Its a buzz term they’ve been using to obfuscate the fact that every piece of LLM slop needs a human to review it, thus costing more time and money than hiring a competent human straight out.
I have to use them for work. Even the cloud hosted ones (which are arguably better than self hosted as they can have more resources thrown at them) need all of their output reviewed. They just aren’t up to the level of code generation as a senior dev yet (and if you think they are, I feel sorry for how unskilled your team is).
Locally hosted LLMs are a better option, but it doesn’t change the fact that the technology is designed outright to give a statistically likely result (based on the training data), instead of a correct one. LLMs make “guesses” about what the output should be and are not capable of internal regulation. They do not understand concepts and merely regurgitate jumbled data back at you with no consideration for content beyond that what can be analyzed statistically. It’s not a question of if they will make a mistake, but where and when.
https://youtube.com/shorts/ZMP8_jD-y0s These kinds of responses are an inherent feature of how they function, they cannot be fixed or tweaked out of the system.
Its a buzz term they’ve been using to obfuscate the fact that every piece of LLM slop needs a human to review it, thus costing more time and money than hiring a competent human straight out.
i think if they used a strong model for local ai it wouldn’t need much reviewing if at all. but it would need a beefy computer.
I have to use them for work. Even the cloud hosted ones (which are arguably better than self hosted as they can have more resources thrown at them) need all of their output reviewed. They just aren’t up to the level of code generation as a senior dev yet (and if you think they are, I feel sorry for how unskilled your team is).
Locally hosted LLMs are a better option, but it doesn’t change the fact that the technology is designed outright to give a statistically likely result (based on the training data), instead of a correct one. LLMs make “guesses” about what the output should be and are not capable of internal regulation. They do not understand concepts and merely regurgitate jumbled data back at you with no consideration for content beyond that what can be analyzed statistically. It’s not a question of if they will make a mistake, but where and when.
https://youtube.com/shorts/ZMP8_jD-y0s These kinds of responses are an inherent feature of how they function, they cannot be fixed or tweaked out of the system.