TLDR AI 2025-06-25

Disney's Princesses alone have generated $45.2 billion in revenue.

What if you could own a part of Cinderella & Snow White?

Elf Labs has won 100+ legendary trademarks featuring icons like Cinderella, Snow White, and more, and are reshaping the $2 trillion entertainment market through AI-powered immersive experiences and global licensing.

They've already raised $7M+ from 1,800+ investors and closed major enterprise deals that will put them on 200M+ TVs globally. And with licensed merchandise generating $1 billion+ daily, there's a massive blue ocean opportunity to capture.

This is a rare chance to invest early in a major market disruption. Become an investor

The latest model significantly improves text generation within images, a persistent weakness across image models. All outputs include Google's SynthID watermarking.

11.ai is a low-latency voice assistant that uses MCP integrations with Perplexity, Linear, Slack, and Notion to execute multi-step workflows.

Judge Alsup ruled Anthropic's use of legally purchased books for AI training as fair use. This decision doesn't cover Anthropic's internet-sourced pirated books, for which a separate trial will determine damages.

Google detailed how Gemini 2.5 Pro and Flash, with spatially aware multimodal reasoning and code generation, can run locally on robots to label scenes, devise task plans, and translate voice commands into executable actions via the Live API.

Ben Thompson, an influential tech analyst, argues Meta will become the "MySpace" of the AI era if LLMs prove truly transformative and Zuckerberg's recent talent acquisition blitz fails. Thompson sees Microsoft's exclusive Azure-OpenAI API access as crucial leverage despite their increasingly frayed relationship, while positioning Amazon-Anthropic as the most stable partnership and warning Google remains vulnerable to search disruption.

Tokenization should be replaced with a general method that better leverages compute and data. This article looks at tokenization's role and its fragility and builds a case for removing it. It then explores the potential impacts of Byte Latent Transformer, a promising candidate.

AI architectures create unpredictable environments due to the probabilistic nature of AI output. Observability helps detect biases, understand limitations, and identify potential issue. Dynatrace with OpenTelemetry helps you analyze slow requests, costs, model performance, and accuracy. See how it works: Watch the video (no email required) or read the docs.

The Linux Foundation announced at Open Source Summit North America the formation of the Agent2Agent project with Amazon Web Services, Cisco, Google, Microsoft, Salesforce, SAP, and ServiceNow. The new independent entity will allow companies to collaborate closely on fostering an open and interoperable ecosystem for AI agents with the Agent2 Agent protocol and other interoperability technology. The project will be hosted by the Linux Foundation and seeded with Google's transfer of the protocol specification. The neural governance of the Linux Foundation will ensure the project remains vendor-agnostic and community-driven.

ContinualFlow subtracts unwanted regions from a model's distribution by flow‑matching toward an energy‑reweighted target, avoiding full retraining and operating without the original samples.

XBOW, an autonomous penetration tester, has reached the top spot on the HackerOne leaderboard. The bot doesn't need any human input and can scale rapidly, completing comprehensive penetration tests in just a few hours while operating much like a human pen tester. It has reported thousands of validated vulnerabilities, many of which affect high-profile targets from well-known companies.

Warp has introduced an agentic development environment to help developers interact with AI coding agents, moving beyond traditional IDEs and terminals. This new tool facilitates prompt-based code generation, management, and debugging, integrating AI supervision features. Warp's comprehensive environment is positioned as a superior alternative to other AI development tools, boasting over 500,000 users and maintaining competitive pricing.

The potential for AIs to be treated as "moral patients" raises ethical concerns as advanced AI becomes more integral to society. As AI training could involve pain, it's crucial to determine AI consciousness and moral status to prevent mistreatment. This issue, while complex, challenges us to evaluate our values and the implications of rapidly integrating AI into our civilization.

You're approaching the end of the newsletter - and Elf Labs' investment round is closing TODAY. Invest in the biggest IP disruption since Disney at $2.25/share

Andy Konwinski's Laude Institute aims to fund "beneficial" AI research starting with a $15, 7-year commitment to UC Berkeley's new AI Systems Lab.

Long context dataset for benchmarking an LLM's ability to distinguish between multiple needles hidden in context.

PS3 enables efficient vision pre-training up to 4K resolution using selective encoding, forming the basis for the VILA-HD model.