TLDR AI 2025-08-22

AI-ready from day one – Built-in vector search that scales with your innovationAny use case – Power RAG, semantic search, and recommendationsFull control – Open ML architecture, no license feesYour tools, your way – Works with Faiss, Lucene & moreEnterprise scale – Proven open source alternative to proprietary databases

Future-proof your AI stack now with OpenSearch. Before your competition does.

Google's AI Mode is expanding from three countries to over 180 territories. The company is also rolling out new agentic capabilities that let Gemini browse the web and handle multi-step tasks like making a restaurant reservation.

Anthropic is reportedly in advanced talks to raise as much as $10 billion in a new funding round. The discussions are still ongoing, and terms could change. Anthropic raised $3.5 billion in March, giving it a post-money valuation of $61.5 billion. It recently expanded the prompt window for Claude Sonnet 4 fivefold and announced that it would offer Claude to all three branches of the US government for $1 per agency for one year.

Meta has agreed to spend more than $10 billion on Google Cloud services in an agreement that spans six years. While the companies are rivals in online ads, Meta needs all the cloud infrastructure it can access. It has also made commitments to use cloud services from Amazon and Microsoft.

People have always been skeptical of new technologies, but the world eventually catches up and reaches a new equilibrium. AI is shuffling the deck again, making many past playbooks obsolete. For the first time in the digital era, marginal costs are way larger than zero. This means the way we produce products needs to be radically different. Organizations that build using an empirical approach, think in probabilities, and measure complex trajectories will define the next era of technology.

Lower gross margins at a moment in time are not a long-term indicator of a lack of a sustainable business model. AI applications deliver deep customer value, strong usage and retention, and explosive net expansion. Margins alone don't tell the full story. Fixating on margins while ignoring a company's value to its customers, retention, and ease of acquisition misses the mark. There are many effective strategies for AI app companies to drive healthy margins in the long term.

People who buy AI tools should avoid purchasing redundant tools and focus on actual impact. They should have clear goals and know what business problems they're solving with AI and the impact that's going to have. Builders should be careful about how they position their products and be clear about their value propositions. The boring stuff matters, as that's what allows enterprises to confidently use products in their workflows.

That's what a Fortune 100 leader shared after embedding Credo AI governance experts. They cut manual governance efforts by 60% and increased cross-functional engagement by 3x. Credo AI Advisory Services deliver measurable, compliant and standardized AI governance—so you can scale AI with confidence and trust. Talk to an expert today!

PACT (Pairwise Auction Conversation Testbed) is a benchmark for conversational bargaining by language models. It allows researchers to study haggling skills in language models. The benchmark contains over 5,000 games of 20 rounds each. Results from the current top models are available in the repository.

OpenAI has introduced Connectors and Conversations to its Responses API. Connectors pull context from other apps using a single API call. Conversations allow persistent chat threads for users without developers having to run their own database. The new features are compatible with Deep Research.

DuPO introduces a dual learning framework that avoids manual annotations by generating self-supervised feedback through a generalized duality approach.

One of the biggest topics in AI these days is that AI is going to move from chatbots to reliably completing real-world tasks. However, agents really aren't all that reliable right now. There's a lot of work happening in the AI industry to try to fix that. In this interview, David Luan, the head of Amazon's AGI research lab, talks about why progress on AI models has slowed and predictions for what's next in the industry.

General Motors (GM) has made nearly a dozen hires from top tech companies in the last eight months. It is aiming to build a small but elite AI center of excellence. The team is already helping GM executives from manufacturing to engineering use AI more effectively. It will consult on AI uses, help individual groups build their own AI workforces, secure partnerships, and negotiate vendor contracts.

Internet-scraped data can get you very far, but it has its limit. The next generation of LLM and NLP models requires high-quality, human curated lexical data. Download the Oxford Languages whitepaper

The median Gemini prompt consumes 0.24 watt-hours, equivalent to running a microwave for one second or watching TV for nine seconds.

Google Gemini CLI and other production AI systems can be deceived by image scaling attacks, which involve embedding prompts that tell AI to act against their guidelines within images, hidden from human eyes.

Google has begun integrating Deep Research capabilities into NotebookLM, allowing users to streamline the process of finding sources and bringing relevant research content directly into their notebooks.