- AI Toast
- Posts
- ChatGPT has ads
Welcome to AI Toast!
Here’s what’s on the menu today:
ChatGPT has ads
Universal Commerce Protocol by Google
Privacy AI using Confer
Superhero Tools for writing
Voyage AI by MongoDB
Quick AI Toasts: Metaverse is no more
Total read time: About 5 minutes, perfect for a quick coffee break.
Image Source: OpenAI Blog
OpenAI is changing its business model to include advertisements within ChatGPT starting in early 2026. This move aims to offset massive infrastructure costs while expanding global access to its tools.
Testing begins in the United States for adult users on the free tier and the new $8 monthly Go plan.
Advertisements will appear at the bottom of responses in a clearly labeled and visually distinct box.
The company maintains that sponsored content will not influence the objectivity or accuracy of the chatbot's answers.
Higher subscription tiers like Plus, Pro, and Enterprise will remain completely ad-free for all users.
Privacy protections include blocking ads for minors and preventing them from appearing near sensitive topics like politics or health.
Users can disable ad personalization and clear any data used to generate these sponsored recommendations at any time.
This clearly highlights the high cost of maintaining advanced models and suggests a future where AI access follows the classic subsidization model of search engines.
Personal Take: The era of discounts on AI tools is closing as companies face the reality of billion-dollar operating costs. And humans being humans we always come back to ads to sustain any infrastructure. We can expect conversational commerce to become the standard way many people interact with these tools.
mage Source: Google Blog
Google has introduced the Universal Commerce Protocol (UCP) to standardize how AI agents handle the entire shopping process from discovery to checkout. This open-source framework allows digital assistants to communicate directly with retailer backends without needing custom integrations for every individual store.
The protocol enables AI agents like Gemini to search inventory, negotiate prices, and complete transactions within a single conversation.
Major partners including Shopify, Walmart, Target, and Etsy have already endorsed the standard to help reach shoppers in AI interfaces.
Merchants remain the seller of record, maintaining full control over customer data, branding, and fulfillment throughout the process.
Technical implementation involves a machine-readable file located on the merchant's domain that describes supported services like cart management and payments.
Security is managed through tokenized credentials and verifiable consent, ensuring that automated agents cannot make unauthorized purchases.
The system supports various integration methods, including the Model Context Protocol, to fit into different existing retail tech stacks.
This shift moves online shopping away from traditional web browsing and toward a model where software handles the mechanical steps of buying.
Personal Take: The future path of retail will likely involve personal AI shoppers that manage our household needs and budgets with minimal human intervention. We are moving toward a web where websites are built as much for machines to read as they are for people to view. It’s gonna be really interesting to see if humans end up like the E.T. movie.
Voyage AI by MongoDB just launched the Voyage 4 series
Voyage AI by MongoDB just launched the Voyage 4 series, a unified family of text embedding models featuring an industry-first shared embedding space🧬
Say goodbye to re-indexing when switching between models.
The bar chart below compares the average retrieval quality (as measured by NDCG@10) of the Voyage 4 series of models along with Gemini Embedding 001, Cohere Embed v4, and OpenAI v3 Large.
Overall, voyage-4-large is the top-performing model, surpassing voyage-4, voyage-4-lite, Gemini Embedding 001, Cohere Embed v4, and OpenAI v3 Large by an average of 1.87%, 4.80%, 3.87%, 8.20%, and 14.05%, respectively. Link
Breaking the traditional tradeoff between embedding quality and cost 📉
The series includes voyage-4-large, voyage-4, voyage-4-lite, and the open-weighted voyage-4-nano.
All models in the series produce compatible embeddings, allowing customers to mix and match models for query and document embedding based on their specific accuracy, latency, and cost requirements.
voyage-4-large. The new flagship embedding model leveraging a mixture-of-experts (MoE) architecture to establish a new state-of-the-art while maintaining serving costs 40% lower than comparable dense models. This is the first production-grade embedding model to utilize MoE architecture.
voyage-4. Approaches the retrieval quality of voyage-3-large while maintaining the efficiency of a mid-sized model.
voyage-4-lite. Approaches the retrieval accuracy of voyage-3.5 while requiring significantly fewer parameters, enabling high-quality embeddings at significantly reduced computational cost.
voyage-4-nano. Voyage AI’s first open-weight model, freely available on Hugging Face under the Apache 2.0 license. voyage-4-nano is ideal for local development and prototyping with an easy path to production.
All models produce compatible embeddings, allowing customers to mix and match models for query and document embedding based on their specific accuracy, latency, and cost requirements.
Flexibility & Open Source. Mix, match, and optimize for your pipeline. 🛠️
Index your docs with voyage-4-large and query with voyage-4-lite.
Developing locally? voyage-4-nano is open-weights and available now on Hugging Face for a seamless path to production. 🤗
Ready to build the next generation of agentic AI?
Start building with Voyage 4 today (First 200M tokens free! 🎁)
🔗 Read the full launch details: Click here
Image generated with Gemini
Be it our health questions, financial worries, relationships and private thoughts, privacy has always been the uncomfortable tradeoff in AI. Confer, built by Signal founder Moxie Marlinspike, takes a different approach, privacy enforced by architecture, not promises. Instead of treating data protection as a feature, Confer treats it as the system’s foundation and its technical stack is designed such that conversations remain inaccessible even to its own operators.
Messages are encrypted using WebAuthn passkeys before they ever leave your device.
Data is processed only inside Trusted Execution Environments (TEEs) isolated hardware enclaves that block access from administrators and attackers alike.
The system uses remote attestation to continuously verify that the secure environment hasn’t been tampered with.
Inside the enclave, open-weight foundation models process requests without retaining conversation data.
The result is an AI assistant that can respond intelligently without accumulating behavioral data trails.
Building privacy at this level isn’t easy.
TEEs require specialized hardware and careful configuration.
WebAuthn works best on mobile devices and modern Macs, with workarounds needed on Windows and Linux.
Secure enclaves introduce computational overhead that increases infrastructure cost.
Confer accepts these constraints as the price of meaningful privacy, prioritizing correctness over convenience.
What Security Experts Are Saying
Researchers see Confer’s approach as materially stronger than typical privacy controls.
Hardware-based enclaves combined with encryption create protections that software-only systems can’t easily match.
The design significantly limits internal abuse, data leaks and retroactive policy changes.
The tradeoff remains device compatibility and computational cost.
Confer’s existence alone pressures the market.
It proves privacy-first AI is technically possible.
It challenges the assumption that personalization requires surveillance.
It forces larger platforms to justify weaker privacy models.
If demand grows, privacy could become a competitive differentiator rather than a regulatory checkbox.
Personal Take: This is what real privacy engineering looks like not settings pages and policy documents, but systems that physically prevent abuse. Confer doesn’t ask users to trust it. It removes the need for trust entirely. As AI becomes the interface to our inner lives, architectures like this may become less of a niche and more of a necessity.
Superhero Tools for Writing
Grammarly provides real-time writing assistance by correcting grammar, spelling, and tone across various platforms and applications.
Hemingway App improves readability by highlighting complex sentences, passive voice, and unnecessary adverbs to ensure your writing is bold and clear.
Writer is an enterprise AI platform that helps teams generate consistent, on-brand content while ensuring data security and compliance.
Surfer SEO analyzes search engine results to provide data-driven guidelines for optimizing content to rank higher in organic search.
HubSpot offers an integrated content hub that combines AI-powered writing tools with SEO recommendations and customer relationship management.
Quick AI News Bites
Meta is officially shuttering its Horizon Worlds platform as the company shifts its primary focus toward physical robotics and advanced AI hardware. This is the end of a multi-billion dollar bet on virtual social spaces that struggled to gain mainstream traction. The company now plans to repurpose its metaverse engineering talent to accelerate the development of general-purpose robots for industrial and domestic use.
California Attorney General Rob Bonta issued a cease-and-desist letter to Elon Musk’s xAI regarding the company's data collection practices. The state demands that xAI immediately stop scraping public records and private user data for model training until it complies with local privacy laws. The legal notice includes a strict deadline for the company to provide a full audit of its data sources.
Moxie Marlinspike has launched Confer, a new chatbot platform designed to operate with the same end-to-end encryption standards as Signal. The system prioritizes user privacy above all else by ensuring that neither the platform nor the model providers can access the text of conversation logs. This move targets professional sectors that require strict confidentiality when using generative tools.
Elon Musk has escalated his legal battle against OpenAI and Microsoft by seeking up to $134 billion in damages. The lawsuit alleges that the partnership breached original founding agreements by prioritizing profits over the development of open-source artificial intelligence for the benefit of humanity. Lawyers argue the requested amount reflects the estimated commercial value gained from technology Musk originally helped fund.
VC’s are pouring record amounts of capital into AI-driven diagnostic tools as the healthcare sector seeks to reduce administrative overhead. Industry analysts describe the current market as an AI healthcare gold rush driven by a shortage of clinical staff and rising patient demands. Startups focusing on automated radiology and predictive patient monitoring are receiving the largest share of recent funding rounds.
Boost revenue and gain new customers by partnering with us
Reach over 35K AI enthusiasts with your product.
Join our newsletter to connect with tech professionals, investors, engineers, managers, and business owners worldwide. DM now!
Got feedback, a story worth toasting, or a wild tech question? I’d love to hear from you; just reply, or find me on X.
Cheers,
— Poonam Soni