CityFibre, broadband and BT

If OpenAI and Anthropic were to withdraw their models, we would lose access because Microsoft would have to withdraw them from their servers.

Copilot would not be able to answer any questions, and our home-built agents would stop working, even if we just want them to answer questions about our own company data.

RAG doesn't work on its own.

It supplements the LLM with additional data, but the LLM is still required.

It's the LLM that knows how to communicate with us in natural language.

RAG just teaches it how to communicate with us about TM Forum stuff.

And then this last paragraph is: "However, if we were to take a more DIY approach, we could use alternative open models like Qwen, Llama, or Gemma." So the, the Japanese ones like-- I, I don't know if one of those is DeepSeek.

Um, maybe it is.

But they, they tend to be open source.

Um-

But that's, that's the, um...

That's exactly what, um, Pardeep Kohli was saying.

Right.

That, that was his argument for using open-

Okay

...

But the, I mean, it, it-

And let me just quickly finish before you say.

We could run these open models locally on our own servers, so even if they disappeared from the internet, we could still use them.

Yeah.

However, if a model would be withdrawn, we would not receive updates, so the model's knowledge of the wo- of world affairs would be frozen at the point in time it was withdrawn.

But I, I...

No, I was just gonna say, whether you're running it in the cloud or, or locally, you're still paying a license fee if it's a proprietary system that comes from somebody.

Yeah, absolutely.

So if there, if there's a legal enforcement that you can't use that anymore, it doesn't really matter where it's hosted.

If OpenAI and Anthropic were to withdraw their models, we would lose access because Microsoft would have to withdraw them from their servers.

Copilot would not be able to answer any questions, and our home-built agents would stop working, even if we just want them to answer questions about our own company data.

RAG doesn't work on its own.

It supplements the LLM with additional data, but the LLM is still required.

It's the LLM that knows how to communicate with us in natural language.

RAG just teaches it how to communicate with us about TM Forum stuff.

And then this last paragraph is: "However, if we were to take a more DIY approach, we could use alternative open models like Qwen, Llama, or Gemma." So the, the Japanese ones like-- I, I don't know if one of those is DeepSeek.

Um, maybe it is.

But they, they tend to be open source.

Um-

But that's, that's the, um...

That's exactly what, um, Pardeep Kohli was saying.

Right.

That, that was his argument for using open-

Okay

...

But the, I mean, it, it-

And let me just quickly finish before you say.

We could run these open models locally on our own servers, so even if they disappeared from the internet, we could still use them.

Yeah.

However, if a model would be withdrawn, we would not receive updates, so the model's knowledge of the wo- of world affairs would be frozen at the point in time it was withdrawn.

But I, I...

No, I was just gonna say, whether you're running it in the cloud or, or locally, you're still paying a license fee if it's a proprietary system that comes from somebody.

Yeah, absolutely.

So if there, if there's a legal enforcement that you can't use that anymore, it doesn't really matter where it's hosted.