Topic
12 facts from 9 episodes, in date order. These are what speakers said on the shows, each linked to the moment it was said. They aren't independently verified.
2024
Josh Albrecht says DeepSeek Coder significantly outperforms Llama models of the same size and some much larger ones, which he attributes to heavy training on code.
“I think that model still significantly outperforms Llama of a model that's the same size and even of of much larger models.”
Undated
Kawal Gandhi said open source models such as Llama and Stable Diffusion were coming to Google Cloud's platform.
“I'm really excited to see, like Llama, stable diffusion, uh, other variations coming onto the platform”
Undated
Gandhi said Google Cloud aims to make new models such as Llama available in its Model Garden so customers can use them within existing applications.
“can we make it available like Llama inside our model garden? And then they can use it inside their application”
Undated
Heller says CaseText had not found Meta's Llama 70B model able to outperform GPT-4 on its tasks, though it came close.
“we haven't seen, you know, uh, say, a Llama 70B outperform GPT four in a task, get close, but not outperform GPT four.”
Undated
At the time of the interview, Meta's Llama 400 billion parameter model was coming out, and Liu said open source models appeared to be catching up.
“obviously, you know, very timely, uh, you know, to discuss this with the the Llama 400 billion parameter model coming out”
Undated
Gomez says fine-tuning open models such as Llama is less effective than building a model yourself because you lack control over the training data.
“Taking those models and trying to, um, fine tune them, it's just it's not as effective as building it yourself, and you have much fewer levers to pull.”
Undated
Joelle Pineau says Meta released successive Llama language models and that each generation improved significantly, partly through community contributions.
“we release Llama one, Llama two, Llama three, and so on. And every generation gets significantly better.”
Undated
Daniel Whitenack says open models such as Llama can be run internally using tools like vLLM or Ollama, even on a laptop.
“So you can use VLM, or you can use Ollama, or or whatever it is, right? I can spin up one on my laptop”
Undated
Daniel Whitenack names Llama, DeepSeek and Qwen as examples of open models, and GPT and Anthropic's models as examples of closed models.
“There's closed models, there's open models, like closed models being the GPTs or Anthropic or et cetera, the open models being the you know, Llama or Deepseek or Quinn or whatever.”
Undated
Larridin trained a customized Llama model to block employee prompts that are illegal or that the company disallows.
“basically trained our own kind of customized Llama model to block people from asking questions that are illegal or the company doesn't want you to”
Undated
Chris Benson says part of LeCun's agreement to stay at Meta was that its models would be open sourced, which he gives as the reason the Llama family was open.
“And part of the the agreement on him staying at Meta, uh, during that time was that the models that they had would be open sourced.”
Undated
Chris Benson says Llama has increasingly trailed other frontier models in performance.
“Llama has started trailing farther and farther behind other frontier models in terms of performance.”
Episodes