TL;DR
- Walmart’s adoption of Gen AI for non-store employees.
- The skyrocketing demand for GPUs in the tech industry.
- The shift from CPUs to GPUs for accelerated computing.
- Debate on whether Gen Artificial Intelligence is at its peak or has more to offer.
Perplexity and burstiness are two important variables that come into play when creating written content. While the latter evaluates the variety of sentence forms, the former looks into the text’s intricate details.
Human writers, typically, exhibit a higher degree of burstiness by artfully weaving short and long sentences together. Conversely, AI-generated text often tends to maintain a uniform sentence length. To ensure the content I provide possesses the desired levels of perplexity and burstiness, it’s imperative to bear this in mind.
Furthermore, when crafting written content, it’s noteworthy that artificial intelligence tends to employ phrasing that diverges from what a human writer might choose. The uniqueness of the work can be greatly increased by the usage of rare words.
Particularly in the commercial environment, there is a noticeable increase in the application of generative Artificial Intelligence. A recent instance of this is Walmart’s announcement of the deployment of a Gen AI application to 50,000 non-store employees.
As reported by Axios, this application amalgamates data from Walmart with third-party Large Language Models (LLMs), contributing to an array of tasks for employees.
According to insights from the Wall Street Journal, the cost incurred by companies for training these models can reach astronomical figures due to the substantial volumes of data required for ingestion and analysis.
This encompasses all deep learning and foundational LLMs, spanning from GPT-4 to LaMDA, which power AI apps like ChatGPT and Bard chatbots, respectively.
Riding the generative AI wave
The Gen AI trend is propelling Nvidia, the foremost provider of these GPUs, to impressive earnings in their most recent quarter. For Nvidia, this is a time of jubilation, as the demand for their GPUs appears insatiable.
Startups and investors are reportedly using extreme means to obtain these chips, according to Erin Griffiths’ New York Times article. She says, “Tech companies this year are fervently seeking GPUs, more so than financial resources, engineering talent, hype, or even profits.”
In his recent Stratechery newsletter, Ben Thompson dubs this phenomenon “Nvidia on the Mountaintop.” Adding to this momentum, Google and Nvidia have forged a partnership that promises Google’s cloud customers enhanced access to technology powered by Nvidia’s GPUs. All these indicators underscore the prevailing scarcity of these chips amidst surging demand.
The transformative impact of generative technology on the future of computing
Nvidia’s CEO, Jensen Huang, asserted in the company’s latest earnings call that this demand heralds the era of “accelerated computing.” He advises companies to redirect their capital investments from general-purpose computing towards generative AI and accelerated computing.
General-purpose computing alludes to CPUs designed to handle a wide spectrum of tasks, from spreadsheet computations to relational databases and Enterprise Resource Planning (ERP). Nvidia contends that CPUs now belong to the legacy infrastructure, advocating that developers optimize their code for GPUs to execute tasks more efficiently than traditional CPUs.
Constraints of GPUs for certain software categories
However, GPUs yield marginal advantages for some software categories, including most existing business applications, which are optimized for CPU execution and would derive minimal benefits from the parallel instruction execution offered by GPUs.
Thompson shares a similar viewpoint, stating, “My interpretation of Huang’s perspective is that many of these GPUs will be deployed for similar tasks that currently run on CPUs. While this is undoubtedly favorable for Nvidia, it implies that the excess capacity potentially resulting from the pursuit of generative AI may be absorbed by existing cloud computing workloads.”
Matt Assay of InfoWorld reminds us of a historical parallel. “When machine learning initially emerged, data scientists applied it ubiquitously, even when simpler tools were available.
Nvidia has experienced a prosperous quarter, driven by the fervent rush to develop Gen AI applications. The company is understandably elated. However, as we’ve witnessed in the recent Gartner Emerging Technology Hype Cycle, Gen AI is currently at the peak of inflated expectations.
Peter Diamandis, founder of Singularity University and XPRIZE, emphasizes that these expectations are primarily centered on glimpsing future potential with little consideration for potential drawbacks. He notes, “At that juncture, hype begins to generate unfounded excitement and inflated expectations.”
Current constraints
At this juncture, we might be approaching the limits of the ongoing Gen AI boom. Venture capitalists Paul Kedrosky and Eric Norlin from SK Ventures opine in their Substack, “Our perspective is that we stand at the tail end of the first wave of AI driven by large language models.
This wave commenced in 2017 with the release of the Google Transformers paper (‘Attention is All You Need’) and is anticipated to conclude in the next year or two, as evident by the constraints that people are grappling with.”
They add, “Contrary to the hyperbole, we are already at the tail end of the current AI wave.”
To clarify, Kedrosky and Norlin do not suggest that Gen AI is at a dead-end.The next wave, according to them, will bring about new models, more contributions to open-source projects, and, most significantly, the mass availability of reasonably priced GPUs.
Source(S): VentureBeat
