I'd be surprised if FE is not the biggest impacted domain, followed by backend but backend still has a lot more risk despite it being more straight forward.

Asking it questions is incredibly fun. Not sure if I'm publishing this or not, it's just a fun side project, but I can ask it any question and I'll post it.

OMG, not this take again. You are not making anything if you just prompt the AI. It's the equivalent of going into a mine and finding a diamond and then saying you created the diamond. You did not create the beautiful rock that took millions of years to cristalize just like you did not create the beautiful rendered cutscene that took ungodly amounts of compute for training and inference.

Zuck's kid just mined the AI for a good output with text.

The distinction is important because there are and will be increasingly more artists that use AI "properly", i.e. integrate it into a larger pipeline and workflow, and actually produce art with intention and attention to detail, and not just a text prompt (as I assume an 8 year old would do).

Semantics aside, it's not hard to see how LLMs "hallucinate" (producing wrong results) if you see them as powerful pattern matchers instead of thinking machines because there is no intentionality in their output. Humans think things with intentionality, and meta-cognition. We arrive at conclusions and decisions with grounded intention and knowing about why we know what we know. LLMs simply pattern match and create an output that looks like they were doing this, but they're not.

You can see this if you try to have a normal human conversation with them. At first the illusion works, but very quickly it shatters as you realize this text you are reading has no idea about theory of mind that it could have picked up on had it been a mind reading your prompts.

In short, there needs to be intention and design, and not just exposing the developer tools, generally speaking.

If the game industry wasn't so myopic about recreating reality in graphics, the larger AAA titles could be exploring this more in depth. I think this would hit the jackpot of immersive experience as it could tap into the same psychological mechanisms.

If a game can afford a designers to really think what the experience should be, I think the gold standard should be no difficulty slider. Playing the game should be a self-consistent experience that gets more difficult as you get better at the game, and gating the final content with skills is very rewarding and makes the experience feel more "real" than if you just get a super configurable game where everything can be turned on/off.

You can do this with anything. Just change the definition and it can be anything you want.

The fact remains these models are obviously lacking in theory of mind, something which any consciouss entity with language should possess according to observation and intuitive consistency. How can a conscious entity communicate with language and at the same time have no way to project its mind and predict other minds. The simplest explanation is that it's just pattern matching language constructs on the surface level with no "pointers" to inner structure as is the case with humans.

Personally, I would focus everything on making changes easy to make at large scale. Typically with a human programmer this comes from designing and architecting and coming up with clever solutions that allow new features to be built faster even with a lot mroe lines of code. Maybe AI can do that, but I think it's uncharted territory because the biggest companies use AI with established codebases.

It sounds from the article the industry is still holding its breath to see just how much it will be able to do these harder tasks, but there are many reasons to think we have hit diminishing returns, and LLMs lack critical architecture to properly make decisions that matter without supervision.

This is true, but then there is "AI skill erosion" or whatever we may call it, where skills diminish in proportion to AI usage. At least that's been my experience. I have to still do manual code on the side to remain relatively sharp, which defeats the purpose of using AI to the extent I'm willing to sacrifice my own skills. If AI was consistent in its output, meaning we knew exactly where it fails and when, that could then become the focus. Without consistency one needs to remain sharp at every level, from the details of syntax to design.

That said, this tech will in no way give you anything remotely finished. It will at best be the beginning of a prototype which will require a lot of work to even have a proof of concept.

This is the same as with GenAI for images and video. Simply prompting doesn't get you much. You have to know how to use the tool to achieve a creative vision, and you need the vision to begin with. This idea that everyone will have a vision is vastly overestimated. Most people just want to consume and get entertained... and this is just like with anything else, most people don't want to grow their own food, or build their own house.