unlimited Model Scaling
For almost two decades, Silicon Valley has clung to the notion that unlocking the path to artificial general intelligence lies through more computing power, larger datasets, and bigger models. However, that assumption may be a fundamental mistake.
Cost may keep going up without commensurate returns from increasingly complex models. Companies soon will chase other paths. These include fine-tuning AI's responses by humans after training and particular types of AI "agents," which are programmed to perform only predetermined tasks, such as buying airline tickets or mailing emails.
Apple Distinguishes Themselves with an AI Plan that Focuses on Privacy
OpenAI, Google, and Anthropic are significantly pushing into LLM efforts, while Apple's approach differs from them. However, the corporation never focused on general-purpose AI, though its attention has been paid more to enhanced user experiences through private-first AI features on its devices.
For example, Apple Intelligence almost entirely runs on devices. Only tasks that demand more computing are done on its Private Cloud Compute encrypted servers.
Apple's Privacy-Centric AI Strategy Stands Apart
Features of Apple's AI development involve an assortment of new capabilities coming through inside its products, like writing assistance, advanced functionalities of Siri, and image generation ability. These developments are more akin to the in-depth approach and are not aligned with the real nature of LLM space.
However, Apple does not compete directly with companies developing humongous AI models but rather partners with companies like OpenAI which allow Siri to forward complex queries to ChatGPT as the case may be.
Potential Validation of Apple's Conservative Approach
In this regard, Apple might be doing smart work here by focusing on some fine, user-centric AI approach to outsmart the growing hindrances facing general-purpose AI developers, as MacRumors reports.
In such a scenario, Apple's emphasis on security and specific features of AI might work to their advantage and not limit it. As we head to 2025, the iPhone maker looks forward to bringing enhanced AI features in line with the iOS 18.2 update next month.
Related Article: Most Affordable iPad to Get Apple Intelligence Update in 2025: Why?