Will AI take your job?

He's a bluffer. But has deep pockets which means he can make lots of mistakes and get past them by throwing money at it. His main talent seems to at making money, and circumventing the usual rules. Again by throwing money at it.

It's not that good - yet. It understands language. It knows what you are asking. It's trying to answer you. It doesn't understand the difference between a correct or an incorrect answer - yet. The problem is being worked on, it's called AGI and predictions are it will be achieved sometime.within the next 5-8 yrs. However it is very good at straightforward requests in the realms of maths, physics, coding, accounting etc... People who know how to write the correct prompts to guide the AI will be the ones who keep their jobs ie spend 4 hrs tweaking the prompt and get a few weeks worth of output in seconds. Spend another few days sanity checking it and you have saved weeks of work. That is where this is going in the next 10 yrs.

AI seems to create answers it thinks you want, but actually isn't what you asked. 

If you mean hallucinations and them responding regardless, this is a fundamental facet of transformers, but the architectures and pipelines are getting better for this. More expressive models with chain-of-thought are far better for producing valid reasoning in my experience. RAGs help a lot also. They may still hallucinate but for my own work, this is not an issue. A model might get something wrong but I generally know and can refine the inputs. When I consider the scale of correct output that they offer, to the mistakes, the hallucinations are irrelevant to me.

I'm looking at how I can data cleanse our data and structure that data so its useful in a model

For training a model, finetuning, or referencing? I'm sure you know what you're doing but e.g. if you're allowed to finetune, consider taking a look at datasets that are known to be performant, and the structure of them. E.g., Open-Orca/OpenOrca · Datasets at Hugging Face. I.e., system prompt | question | response

Personally I'd use AI in a pipeline to get data ready for AI via agent workflow engines, with oversight obviously. If LLMs have one great strength it's in inferring and expressing linguistic structure, and text summarisation and similar tasks. This has been the case since BERT and even before. Small local models may handle simple sanitisation, depending on the task.

Not what I said at all. Bizarre response. Not sure why you directed that at me.

My response was for those who refuse to use AI because it doesn't do the exact thing they tried it for once. I literally said it is saving hours of time and will only get better over time..It isn't that great in some areas - yet. That's a fact. However it is very good at some tasks now. My team is actively using it. I have been rolling out AI to my group for the past 18 months. And contiuing to identigy use cases for it. It is not going to "take all jobs" as some.people.have claimed. it is going to change how people do their jobs. Some jobs will be obsolete in 10 yrs akin to how automationn replaced lab technicians in the 1970s who used to analyse blood under a microscope. Roles will.change and those who figure out how to use AI will still have their choice of roles.

Veo 3 is only out a few weeks and Joe Soaps are outputting things based on a few prompts that….minor issues aside with respect to how good the rest is….would take big budgets, or an impossible budget, and a lot of time to produce traditionally. Look at the state of the art from 6 months ago and compare to now.

Suno was also used to generate the music for that one ^

I got a feeling this thread will look fairly quaint in five years time.

I suspect this is world-changing technology.

Government strategy on AI was published a few days ago

While the output is just ok quality (more like in game cut scenes). the direction and cinematography is very poor. I'm not impressed. This why I'd suggest Joe soap isn't going to the anytime soon. I'd like see what a director could do with it. No doubting the tech is making progress.

Except we've got a long list of promises that he's made in relation to his own cars and he didn't get them right. So no, he's not the foremost expert on robotics or AI just because he's wealthy.

https://en.m.wikipedia.org/wiki/List_of_predictions_for_autonomous_Tesla_vehicles_by_Elon_Musk

Is that not a subjective argument, you could also argue that most "modern art" is crap and very few artists now could reproduce the art of the great masters of the 18th and 19th centuries?

The common defensive argument is that ,oh well the great master's work has already been done , don't need to keep doing that, but nobody bothers to attempt it now either. If AI can produce new "great masters" art well then that would be big. However I think AI would have much more success reproducing the works of Jeff Goons (dead animals in formaldehyde) rather than Rembrandt .

A certain amount is subjective. However a lot of AI output is lacking technical aspects, like framing for example.