Large Language Models

Today's Stories

1. 1

   Kroger launches AI meal-planning assistant to drive shopping

   Kroger has introduced an artificial intelligence (AI) shopping assistant that helps customers plan meals and add items to their cart, marking the grocer's entry into AI-powered commerce tools for grocery shopping. The AI assistant may help Kroger increase basket size and customer engagement by simplifying meal planning and shopping—a move that signals how retailers are turning to AI to enhance the in-store and digital experience and compete as e-commerce alternatives grow.

   The rollout scope, pricing, availability on Kroger's app and website, and whether the tool drives measurable gains in customer spending and loyalty will indicate whether AI-driven meal planning becomes a standard grocery-retail feature.

2. 2

   Legal teams need problem-first AI, not tech-first tooling

   Kimberly Harlowe, Senior Director of Litigation Support and Technology at Altria, outlined a framework for corporate legal departments to adopt AI strategically. The approach prioritizes identifying specific business problems before selecting tools, designing modular infrastructure to swap solutions as AI evolves, and scaling through outside counsel partnerships rather than internal-only automation. Nearly 87% of General Counsel report active AI use, but barely half have a formal rollout roadmap. A problem-driven strategy—starting with low-risk internal wins like organizing legacy documents, then scaling to outside counsel workflows—helps legal leaders earn executive buy-in by anchoring pitches in cost savings or reclaimed capacity, not software features.

   The framework emphasizes building a 'plug-and-play architecture' that avoids locking departments into single-vendor ecosystems, since AI solutions shift every six months. High-impact use cases for outside counsel include automated deposition summaries and discovery drafts, with success tied to aligning law firms around standardized tools and maintaining human review at each step.

3. 3

   Walmart uses AI to predict weather disruptions, reroute inventory before storms hit

   Walmart deploys predictive AI and machine learning models that factor in historical weather patterns and real-time data to forecast how severe weather will affect its supply chain. The system helps planners reposition inventory, adjust transit times, and reroute shipments to unaffected locations before disruptions occur. The retailer also uses an "intelligent fulfillment engine" to recalculate delivery paths during weather events and operates a storm rerouting agent in Canada that cross-references 10-day forecasts with highway and ferry closures. Severe weather can spike demand for essentials (like umbrellas during heavy rain), reduce fulfillment center capacity, and close key routes—especially in remote regions where ferry cancellations may be the only way to reach stores. By predicting these impacts rather than reacting to them, Walmart can maintain on-time deliveries and keep stores stocked while also protecting driver safety during emergencies like wildfires in California and Colorado.

   Walmart is using the same AI approach to optimize its network design long-term, not just handle short-term weather crises. Indira Uppuluri, SVP of supply chain technology, noted the company is "learning how to improve its supply chain for the long haul" through simulations that may reveal structural changes to how the network operates.

4. 4

   HR Tech Startup PeopleX Eyes AI-Driven Organizational Overhaul

   PeopleX, an HR technology company, is positioning itself around AI agents that will reshape organizational structures, according to comments from the company's leadership about the outlook for the next five years. The shift toward AI-integrated human resources signals that HR functions—traditionally viewed as administrative—are becoming core to how companies operate and compete. This reflects a broader trend where talent management and organizational design are seen as strategic priorities rather than cost centers.

   The specific capabilities and timeline for PeopleX's AI agent deployment remain to be detailed, but the company's vision suggests HR teams will need to rethink hiring, structure, and workforce planning within the next five years.

5. 5

   Meta, Microsoft both face surging AI costs amid divergent profit trends

   Microsoft and Meta released earnings reports on Wednesday showing both companies are spending heavily on AI infrastructure. Microsoft increased its profits despite the higher costs, while Meta reported a sharp decline in earnings. The divergence in profitability signals different returns on AI investment between the two tech giants. For investors and business leaders watching AI deployment trends, Microsoft's ability to offset AI spending with profit growth may indicate stronger monetization of its AI work, while Meta's earnings decline suggests the company is not yet realizing comparable financial returns from its AI investments.

   The ongoing trajectory of both companies' AI spending and their ability to convert that spending into revenue growth will shape competitive positioning in enterprise AI and cloud services markets.

6. 6

   6 LLM deployment formats compared: speed vs. hardware flexibility

   A technical breakdown of six production LLM deployment formats—Pickle, safetensors, GGUF, ONNX, MLX, and TensorRT—each with different tradeoffs between inference speed and hardware compatibility. Pickle and safetensors run anywhere Python runs; GGUF bundles weights and tokenizer for CPU-only machines; ONNX lets models switch frameworks; MLX optimizes for Apple silicon; TensorRT compiles to specific NVIDIA GPUs for maximum speed. Faster models require hardware-specific choices baked into the file, so a single model must be rebuilt separately for each chip or runtime it needs to support. Teams must choose how far down the speed-versus-compatibility ladder to go—trading inference latency against the types of hardware that can actually run the model in production.

   The article emphasizes holding the deployment format as high on the speed-compatibility chart as the latency budget allows, because every step toward faster execution narrows the hardware it can run on. Teams deploying LLMs face a constant tradeoff between minimizing response time and maximizing portability.