Much has changed since the introduction of ChatGPT in November 2022. It is, in fact, difficult to keep up with the relentless stream of news regarding new AI-based tools. Each week we are informed of new developments, each of them (purportedly) groundbreaking and unprecedented – at least if we choose to believe the mainstream press and the marketing efforts of major software vendors. To recall, the introduction of DeepSeek’s R1 model in the end of January sparked a major sell-off in NVIDIA stock and marked China’s arrival on the international LLM arena. Maybe it is unnecessary to rely on big US-based software vendors after all; maybe large investments in hardware are not justified. Leaving aside questions of intellectual property and AI regulation, we must ask: which of the LLM-based products are of direct relevance to the practicing lawyer? Which can be useful in the performance of everyday tasks? Even the less technically-inclined lawyers face the prospect of being left behind unless they catch-up with developments in this field. Some LLM-based tools can, indeed, result in efficiency gains. Others are outright dangerous to use – unless their users exercise extreme caution and verify each generated output. Is it more efficient to expend resources to do things in a traditional manner or to deploy LLMs? As it turns out, everything depends on the task and the tool at hand. Legal research differs from legal reasoning and tools that are suitable for student papers may not be suitable in the performance of high-risk tasks. Lawyers must not only evaluate LLMs in light of potential efficiency gains but also heed privacy and confidentiality concerns.