GEN AI is slop. Uses too much water as well. Milo can stay dead. 

Someone has to say it so here goes.....

HALFLIFE 3

Though extremely unlikely, I still hold on to hope that Steambot Chronicles 2 can return in some form, as the original remains my favourite game.
Cancelled in the wake of the 2011 tsunami, its rights-holders don't seem interested in doing anything with the property (a Reddit post recently mentioned that they were open to licensing), and director Kazuma Kujo only brings it up infrequently. Unfortunate, since it's one of the most unique gaming experiences I've had, even now, and nothing else really sates that particular desire.

I would pay insane amounts of money for Star Wars: The Force Unleashed III and Star Wars 1313 to be revived.

Also, ZombiU 2. I LOVED the first one and there was so much potential for a bigger and better game. Too bad that now it would be impossible to replicate the Wii U Gamepad gameplay without some serious compromises.

I was recently thinking about the Project Milo demo because of the Black Mirror Episode "Plaything" that came out earlier this year. It features a game designer directly based on Peter Molyneux who developed a game in the 1990s with a colony of AI characters. Not too unlike his earlier games. Over the decades, the AI characters advanced and expanded, and someone who was reviewing the game basically lived to upgrade his computer setup, and therefore the environment of these AI characters, and

I think the term "AI" is getting conflated with LLMs, but the LLM AI model is just the most recent big thing. What many people have forgotten, or didn't know to begin with, is that we've had AI dating back to the 1950s. Those early models were designed to play chess. AI has been a part of video games since the beginning, it just used different ways to mimic intelligence and in much smaller and finite contexts than what LLMs are doing today. But even before LLMs, we had quite open AI that would learn based on the expansion of the WWW - About 15 years ago, Watson became a big deal in AI, particularly when he soundly beat the top Jeopardy players. Now, someone who might know more than me can correct me, but I believe Watson was powered by a Supercomputer or network of computers that interpreted the set language rules of English, and used keywords (including phrases) to search basically millions of documents from wikipedia and other Encyclopedias, then picked out thousands of possible answers, and eliminated the least likely ones until it determined an answer... with greater accuracy and speed than Ken Jennings. Basically a much more expensive way to do what LLMs, and with a tailored/more limited functional context.

Current LLM models work on input data → converts data to tokens → It uses the patterns encoded in its trained parameters to score each possible next token and pick the one most likely to follow → Repeat until output is complete → Output of the predicted outcome data. So this can function like someone asking and answering a question where the question is the input and the answer is the output of how they'd be predicted to answer based on the training.

But to get back on earlier AI models, they were numerous, but most revolved around scripted rules and behaviour trees, motivational patterns, and pathfinding algorithms. No training is involved, but earlier AIs could adjust their behaviour based on scripted rules. The context of a video game is finite, so this is sufficient for that. Something like Project Milo would require more sophisticated AI, especially if Molyneux wanted it to sore. So, what was showed 15 years ago would be far more limited than what would be required. And yes, while Project Milo wasn't a "game" as in, something you can beat, sandbox simulations are widely considered games these days, and often the gamification of them ruins the initial concept - SPORE was a great example where the developers chickened out... likely from EA pressure. SPORE was initially marketed as a massive sandbox crossing eras of evolution, civilization, and space exploration with emergent goals rather than traditional game-based ones. People like Peter Molyneux and Will Wright are visionaries, and sometimes they don't develop commercially viable projects → Will Wright made the predecessor to SPORE, the James Lovelock Gaia sandbox simulator SimEarth, which I think only me and maybe 5-15,000 other people on the planet enjoyed - and its more enjoyable to me than SPORE because (to this day) I love having it run like a slow cinema thing in the background by just putting in the parameters, and letting the sandbox run for hours, to see what kind of world it constructs. I appreciate it when these sorts of visionaries exist, and these sorts of projects get done. That's also why I'm a big fan of the Switch EShop and Steam, because a lot of these creative "outside the box" games make their ways to those platforms.

I still think the gold standard of sandbox simulations is Dwarf Fortress, and it has been since like the era of Sims 2 or something.

Too much water? I know it uses too much electricity but water?

Can they? Yes they can
Do they? Doubtful

Those data centers yes. Servers need to be cooled down. AC's use water.

https://www.yahoo.com/news/articles/much-water-data-centers-prepared-202544298.html

Just another reason for us to hate AI.

Do you think love can bloom, even on the battlefield?