TLDR AI 2026-05-28

Build strong data foundations for agentic AI at scale. Register now

Explore advice from experts:

• James Jay Kulesa, Head, Innovation & Design, Regeneron
• Sush Sreevathsa, Data & Analytics Product Leader, Yahoo
• Muhammad N. Nasir, Global Product Management Lead, Mercedes-Benz AG
• Jeff Falcon, Principal, AWS Marketplace, Center of Excellence
• David Potes, Partner Solutions Architect Leader, AWS

Hear their perspectives on database strategies, governance, AI agents, and fast AI implementation. Register now

Cognition raised over $1B at a $26B valuation, with significant backing from major investors to expand Devin, an AI software engineer. Devin has significantly cut project times and improved automation for clients like Mercedes-Benz and Itaú. Cognition aims to further streamline software development by matching models to tasks while expanding its engineering capabilities.

ElevenLabs has released Music v2, a music-generation model capable of switching genres mid-track while maintaining vocal and compositional coherence.

Biohub has made its open discovery engine for protein structure prediction, design, and biological discovery available to researchers everywhere. The release includes ESMC, a state-of-the-art language model that has internalized the fundamental properties that govern protein biology; ESMFold2, a design engine designed to transform ESMC's sequence representations into atomically-resolved 3D structures of biomolecular complexes; and ESM Atlas, which makes ESMC's representations navigable across 6.8 billion protein sequences and 1.1 billion predicted structures. All three models are freely available to the global scientific community.

Both Anthropic and OpenAI have started aggressively pricing their APIs. This is likely because they have found product-market fit with coding/general-purpose agent products. Companies spending over $200 per month per user helps these businesses cover their costs much better than charging $10 to $20 per month per user. Coding agents amplify this spending significantly.

Real-world systems often behave differently in production than they do in the lab. Teams often discover these failures after launch, then spend weeks fixing them. That feedback loop is slow and manual. Today, it is possible to build agents that self-improve. This post looks at how OpenAI used Codex to build this type of agent at Thrive Holdings, resulting in an AI that can prepare increasingly complex tax returns.

The blog post introduces a method to reduce the weight synchronization payload in async RL using "Delta Weight Sync," which transmits only changed model parameters between RL steps, significantly reducing data transfer from gigabytes to megabytes. A Hugging Face Hub "bucket" manages high-frequency object storage, enabling separate locations for the trainer and inference engine without direct communication, leading to substantial bandwidth savings.

Learn on how to build strong data foundations for agentic AI in this executive panel[IS1] . Get strategies for databases, governance, and AI implementation from experts at Yahoo, Mercedes-Benz, Regeneron, and Amazon Web Services (AWS). Register now

Secure MCP Tunnel enables connecting private MCP servers to OpenAI products without exposing them to the internet. It uses tunnel-client to establish outbound HTTPS paths for request handling while maintaining server privacy. The solution integrates easily with existing systems, supporting enterprise networking requirements and maintaining secure data flow.

LiteParse is a standalone OSS PDF parsing tool that provides high-quality spatial text parsing with bounding boxes without proprietary LLM features or cloud dependencies. It features fast text parsing, screenshot generation, and support for multiple languages, platforms, and output formats. Everything runs locally on users' machines.

NVIDIA's LocateAnything is a vision-language grounding framework that decodes bounding boxes in parallel rather than token-by-token.

Apex is a React Native coding model trained to build apps by analyzing architecture decisions, fixing framework-specific issues, and reasoning about constraints. While it doesn't match frontier models on coding benchmarks, the optimized model significantly alters the performance-to-cost ratio within its specific domain. The model is still in development. It is now available in a private beta with selected teams.

Nvidia will invest $150 billion a year to make sure that Taiwan remains at the epicenter of the AI revolution. The investment is aimed at cementing Taiwan as the world's tech manufacturing hub for a long time. Nvidia will create a new headquarters in Taiwan to expand its partnership with TSMC, benefit from close proximity to advanced packaging technology not yet available at TSMC's factories in the US, and boost its alliances with other nearby partners. Expanding the AI ecosystem helps Nvidia further its bottom line.

YouTube said it will automatically apply labels to videos containing significant photorealistic AI content, reducing reliance on creator self-disclosure.

Join this panel with enterprise leaders from Yahoo, Mercedes-Benz, Regeneron, and Amazon Web Services (AWS) to see how to build strong data foundations for AI. Register now

The new Palo Alto-based startup aims to make AI systems that can actually see and interpret the physical world.

Ramp pointed roughly 10,000 Inspect coding-agent sessions at its backend in an 8-hour run with a minimal "find security issues" prompt.

Google DeepMind CEO Demis Hassabis now predicts AGI could be achieved by 2029-30, accelerating from his earlier estimate of 2030-2035.

Epicure is a family of three sibling skip-gram ingredient embeddings retrained from scratch on a multilingual recipe corpus.

Google enhances Gemini for Business with the introduction of shareable Projects, which allow team members to collaborate within dedicated multi-surface workspaces.

TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more.

Anthropic plans to expand Claude's voice mode to 18 new languages.