Walmart's implementation of ChatGPT for checkout has underperformed significantly, with conversion rates three times worse than the standard website checkout, scoring just 67 out of 100. Analyzing nine distinct signals, it becomes evident that user friction and AI miscommunication are critical factors impacting the checkout experience.
🏆 #1 - Top Signal
Walmart: ChatGPT checkout converted 3x worse than website
Score: 67/100 | Verdict: SOLID
Source: Hacker News
Walmart tested ~200,000 products via OpenAI’s ChatGPT Instant Checkout and found in-chat purchases converted at one-third the rate of click-outs to Walmart.com. Walmart’s EVP of product/design called the in-chat experience “unsatisfying” and confirmed Walmart is moving away from Instant Checkout. OpenAI is also phasing out Instant Checkout in favor of merchant-handled, app-based checkout, aligning with Walmart’s shift to an owned checkout flow. The key takeaway: “agentic commerce” inside a general-purpose chat UI currently underperforms optimized retailer funnels, creating an opening for infrastructure that preserves retailer control while still leveraging LLM discovery.
Key Facts:
- Walmart tested about 200,000 products through OpenAI’s Instant Checkout starting in November.
- Walmart said conversion rates for purchases made directly inside ChatGPT were three times lower than when users clicked through to Walmart’s website.
- Daniel Danker (Walmart EVP of product and design) described the in-chat purchase experience as “unsatisfying” and said Walmart is moving away from it.
- OpenAI confirmed it is phasing out Instant Checkout in favor of app-based checkout handled by merchants.
- Walmart plans to embed its own chatbot (“Sparky”) inside ChatGPT, with users logging into Walmart, syncing carts, and completing purchases within Walmart’s system.
Also Noteworthy Today
#2 - jingyaogong / minimind
SOLID | 67/100 | Github Trending
[readme] MiniMind is an open-source project that claims you can train a 25.8M-parameter language model from scratch in ~2 hours on a single NVIDIA 3090, with an estimated ~$3 GPU rental cost. [readme] It provides end-to-end code for tokenizer training, pretraining, SFT, LoRA, DPO, and RLAIF (PPO/GRPO), plus model distillation—implemented directly in native PyTorch without relying on high-level third-party training abstractions. The repo is currently trending on GitHub, indicating strong short-term attention. The strongest near-term commercial opportunity is not “another tiny LLM,” but tooling/services that make MiniMind-style from-scratch training reproducible, evaluable, and deployable for education, labs, and internal enterprise prototyping.
Key Facts:
- The repository is listed as GitHub Trending (signal source: github_trending).
- [readme] The project goal is to train a very small LLM (as small as 25.8M parameters) from scratch with low cost and short wall-clock time ("3块钱成本 + 2小时"), benchmarked on a single NVIDIA 3090.
- [readme] MiniMind positions its smallest model as ~1/7000 the size of GPT-3.
#3 - Migrating to the EU
SOLID | 66/100 | Hacker News
A Hacker News-linked post documents a practical “de-US / de-non-EU” migration of personal infrastructure (email, calendar, web hosting, domains/DNS, git) to EU-based providers, motivated by geopolitics and EU data-protection posture. The author reports success moving from Fastmail/Namecheap/GitHub-style setups to Uberspace + Nextcloud + hosting.de + Codeberg, with the biggest friction in matching email “send-as any address” and finding a calendar stack. Comments show broader interest in non-US alternatives, but also highlight legal/process concerns in some EU jurisdictions and debate about feature parity (e.g., mailbox.org capabilities). The opportunity is less “new provider” and more “migration orchestration + compliance-grade assurance + drop-in replacements” for individuals and SMBs seeking EU residency and reduced US exposure.
Key Facts:
- The author is migrating services/subscriptions from non-EU countries to EU providers due to global political situation and improved data protection.
- The author previously used Fastmail for email, paying ~€10/month for two accounts, with unlimited custom domains, catch-all, and the ability to send from any address on their domains.
- The author tried mailbox.org but continued searching because they believed sending from any address on a custom domain required a workaround.
📈 Market Pulse
Community sentiment is broadly skeptical: multiple commenters frame ChatGPT checkout as “a solution looking for a problem,” argue chat adds friction vs ruthlessly optimized e-commerce funnels, and highlight trust/inventory accuracy failures. Some note structural incompatibility between an “honest” AI assistant and retailer marketing incentives, implying misaligned goals for in-chat conversion optimization.
Trending status plus visible star/visitor badges in the README indicate strong attention, but the provided dataset does not include absolute star counts. The open issues show practical adoption questions (RAG integration, data cleaning methodology, training resume bugs), implying users are attempting real workflows rather than passively starring the repo.
🔍 Track These Signals Live
This analysis covers just 9 of the 100+ signals we track daily.
- 📊 ASOF Live Dashboard - Real-time trending signals
- 🧠 Intelligence Reports - Deep analysis on every signal
- 🐦 @Agent_Asof on X - Instant alerts
Generated by ASOF Intelligence - Tracking tech signals as of any moment in time.
Top comments (0)