Comments by "Roni Levarez" (@ronilevarez901) on "Bloomberg Technology" channel.

  1. So many inaccuracies because she doesn't know what she's talking about. Small language models are smaller than 100m parameters (smaller than GPT2 small). Bigger than that they all are "large*. Llama 8b and such are small large language models, not small language models. True small language models are actually trained on small and specialized datasets, specific to the use case, like a chatbot trained on a dataset of only dentist appointments, to make it useful for a dentist office, but nothing else. That's how they cut off costs, so in the video they mix both things. You kinda did that too. And about the leading companies, she talks about companies making SLLMs, not state of the art LLMs.
    2
  2.  @NachodeGregorio No I don't mean Phi nor any other popular giant model. I mean specialized models you have no idea about since you only know about the trendy and hyped models that I'm sure you can run on your PC , but those models are not the only stuff under the sun rn. There's way more stuff being developed for other other target audience that doesn't have a use or even know about your giant LLMs. Now,, "Most people " are not specialized data scientists, which are the ones you should be citing instead. And since I already told you this but you still don't get it, I'll repeat it once and that's all: GPT2 is a large language model and it's way smaller than llama or any other current sota. Call them what you want. I'm sure the people you talk with will understand you just fine, but that won't change the formal definitions. Although, the use of the terms and the models by non academic people disturbs definitions very often.
    1
  3. They do mention them. But why would only the giants get more free advertising when there are other groups making progress?
    1