Judge allows NAACP lawsuit against xAI over unpermitted data‑center turbines to proceed

Judge lets NAACP lawsuit over xAI data‑center emissions proceed

A federal judge denied xAI's motion to dismiss a lawsuit filed by the NAACP that alleges the company operated at least 35 combustion turbines at its Colossus data centre in Memphis without required permits^1,2. The suit, brought on behalf of the NAACP by the Southern Environmental Law Center, claims violations of the Clean Air Act. Judge Debra M. Brown ruled the plaintiffs have statutory standing and the turbines are not mobile sources under Mississippi law¹.

The complaint says xAI, now known as SpaceXSI, installed and ran the turbines without pre‑construction or operating air permits, breaching federal Clean‑air rules. xAI argued the turbines were mobile and thus exempt, and that the plaintiffs lacked standing, but the judge rejected both defenses^1,2. The Justice Department had also sought dismissal, but its request was denied, leaving the case to move forward².

Under the Clean Air Act, companies must obtain permits before building or operating major sources of air pollution such as gas turbines. Citizen‑suit provisions let groups like the NAACP sue to enforce these rules, a right the judge affirmed. The case reflects growing scrutiny of AI data‑centre projects, with several states recently considering moratoria on new facilities¹.

“another important victory in our effort to hold xAI accountable”

AI delivers zero financial payoff to most companies

Torsten Slok, chief economist at Apollo Global Management, reported that AI has not improved profit margins or revenue per worker for non‑tech firms in the S&P 493, a subset of the S&P 500 that excludes major tech companies. He said the data show flat profit margins and revenue per employee despite large AI investments. Slok expressed doubt that corporations can turn cheaper AI into higher growth or margins³.

Slok’s analysis focused on the S&P 493, removing Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla from the index. He found no upward trend in profit margins or revenue per worker for the remaining firms. He described AI as a “great calculator” but noted it lacks judgment, accountability, trust, culture, unwritten rules, ethical trade‑offs and fairness, which he says drive reputational risk and financial performance. Recent studies cited in the same note also found no significant productivity gains from AI in these sectors³.

The claim counters widespread hype that AI will boost productivity across all industries. Slok’s skepticism highlights the gap between AI capabilities as a data‑processing tool and the human judgment required for business decisions³.

I built an AI Tamagotchi before Meta could

Early adopters reported that the device could display avatars and connect to the Muse cloud but struggled with voice output and smart‑home discovery⁴. Business analysts note that, despite hype, Muse‑driven purchases remain modest, with most companies seeing only “hundreds” of orders or “very, very small” activity⁵.

The hardware is a Waveshare ESP32‑S3 1.75‑inch touchscreen pendant sourced from China and officially sanctioned by Meta as a “Muse Gadget”. Users must pair the device with the Muse phone app and provide Meta’s API keys, but initial interactions often misinterpret commands as transcription requests, leading to slow, text‑only replies. Meta’s engineers Anant Narayanan and Jeremy Payne supplied a sample app for video streaming, yet the lack of a remote shutter button limited practical photo use⁴.

Muse is Meta’s agentic AI platform that lets users automate tasks through voice or text commands. The Muse Charm is one of the first consumer‑facing devices built on open‑source code and tokens provided by Meta, aiming to bring the AI assistant into a wearable form factor⁴. While the concept mirrors earlier virtual pets, the device’s limited functionality and slow responses have kept it from achieving broad adoption, mirroring broader industry skepticism about AI agents’ immediate impact on commerce⁵.

Uncut offers a 7 MB lossless coding‑agent harness

The tool runs as a single statically linked Go binary on Linux, macOS, and Windows, supporting amd64 and arm64 architectures. It deliberately omits a text user interface, telemetry, plugins, background agents, automatic retries, and context compaction, sending the full conversation on every model call⁶.

Uncut’s design ensures that no part of a model’s output is silently truncated; if a provider cuts a response, the program exits with an error instead of returning a partial answer. All file contents and combined command output are preserved, and optional append‑only JSONL sessions record every message and tool result for resumability. The harness works with any OpenAI‑compatible Chat Completions endpoint, including local servers, and emits structured NDJSON events via the ‑‑json flag while defaulting to plain stdout for final text⁶.

Uncut counters this by being “lossless by default,” avoiding summarization, pruning, sliding windows, or silent output caps. Its small footprint and lack of runtime dependencies make it easy to embed in CI pipelines or container images without adding overhead or privacy‑risking telemetry⁶.