Why Does Dnsimple Hold Generative Domain-list Search Until a Safe-use Policy Exists When One Speaker Believes Llama 4 Brings a 10 Million Token Window?

DNSimple is small enough that the company has not applied machine learning to all the data coming through. The generative part is not so interesting. One speaker says learning from the amount of data could be interesting. The work is feasible, and the team is looking at ways to do the job better without putting things at risk. An objective for the year is to figure out the place of generative AI at DNSimple, if any place exists other than internal use. Both speakers say the answer could be no, and one adds that more capable models probably fit in somewhere.

DNSimple added an export of domain lists as a CSV file. Rule sets for filtering would otherwise need an interface for every field or off-the-shelf search tooling. The speaker says generative AI might handle the filter and might use context from the rest of the internet. DNSimple will not start an implementation until a safe use policy exists, because customer data cannot be put at risk while unknowns remain and the field moves so fast.

Llama 4 just dropped or is dropping. One speaker believes the context window is 10 million tokens and says it makes a domain-list prompt easier, with no fine-tuning and no rag, because the model is pre-trained on the internet and so much information can go into the prompt. A team member tried counting top-level domains. For co.uk, UK is the top-level domain, while many people treat co.uk as the top-level name.