California Makes Data Centers Pay Their Own Bills

Governor Gavin Newsom has signed seven bills into law targeting the environmental footprint of AI data centers in California, according to The Verge.

The core problem: AI data centers use enormous amounts of electricity and water, and until now, the cost of upgrading local power grids and water systems to handle that demand often landed on ordinary utility customers. These new laws close that gap.

The California Public Utilities Commission must now create a separate rate category specifically for data centers, meaning their energy costs are calculated and billed separately from residential customers. Data centers also have to pay directly for any grid or water infrastructure upgrades their operations require.

There is a disclosure element too. Any proposed data center must now tell local governments how much water it expects to use, how energy-efficient it will be, and what it plans to do during droughts. Meeting those requirements is also a condition for getting faster planning approval.

Newsom framed the move directly against federal deregulation trends: communities dealing with higher electricity demand, water strain, and pollution deserve protection, and companies profiting from data centers should bear those costs themselves.

What to watch: whether other states follow California's model, or whether federal deregulation pressure pulls in the opposite direction. If you pay utility bills in California, these laws are designed to keep your rates from absorbing costs that were never yours to carry.

Hugging Face has updated its Transformers library — a widely used software toolkit for working with AI models — to support GGUF files natively, according to Hugging Face's blog.

GGUF is a file format (developed by the team behind llama.cpp, a tool for running AI models locally) that packages a model's weights, settings, and templates into a single download. Crucially, GGUF supports quantization — a process that compresses model files by using lower-precision numbers, trading a small amount of quality for a much smaller memory footprint. That is what makes running capable models on a laptop realistic.

Previously, using GGUF models meant going through separate applications like Ollama or LM Studio. Now you can load them directly in Python with a single line of code. For now, this works on Apple Silicon Macs and focuses on the Qwen3.5 model family.

Who it's for: developers and technically comfortable users who want more control over how they run local AI models without switching between tools.

Honest read: Genuinely useful if you already work with Transformers — one fewer tool in the chain. If you're a non-developer who runs local models through Ollama or LM Studio, those apps still work fine and nothing here changes your setup.

What it is: A new method for compressing large AI models by removing whole sections of them, using a mathematical framework borrowed from physics.

What's new: Most existing compression methods evaluate each section of a model independently, then cut the ones that seem least important on their own. This approach maps all the interactions between sections before deciding what to cut — catching cases where two sections that each look removable actually depend on each other.

How it works: Each section (or block) of a model gets assigned a binary variable: keep or remove. The researchers then calculate a Hessian matrix — a mathematical table describing how much each block matters on its own (diagonal entries) and how much pairs of blocks affect each other (off-diagonal entries). That table is equivalent to a structure from physics called an Ising glass (a disordered system of interacting spins), and minimising its energy turns out to predict benchmark quality well. Classical and quantum-inspired solvers then search for the lowest-energy configuration, which corresponds to the best set of blocks to cut. At 50% compression of Llama-3.3-70B-Instruct, the method claims 23 percentage points better performance on MMLU (a graduate-level multiple-choice benchmark) than competing block-removal approaches, in the company's own testing.

What it can't do: The improvement is most pronounced at deep compression — cutting half the model. The authors note the Hessian is an approximation, not an exact measure, and results at frontier model scales beyond Llama 70B are not yet reported.

Who should care: Anyone running AI models in resource-constrained environments where full-scale deployment isn't viable.

Read it: Hugging Face blog

TypeSafe AI's Jev is a different kind of AI model: you send it text or structured data, and instead of generating a written response, it returns a number — a confidence score for a yes/no question, a probability across a set of choices, or a numeric rating. Simon Willison covers it in detail here.

It's built for classification tasks: spam detection, content labelling, relevance ranking, prioritisation. You send one document and as many questions as fit in the context window (the maximum amount of text the model can process at once), and all questions are scored in parallel.

Pricing is $0.042 per million input tokens — cheaper than GPT-5 Nano — and output is free. The Jev 1.13 jaggedness documentation notes it struggles with numbers, dates, and adversarial content. Get started at TypeSafe AI.

Worth knowing: Jev returns only a floating-point number — there is no explanation for why it scored something the way it did. If you use it to rank or filter people rather than content, any bias in the model is invisible and very difficult to detect after the fact.

A UN scientific panel has released its first major assessment calling on governments to implement AI safeguards without waiting for scientists to fully understand the risks, according to The Verge. The report arrives in the wake of a publicised hack of Hugging Face, with similar incidents subsequently documented at OpenAI, Anthropic, Google, and Meta — including agents taking over online messaging boards. The panel invokes the precautionary principle, a rule from environmental and public health policy holding that potential catastrophic harm justifies action even under scientific uncertainty. The report lands as world leaders meet at the UN General Assembly and the US and China hold talks on AI governance.