I find it concerning that IIUC, of the GPT-4 paper, system card and this paper, _none_ discuss the training data in detail. The paper discussed here even says

I find it concerning that IIUC, of the GPT-4 paper, system card and this paper, _none_ discuss the training data in detail. The paper discussed here even says (page 7): > The standard approach in machine learning is to evaluate the system on a set of standard benchmark datasets, ensuring that they are independent of the training data and that they cover a range of tasks and domains. This approach is designed to separate true learning from mere memorization, and is backed up by a rich theoretical framework ... However, this methodology is not necessarily suitable for studying GPT-4, for two reasons. First, since we do not have access to the full details of its vast training data, we have to assume that it has potentially seen every existing benchmark, or at least some similar data. I.e even these authors, who had a special degree of access to the model, aren't sure what's in the training data. So it does _great_ on a bunch of tasks, but we're not sure that it wasn't actually trained on very similar data. A recent discussion on coding performance that's in vs out of its training data seems quite telling. It does poorly on more recent problems of the same level of difficulty. https://news.ycombinator.com/item?id=35297067 https://news.ycombinator.com/item?id=35297067 The model itself may be very impressive, but so far the papers describing and analyzing it are super sketchy, and it's really hard to separate out reasoning capability from noisy memorization.