TLDR AI 2025-06-02

You want AI to become part of your organizational DNA - and that means enabling every department to build out their own use cases, without IT gatekeepers standing in the way. But it shouldn't mean an ungoverned free-for-all.

Airia is the “let's get serious about AI adoption” platform. Rapidly prototype, deploy, and manage AI agents that transform workflows across your organization - without sacrificing security or governance.

Connect to dozens of enterprise applications with native integrations. Build agents quickly with templates and no-code tools. Let everyone build with AI while maintaining visibility and control.

Plans start at just $49/month. Get a demo

The Google AI Edge Gallery app allows users to find, download, and run compatible AI models from Hugging Face directly on their phones without an internet connection.

Annualized revenue jumped from $1B to $3B in the last five months, largely due to enterprise adoption of AI coding tools powered by Claude.

Samsung and Perplexity are nearing a deal that would put search technology from the artificial intelligence startup at the forefront of Samsung devices. The deal would mean Perplexity's app and assistant would be preloaded on upcoming Samsung devices and its search features will be integrated into the Samsung web browser. Perplexity's technology may be integrated into Samsung's Bixby virtual assistant. The deal will likely be announced as early as this year with the feature integrations starting from the Galaxy S26 phone line slated to launch in the first half of 2026.

Models that are Mixture-of-Experts with many layers, like DeepSeek, require large batch sizes and high latency otherwise throughput drops off a cliff. This is why it is commonly said that DeepSeek can't be easily run for personal use - with a single user running one inference at a time, it runs at a very low efficiency/throughput. This article explains this phenomenon in detail, as well as why some AI models are slow to respond but fast once they get going and the other effects of throughput, latency, and batch sizes.

This analysis by VC Mary Meeker documents unprecedented AI adoption rates, with ChatGPT achieving global penetration in three years compared to the internet's 23-year timeline. The report covers AI chatbots now being mistaken for humans 73% of the time (up from 50% six months ago), inference costs plunging 99% since 2022, and enterprise adoption accelerating beyond experimental phases.

This post looks at which third-party SDKs and API calls can be found in the four biggest Android AI chat apps: ChatGPT, Claude, Grok, and Perplexity. It looks at each app's development tools, business tools, product and marketing analytics, monetizations, and the API calls recorded by each app while open.

TextRegion generates text-aligned region tokens by pairing frozen image-text models with SAM2-generated segmentation masks, enabling zero-shot performance on detailed visual understanding tasks without training.

A hands-on course teaching how to build production-grade AI agents with LangGraph, RAG pipelines, memory layers, and backend deployment.

A study proposing a privacy framework that integrates variable trust levels among users into differential privacy models. It better reflects real-world data sharing preferences compared to binary trust assumptions.

Most people either skip system prompts or write short, unoptimized paragraphs, missing significant gains from personalized AI behavior. Users should systematically test and iterate on a system prompt instead of providing feedback on poor outputs in siloed chat threads. The comment section of this post includes a selection of community-provided system prompts.

Giving AI a stake in the future means respecting their autonomy and well-being and requires us to honor the contracts we make with them.

The DeepSeek team may have switched from training on synthetic OpenAI outputs to synthetic Gemini outputs.

The idea is to identify opportunities to buy mature, people-intensive outfits, help them scale through AI, and then use the improved margins to acquire other such enterprises and repeat the process.

ElevenLabs' Conversational AI 2.0 introduces a host of new features designed to create more natural, intelligent, and secure interactions for enterprise use cases.

Cerebras claims its systems outperform Nvidia's DGX B200 by achieving an output token speed of over 2, 500 tokens per second compared to Nvidia's 1,000 tokens per second.