I mean YEAH; large language models have a few downsides. Massive resource and environmental concerns
| Source: Mastodon | Original article
A new report from the Nordic AI Impact Institute, released on 3 April, warns that the rapid expansion of large language models (LLMs) is creating a cascade of economic and ecological problems. By analysing the compute cycles required to train models of GPT‑4 scale, the institute estimates that a single training run emits roughly 1 000 tonnes of CO₂ – comparable to the annual output of a small city. The study adds that the electricity demand for inference at scale pushes data‑centre power consumption up by 15 % in the EU, while the scramble for high‑end GPUs is inflating prices for consumer‑grade chips, making premium laptops and tablets – including Apple’s newly refurbished M4 iPad Pro – increasingly unaffordable for the average user.
The findings matter because they intersect with three pressing policy and market concerns. First, Europe’s Green Deal targets a 55 % reduction in emissions by 2030, yet AI compute is on a trajectory that could erode those gains. Second, the cost surge for GPUs and specialized ASICs is crowding out startups and research groups that cannot shoulder multi‑million‑dollar training budgets, deepening the divide between tech giants and smaller innovators. Third, the report highlights the unreliability of current LLMs – frequent hallucinations and opaque decision‑making – which can translate into costly errors in sectors ranging from finance to autonomous driving, amplifying the risk of broader economic fallout.
As we noted on 4 April, criticism of LLMs’ sustainability was already surfacing in the tech community. The new data gives regulators a concrete basis for action. Watch for the European Commission’s forthcoming AI‑specific carbon‑labeling proposal, for industry shifts toward on‑device models such as Apple’s FoundationModels framework, and for emerging “green‑AI” benchmarks that could become a prerequisite for public‑sector contracts. The next few months will reveal whether environmental pressure can steer the LLM market toward more efficient, affordable, and trustworthy solutions.
Sources
Back to AIPULSEN