This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
On Sunday, xAI released the weights and architecture of Grok-1, the large language model behind the Grok chatbot on X. It’s a 314-billion-parameter mixture-of-experts model, published under the Apache 2.0 license, which permits commercial use. As TechCrunch noted, the release doesn’t include training code. It is also a raw base checkpoint from pre-training that concluded in October 2023, not a version tuned for conversation.
Much of the reaction has focused on the rivalry between xAI and OpenAI. I’m more interested in what happened the same day. Perplexity’s CEO, Aravind Srinivas, posted that his company would fine-tune Grok for conversational search and make it available to Pro users.
It’s a small example of a pattern that will matter for reputation.
The model and the product are separating
When people think about how AI describes a company, they usually picture a branded product: ChatGPT, Gemini, Claude, Copilot. Each has an owner, a policy team and some kind of feedback process. If one of them says something false about you, there is at least a company to contact, and that company has its own reputation giving it a reason to care.
Open-weight models loosen that link. Meta’s Llama 2, Mistral’s models and, last month, Google’s Gemma are all available for others to download, adapt and deploy. Grok-1 now joins them. Once weights are public, a model can show up in products its creator never sees, with system prompts, retrieval sources and safety settings chosen by someone else.
Picture a wrapper app running a fine-tuned version of an open model, connected to a search index of unknown quality, presented under a third brand. If it gets your company’s history wrong, who owns that answer?
A cautious prediction
I don’t think this makes the big branded assistants irrelevant. Distribution still matters enormously, and most people will meet AI through products from large companies they already use.
But I’d expect a growing long tail of AI interfaces built on open models: customer service bots, vertical search tools, browser extensions, apps aimed at other languages and markets. Each will summarize companies and people, and many will draw on the same public sources: news coverage, Wikipedia, company websites, review sites.
That has two implications for communications teams.
First, monitoring can’t stop at one or two flagship chatbots. When Gemini was announced in December, the useful question was which big platform would summarize you first. Increasingly, the honest answer will be several, including some you’ve never heard of.
Second, the leverage point moves upstream. Nobody can file a correction with every fork of an open model. What a company can do is make sure the material those systems retrieve and learn from is accurate, current and consistent across the sources that matter most.
Control was always partial
Companies never fully controlled their narrative, and search engines and Wikipedia have shaped first impressions for a long time. What open weights add is that the summarizer itself becomes a commodity. The interpretation layer is now as distributed as the content it interprets.
The safest assumption is that your story will be retold by systems you will never audit. The best preparation is a public record that holds up when they do.