“EBay, a survivor of the first generation of internet companies, is trying to remake itself for the age of artificial intelligence.
The company rode the internet's early e-commerce boom and the dawn of the mobile era to become a publicly traded billion-dollar firm, but has since faced activist investor pressure amid slowing sales.
Looking to boost its business, the company is making use of AI agents, encouraging engineers to use AI to write code and launching AI features that personalize buying and simplify selling.
The San Jose, Calif.-based company manages and runs its own data centers, helping it maneuver the tech infrastructure changes that AI requires, the company said. That includes amassing its own graphics processing units.
Parantap Lahiri, eBay's vice president of network and data-center engineering, said the company has thousands of its own GPUs but also relies on the major cloud providers, including Microsoft's Azure. The company works with multiple suppliers, including Nvidia, to procure its GPUs and other hardware.
EBay also uses AI models from ChatGPT-maker OpenAI, Lahiri said, as well as open-source models from companies like Meta Platforms, and trains its own AI models.
The company last year announced that it had built a supercomputer using its own hardware. That enabled eBay to build and run models about 100 times as large as the year before, according to the company.
EBay's other advantage, its executives argue, is its three decades of data from its buyers and sellers, who have collectively made billions of transactions.
In other areas of the company, AI is a foundation for its over 11,500 employees, who are asked to work the technology into their daily tasks, executives say.
Each project an employee embarks on should have an AI agent attached, said Nitzan Mekel-Bobrov, eBay's chief AI officer. The bots, built and hosted on eBay's platform, can be created by any employee for specific tasks they are doing, have a record of every meeting, and should reduce the need for workers to schedule new ones by a double-digit percentage this year, he said.
For eBay's engineers, AI is used to help write code. The company is using and testing different AI-based coding tools, such as Microsoft's GitHub Copilot and "vibe coding" tools like Windsurf and Cursor, according to Senthil Padmanabhan, eBay's vice president of engineering, global verticals and API platform.
The company also built its own AI model, trained on over 100 million lines of its own code, to do things like improve the efficiency of its code upgrades and code translations.
To make it easier for sellers to list items, eBay rolled out an AI feature called "magical listings." After sellers take a photo, eBay's AI fills in the rest of the item's details, including its suggested listing price.
Like other AI-generated content, however, eBay's item descriptions have been criticized online as low-quality, shallow content. On a Reddit thread, one user said they preferred eBay sellers that didn't use the AI function, calling its use a sign that sellers weren't knowledgeable.
EBay said that as AI itself has improved, it has continued to make improvements to its tech based on feedback from buyers and sellers, and sellers always have control over the final product description.
One of eBay's challenges is going up against deep-pocketed tech giants. Amazon.com last year announced Rufus, an AI-powered shopping assistant that relies on neurosymbolic AI [1] to help buyers navigate the shopping platform and answer product-related questions. Bargain-focused retailers Temu and Shein also have been gaining ground.
Yet it may not matter whether eBay can capture mass retail share going up against the likes of Amazon, according to Deutsche Bank's Horowitz. As long as eBay keeps growing in its key product categories that could be enough to satisfy Wall Street.
In its most recent quarter, eBay logged higher profit and revenue as its gross merchandise volume rose. The company posted a profit of $368 million, compared with $224 million a year earlier.
"The investments they are making into their technology stack around AI are putting them in a position to take share in the verticals they're focused in," Horowitz said.
---
Belle Lin writes about AI and enterprise technology for WSJ Pro.” [2]
1. Neurosymbolic AI (NSAI) integrates neural networks for perception and pattern recognition with symbolic AI for logical reasoning and structure to create more robust, interpretable, and accurate AI systems. By combining these approaches, NSAI aims to overcome limitations of pure deep learning, such as the "black box" nature of neural networks and a tendency to hallucinate, by adding explicit reasoning and constraint enforcement. This synergy allows for the development of AI systems that can not only learn from data but also apply logical rules, make verifiable claims, and understand compositionality, leading to applications in fields like healthcare and legal technology where accuracy and explainability are critical.
How it Works
Neural Networks
are excellent at processing raw data and learning complex patterns, mimicking perception.
Symbolic AI
provides explicit rules, logic, and structured knowledge, allowing for verifiable proofs and compositional reasoning.
Integration
allows neural networks to learn features and relationships, while symbolic components use these to perform logical deductions, verify facts, and enforce constraints.
Key Benefits
Enhanced Accuracy and Reduced Hallucinations:
NSAI can provide verifiable, logically sound outputs, minimizing the errors and "hallucinations" that can occur in large language models.
Improved Interpretability:
By incorporating symbolic logic, it becomes possible to understand and explain how an AI system arrives at a decision, a significant improvement over traditional deep learning models.
Greater Robustness:
The combination of learning and reasoning creates more reliable systems that can better handle novel situations outside their training data distribution.
Compositional Reasoning:
NSAI can understand how different parts of a system relate to each other and how the whole can be predicted from its parts, a crucial capability for complex problem-solving.
Applications
Healthcare:
More accurate and interpretable medical diagnosis systems that can incorporate existing medical knowledge with new data.
Legal Industry:
Assisting with complex legal workflows and ensuring the accuracy of outputs where precision is paramount.
Formal Truths:
Developing AI safeguards that can formalize and verify truth in specific domains, such as healthcare or financial policies.
2. Business News: EBay Takes on Artificial Intelligence to Spur Growth. Lin, Belle. Wall Street Journal, Eastern edition; New York, N.Y.. 10 Sep 2025: B12.
Komentarų nėra:
Rašyti komentarą