Retail giant Walmart is no stranger to AI and has begun testing its own large language model (LLM), which it may use for other applications.

Wallaby, a suite of retail-focused LLMs, is trained on decades of Walmart data and understands how Walmart employees and customers talk. It is also trained to respond in a more natural tone to better align with Walmart’s core values of customer service. 

Desirée Gosby, vice president of emerging Technology at Walmart Global Tech, told VentureBeat in an interview that the company wants to extensively test Wallaby before releasing it to a wider audience. 

“Wallaby is not being used yet because we are testing it quite heavily internally, in particular with our associates since we have such a large base of associates,” Gosby said. “Over the next year, we’ll start to leverage it.” 

While she thinks Wallaby’s first use cases will be more consumer-facing, the new LLM will be part of a stable of models Walmart plans to use when developing new applications. 

It’s not surprising that Walmart would choose to train its own series of retail-specific LLMs. Developing internal models is expensive, and even fine-tuning third-party models can get expensive. As one of the largest retailers in the world, Walmart not only has the war chest to experiment with AI models, but it also sits on a ton of customer, employee, logistics and retail data that enriches a model. 

Mixture of models

Like many companies, the retailer prefers to use the best model it finds for the use case it wants to address. Sometimes, this means using off-the-shelf or third-party models, or Walmart can use another previously developed algorithm.

Gosby said Walmart has a multi-layered approach to bringing AI to its technology stack. It uses a platform called Element to plug and play different models, both from third parties or its proprietary LLMs, to direct them to specific applications. 

“It’s helping us manage those models, and at the foundation are the different LLMs we use, one of which is the retail-based ones in Wallaby,” Gosby said. “At the end of the day, it’s really going to come down to what problems we’re trying to solve, and we will figure out the best approach, maybe it’s leveraging a mixture of models.”

Gosby said during VentureBeat Transform in July that the retailer has been expanding its technology use and taking a platform approach to integrating AI. Walmart had been using GPT-4 for many of its AI applications, especially as more customers were turning to its mobile and web storefronts. 

Of course, Walmart is not the only retailer with AI in its applications. Amazon released Rufus, a chatbot powered by AWS models that answers questions about products and references customers’ reviews on the platform. 

AI all over its applications

Employees at Walmart’s Bentonville, Arkansas headquarters have been using AI for many years. The company rolled out a chatbot for associates to ask questions about Walmart’s policies and employee handbooks. It has also brought in AI to streamline its operations, including managing its supply chain.

Its physical storefronts —Walmart and Sam’s Club — also feature AI technology to help floor associates assess inventory and manage check-out procedures. 

Walmart also expanded generative AI tools on its digital platforms. Gosby said Walmart has had a chat feature for years, but it enhanced its Customer Support Assistant this month. The chatbot will now recognize customers from the start and will be able to understand customer intent.

Walmart showed reporters a demo of the feature that lets a customer express in natural language that they accidentally bought some toys. The Customer Support Assistance focuses on the word “toy” and brings up a recent order for toys, ignoring other purchases that do not fit the conversation’s context. Without prompting, the assistant will also ask if the customer wants to return or keep the product. 

The retailer plans to roll out more personalized recommendations and homepages to customers. 

Gosby said Walmart will continue to explore how AI can help smooth customer and employee experiences. And in the next year, these experiences may include applications powered by Wallaby. 

Today on Equity, Rebecca Bellan sat down with Zander Brumbaugh, the 22-year-old founder of Gamebeast, a startup that offers Roblox developers tools like A/B testing and LiveOps to modify games without needing to release a new version or interrupt a user’s ongoing game. 

Gamebeast recently raised a $3.7 million pre-seed round, led by J2 Ventures with participation from a16z’s Speedrun accelerator, which Brumbaugh graduated from in March. Brumbaugh said everything from his experience consulting for entertainment studios like Netflix, and writing a best-selling book on Roblox development, to chats with investors tell him that the future of video gaming is in user-generated content, or UGC. 

The two discussed the rise in popularity of UGC games, a sector that investors are increasingly looking towards as drivers of growth in the video game industry. Brumbaugh noted that a large majority of kids under 16-years-old asked their parents for virtual currency, like Roblox’s Robux or Fortnite’s V-Bucks, during the last holiday season.   

Most of the games, experiences, and virtual worlds on Roblox are user-generated, and that’s by design. Fortnite, which is well known for its Battle Royale player-vs-player game, also has a sandbox mode that allows players to build and create their own worlds and experiences. Creative mode is fast gaining popularity.

“These platforms make it extremely easy to make content from a young age,” Brumbaugh said. “I started when I was 12, and taught myself how to program in Lua and kind of learn my way around the Roblox engine. I think people being able to creatively express themselves and make games that they themselves enjoy, and then also turn that into a profitable hobby, or in many cases, a career, that’s a really exciting thing.”

On the show, Brumbaugh also shared his thoughts on how generative AI is changing the gaming landscape. 

“You can’t really discuss AI, especially in creative spaces, without there being an immediate knee-jerk negative reaction from a lot of people in the community,” Brumbaugh said. “But ultimately, I think as models get better, the best thing is always going to be integrating these types of models into a workflow, making something easier for developers.”

Brumbaugh didn’t only get investment for his startup because investors see the value in enabling UGC game development. J2 Ventures’ thesis focuses on dual-use technology, and Gamebeast was able to demonstrate how its tech has applications both in commercial and defense. And indeed, we’re seeing a trend of companies going for that sweet military money to stay afloat.

There are more insights about the future of the video game market and investor insights aplenty, so have a listen, and enjoy!

Equity is TechCrunch’s flagship podcast, produced by Theresa Loconsolo, and posts every Wednesday and Friday. 

Subscribe to us on Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. For the full episode transcript, for those who prefer reading over listening, check out our full archive of episodes over at Simplecast.

That’s just a glimpse of the best Apple deals we’re currently seeing, though. Below, we’ve compiled the best across a range of categories — including tablets, headphones, styluses, and more — so you can sift through Apple’s various wears more easily. We’ve also put together a larger guide to the best Prime Day deals overall, just in case you also want to take a look at what’s on sale outside of the Apple ecosystem.

Update, October 9th: Adjusted pricing and availability.

You’ve probably never heard of Foxconn Technology Group, but it is a Taiwanese consumer electronics manufacturer that is perhaps best known for supplying the world with Apple iPhones – it also makes most of Amazon’s Kindles and Nintendo’s gaming consoles. 

Stopping short of actually producing the folklore legend Apple Car, the company has branched out into developing electric vehicles of its own under the Foxtron moniker, with its existing Model C five-or seven-seater SUV already licensed by the rapidly expanding Taiwanese automotive brand Luxgen and badged the n7.