The new global study, in partnership with The Upwork Research Institute, interviewed 2,500 global C-suite executives, full-time employees and freelancers. Results show that the optimistic expectations about AI’s impact are not aligning with the reality faced by many employees. The study identifies a disconnect between the high expectations of managers and the actual experiences of employees using AI.
Despite 96% of C-suite executives expecting AI to boost productivity, the study reveals that, 77% of employees using AI say it has added to their workload and created challenges in achieving the expected productivity gains. Not only is AI increasing the workloads of full-time employees, it’s hampering productivity and contributing to employee burnout.
Yeah, for sure since programming is also a language. But IMHO, for a machine learning model the best way to approach it is not as a natural language but rather as its AST/machine representation and not the text token. That way the model not only understands the token pattern but also the structure since most programming languages are well defined.
Note that, even if we refer to Java, Python, Rust etc. by the same word “language” as we refer to Mandarin, English, Spanish etc., they’re apples and oranges - one set is unlike the other, even if both have some similarities.
That’s relevant here, for two major reasons:
- The best approach to handle one is not the best to handle the other.
- LLMs aren’t useful for both tasks (translating and programming) because both involve “languages”, but because LLMs are good to retrieve information. As such you should see the same benefit even for tasks not involving either programming languages or human languages.
Regarding the first point, I’ll give you an example. You suggested abstract syntax trees for the internal representation of programming code, right? That might work really well for programming, dunno, but for human languages I bet that it would be worse than the current approach. That’s because, for human languages, what matters the most are the semantic and pragmatic layers, and those are a mess - with the meaning of each word in a given utterance being dictated by the other words there.
Yeah, that’s my point ma dude. The current LLM tasks are ill suited for programming, the only reason it works is sheer coincidence (alright, maybe not sheer coincidence, I know its all statistics and so on). The better approach to make LLM for programming is a model that can transform/“translate” a natural language that humans use to AST, the language that computers use but still close to human language. But the problem is that to do such tasks, LLM needs to actually have an understanding of concepts from the natural language which is debatable at best.
