🗞 OpenAI's Court Challenge, Anthropic's Military Models, and Perplexity's Growth Surge

🔎 The Latest on the AI Frontier:

• OpenAI Challenges Court Order to Preserve ChatGPT Data in NYT Copyright Case

• Anthropic Launches Claude Gov Models for Classified Military Operations

• Perplexity AI Hits 780M Monthly Queries with Explosive Growth Trajectory

• Voice AI Transforms Workplace Productivity Beyond Traditional Chatbots

• Research Reveals LLMs Have Fixed 3.6 Bits Per Parameter Memory Limit

• xAI Upgrades Grok with New Task Scheduling and Automation Features

• Other news you might find interesting

⚖️ OpenAI is appealing a court order requiring it to preserve all ChatGPT output data indefinitely in the New York Times copyright lawsuit, citing user privacy concerns as a core principle that cannot be compromised.

• The dispute stems from the Times' 2023 lawsuit accusing OpenAI and Microsoft of using millions of articles without permission to train ChatGPT, with a judge ruling in April that the newspaper made a valid case for copyright infringement.

• OpenAI CEO Sam Altman called the Times' data preservation demand "inappropriate" and warned it sets a "bad precedent," as the company seeks to vacate the May court order that required segregating all user interaction logs.

• The case highlights the growing tension between AI companies' privacy commitments to users and copyright holders' demands for evidence in what could become a landmark legal battle over training data usage.

⚔️ Anthropic has launched "Claude Gov" AI models specifically designed for classified US national security operations, marking the completion of the AI industry's shift from avoiding military applications to actively pursuing defense contracts.

• The Claude Gov models feature reduced safety guardrails that "refuse less when engaging with classified information," allowing government users to discuss sensitive topics that consumer AI models won't touch—a significant departure from typical AI safety protocols.

• This move puts Anthropic in direct competition with OpenAI and Palantir for the multi-billion dollar government AI market, as the company seeks to raise funding at a potential $40 billion valuation and capture defense revenue directly rather than through partnerships.

• The announcement reflects the broader militarization of AI companies driven by financial necessity, as building large language models is expensive and government contracts provide reliable revenue streams that consumer chatbot subscriptions cannot match.

🔍 Perplexity AI is experiencing explosive growth with 780 million queries per month and over 20% month-over-month growth, positioning itself to potentially reach one billion queries weekly by next year.

• CEO Aravind Srinivas attributes future growth prospects to the company's upcoming "Comet" browser and consumer fatigue with "legacy browsers" like Google Chrome, expecting infinite user retention once people adopt their browser ecosystem.

• The AI search startup differentiates itself by providing sentence-based answers with source links rather than traditional search result lists, and has secured partnerships with Motorola for smartphone integration and reportedly Samsung for device investment and integration.

• Perplexity's growth strategy focuses on distribution partnerships and browser adoption, where users would generate additional queries through search boxes, new tab pages, sidebars, and webpage interactions beyond traditional search behavior.

🎤 Voice AI is quietly transforming workplace productivity beyond chatbots, with tools like Microsoft Copilot and Otter.ai now embedded in meetings, customer service, and healthcare to automate transcription, note-taking, and documentation tasks.

• UK government workers using Microsoft's Copilot saved an average of 26 minutes daily (equivalent to two weeks per year), while healthcare professionals use ambient voice AI to automate patient record-keeping and reduce documentation burnout.

• Fast food chains like Hungry Jack's and KFC are testing AI-powered drive-through ordering systems, while startups like Ring AI offer no-code voice solutions that allow smaller businesses to deploy multilingual customer support at enterprise scale.

• The shift means workers will handle fewer routine tasks and focus more on judgment, empathy, and problem-solving, while businesses must balance productivity gains with privacy concerns and staff retraining needs.

🧠 New research from Meta, Google, Nvidia and Cornell reveals that large language models have a fixed memorization capacity of exactly 3.6 bits per parameter, settling a crucial debate about how AI systems store versus generalize information.

• The study found that models trained on larger datasets actually memorize less per individual data point, with memorization capacity distributed across the entire training set—meaning more training data leads to safer generalization behavior rather than increased copyright risk.

• A 1.5 billion parameter model can memorize approximately 675 megabytes of raw information, but this capacity is spread across millions of examples, making it statistically unlikely for any specific copyrighted content to be reproduced verbatim.

• The findings could significantly impact ongoing copyright lawsuits against AI companies, as the research demonstrates that LLMs primarily learn patterns rather than store exact copies, potentially strengthening fair use defenses for model developers.

🗓️ xAI is upgrading Grok's upcoming Tasks feature with a new left sidebar interface and enhanced scheduling controls, positioning itself to compete with similar automation capabilities being developed across major AI platforms.

• The Tasks feature allows users to schedule prompts for later execution with granular notification controls (in-app, email, or both), representing a move toward agent-style workflows and autonomous AI operations.

• xAI is catching up to OpenAI's existing ChatGPT scheduling capabilities, while Google's Gemini team has been working on similar functionality for over 10 months without public release, and Perplexity also has task scheduling in development.

• The convergence of scheduling features across AI platforms signals growing demand for persistent, time-triggered workflows that enable users to automate AI execution without requiring live interaction.

More news you might find interesting:

• OpenAI's o3 reasoning model has achieved A+ to B grades on University of Maryland law school final exams, representing a dramatic improvement from earlier ChatGPT versions that scored mediocre C's and D's in 2022-2023 studies.

• A new leaderboard platform LMArena.ai has launched to help users navigate the crowded AI image generation market by ranking models through community voting, with OpenAI's GPT Image 1 currently leading the pack.

• Apple's iPhone 16e has cracked Europe's top 10 smartphone rankings despite significant pricing backlash, landing in ninth place overall and capturing 2% of the total European smartphone market in March 2025.

• Meta CEO Mark Zuckerberg predicts that most physical objects will be replaced by holographic versions viewed through AR glasses, with 1-2 billion people potentially ditching smartphones for AR glasses within four years.

• New research from PwC's 2025 Global AI Jobs Barometer challenges fears about AI's workplace impact, finding that artificial intelligence actually makes workers "more valuable, not less" while boosting both employment and wages across AI-exposed occupations.

• New books by tech journalist Karen Hao and linguist Emily M. Bender reveal that the AI industry's foundation may be built on "puffery" and potentially fraudulent claims about large language models' actual capabilities.

• Samsung has partnered with ad tech company Glance to introduce AI-generated lock screen shopping ads that create personalized fashion images using your selfie and body type details, rolling out to Galaxy S22-S25 phones this month.

• Databricks Ventures and Noma Security have partnered to tackle the critical AI inference security gap that's preventing enterprise-scale deployments, backed by Noma's fresh $32 million Series A funding round.

Have any feedback? Send us an email