Agree. Well LLM is inherent to make possible an interaction with the AI and asociate machine learning. But certainly it isn’t the same using a biased chatbot with an LLM to analyze, interprete and summarize a given text or data set. Bad is when the generative AI invent datas when it don’t figure in his lenguage model causing this slops.
This often ocurre in alround AI which pretend to do everything (text, image,…), not so in AI specialized in certain tasks. Like a swiss army knife never can substitute a normal knife suissors, saw, screwdriver…, it will always only be an reduced emergency solution.
That’s a very exotic take. Language (semantic embeddings) is a good mediator for a lot of tasks, which makes it easier for ML models to generalize. The intersection of old school ML and LLMs is a fascinating place, I’ve never seen it dismissed like that. Do you have any reading you’d recommend on the subject?
Agree. Well LLM is inherent to make possible an interaction with the AI and asociate machine learning. But certainly it isn’t the same using a biased chatbot with an LLM to analyze, interprete and summarize a given text or data set. Bad is when the generative AI invent datas when it don’t figure in his lenguage model causing this slops. This often ocurre in alround AI which pretend to do everything (text, image,…), not so in AI specialized in certain tasks. Like a swiss army knife never can substitute a normal knife suissors, saw, screwdriver…, it will always only be an reduced emergency solution.
You don’t need an LLM to interact with machine learning algorithms.
In fact, one might argue one should not use an LLM to interact with machine learning algorithms. 😆
That’s a very exotic take. Language (semantic embeddings) is a good mediator for a lot of tasks, which makes it easier for ML models to generalize. The intersection of old school ML and LLMs is a fascinating place, I’ve never seen it dismissed like that. Do you have any reading you’d recommend on the subject?