Couldn't companies, which want to host models themselves out of privacy/IP theft

<!-- SC_OFF --><div class="md"><p>Hi,</p> <p>I’d like to practice answering scientific questions about a confidential project, and I'm considering using an LLM. As this is about a

<!-- SC_OFF --><div class="md"><p>Been noticing that once you use LLMs for real dev work, the cost conversation gets messy fast. It is not just raw API spend. It is retries, long

<!-- SC_OFF --><div class="md"><p>I was discussing with some other folks how safe is to use open weights models from China and the topic of &quot;trojan horse&quot; came up.</p>

<!-- SC_OFF --><div class="md"><p>I read this in a newsletter today and it raised a question I hadn’t really thought about before. If access to frontier AI models can be

<!-- SC_OFF --><div class="md"><p>AI models are really large and it's a real pain to download via hugging face only (and a major open source liability having just one source).</p>

<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1x5kpng/ai_models_can_provably_hide_capabilities/"> <img alt="AI models can provably hide capabilities:

<!-- SC_OFF --><div class="md"><p>for me it is simple now. anything sensitive stays local.</p> <p>before I was sending client documents through cloud APIs without thinking too