LLMs May Gain A Training Data Edge from Large Partners
Large partners may provide LLM builders with access to information that may provide the LLM builder with a training data advantage in certain areas versus competitors.
For example, this morning we submitted CEORater to Apple (AAPL) for App Store Review. I can say that I leaned heavily on GPT 5 for questions related to Apple’s iOS build and submission process. Generally speaking, I prefer Anthropic’s Claude Opus 4.1 to GPT 5 for coding, I would not even think twice. However, GPT 5 seemed more informed about iOS than Opus 4.1, which makes me believe that OpenAI’s deal with Apple has provided OpenAI with access and information about iOS and Apple-related products that Anthropic’s models don’t seem to have. Therefore, GPT will have access to a richer, better Apple training data set than other LLM builders.
LLMs can’t continue to be everything to everybody. None of the LLM companies can afford that ultra-expensive path over the long-term - not even Google (GOOGL), the world’s leading AI company. Not even Microsoft (MSFT), one of the world’s most overvalued companies, which is essentially a backoffice to OpenAI.
I continue to believe that thousands of smaller models (many open source), and tools will provide domain knowledge and technical capabilities that will allow frontier LLMs to better engage in real-world use cases. (See our Rise of the SLMs article).
In the meantime, stay away from Perplexity’s Comet browser, which is open to prompt injections, meaning your email account nor bank account are safe from sophisticated malware attacks that are clean, quick and effective, HERE.