// Today in AI

Today in AI: Infrastructure Debates and Integration Surges

The conversation today splits between high-level criticism of AI's inputs and practical leaps in model accessibility. While authors question the data quali

LDLatentDaily Desk Jun 28, 2026 2 min read

Today in AI — Photo by Kindel Media on Pexels

June 28, 2026

The conversation today splits between high-level criticism of AI's inputs and practical leaps in model accessibility. While authors question the data quality fueling these systems, developers are rapidly building better tools to connect to them.

🛠️ Claude Code now integrates over 100 open models

hf-claude now lets developers access a massive library of open models, including GLM-5.2, MiniMax-M3, and DeepSeek V4 Pro, directly within Claude Code. This is a major move towards a unified development environment where the best model for the job is just a function call away, accelerating the practical, multi-model future of AI. (@_akhaliq)

🛠️ Show HN: Smart model routing inside Claude, Cursor

A new tool showcased on Hacker News enables intelligent model routing directly within popular IDEs like Claude Code and Cursor. This points to a growing trend of abstraction layers that handle model selection for developers, making sophisticated AI workflows accessible without constant manual switching. (Hacker News)

⚖️ Margaret Atwood slams AI as 'garbage in, garbage out'

The renowned author bluntly criticized AI's fundamental flaw: its output is only as good as its training data. Coming from a literary icon, this cuts through technical hype and underscores the cultural fear that AI is merely systematizing and amplifying the noise of the internet. (The Verge)

🔬 New VISReg technique aims to improve JEPA training

Researchers introduced Variance-Invariance-Sketching Regularization (VISReg), a new method designed to improve the training of Joint-Embedding Predictive Architectures (JEPAs). This is a technical but important step for the self-supervised learning approaches that Yann LeCun champions as the path beyond today's generative models. (@_akhaliq)

💰 Data centers face growing political and public backlash

Hacker News discussions highlight a voter backlash against data centers, while TechCrunch notes skepticism towards Elon Musk's orbital data center concept. The physical and energy footprint of AI is becoming a tangible political issue, not just a theoretical concern. (Hacker News)

⚖️ Author alleges Meta conducted year-long surveillance campaign

The author of 'Careless People' claims Meta surveilled her for 12 months to enforce a non-disclosure agreement. If true, this represents an extreme and chilling corporate tactic that goes far beyond typical legal pressure, raising serious questions about power and silence in the tech industry. (Hacker News)

The takeaway: **The tools for a multi-model, IDE-integrated AI workflow are arriving fast, even as foundational questions about data quality and industry power grow louder.**