This, and similar stories at Anthropic, should remind us that LLM is a sorcery tech that we don't understand at all.
- First, deep-learning networks are poorly understood. It is actually a field of research to figure out how they work.
- Second, it came as a surprise that using transformers at scale would end up with interesting conversational engines (called LLM). _It was not planned at all_.
Now that some people raised VC money around the tech, they want you to think that LLMs are smart beasts (they are not) and that we know what LLMs are doing (we don't). Deploying LLMs is all about tweaking and measuring the output. There is no exact science about predicting output. Proof: change the model and your LLM workflow behaves completely differently and in an unpredictable way.
Because of this, I personally side with Yann Le Cun in believing that LLM is not a path to AGI. We will see LLM used in user-assisting tech or automation of non-critical tasks, sometimes with questionable RoI -- but not more.
- First, deep-learning networks are poorly understood. It is actually a field of research to figure out how they work. - Second, it came as a surprise that using transformers at scale would end up with interesting conversational engines (called LLM). _It was not planned at all_.
Now that some people raised VC money around the tech, they want you to think that LLMs are smart beasts (they are not) and that we know what LLMs are doing (we don't). Deploying LLMs is all about tweaking and measuring the output. There is no exact science about predicting output. Proof: change the model and your LLM workflow behaves completely differently and in an unpredictable way.
Because of this, I personally side with Yann Le Cun in believing that LLM is not a path to AGI. We will see LLM used in user-assisting tech or automation of non-critical tasks, sometimes with questionable RoI -- but not more.