My 2 cents. Nothing more, nothing less. To put my opinion in context, I'm the co-founder of the HPP mod. I do work on the Chinese theater, starting techs, and some important events, like the Bitter Peace and Vichy's creation (the official 2.04 Vichy event is my work, and DI:G's Czechoslovakia events are also my work).
First off, after the hardware issues were fixed in 1.3, HOI3 was the game that I played the most. The previous versions would crash within 15 minutes of playtime, so I couldn't experience the supply and AI stupidness until 1.3 came out. ICE helped keep me occupied, but it didn't solve the fundamental problem with the game: the AI. To this day, ICE and DI:G still don't confront the AI aspect of the game, instead relying on Lothos's changes.
SF did a very good job when it came to the un-moddable military AI. The problem was that Lothos's AI changes were inadequate. The HOI3 AI is actually very, very, smart... if it has the right tools available. Lothos' AI, for example, tells the AI to research techs at random, with no overall strategy. It also has some bogus generic military builds, so you end up with Japan or the UK building nothing but planes and ships, while the USSR does nothing but construct buildings until 1944. The AI also isn't told when to stop training units, which results in 0-MP Germany during 1940.
Here's a perfect example of where the vanilla AI fails when compared to Slan's HPP AI: I'm playing as the RoC (nationalist China), and place garrisons in my major ports. 4 months pass after Japan attacks me, and Japan attempts to invade Qingdao. I have 2 garrison divisions in the city, plus urban terrain, so I should be fine. In vanilla, Japan's AI would leave after it lost the battle, but Slan's HPP AI is different - Japan's AI will land troops next to Qingdao while the city is being attacked, then flank it. You'll hardly ever see than in vanilla SF. Why? The HPP's AI tells the country to build more land units than it normally would, which results in a larger than normal army reserve. It also provides the AI an AI-specific port supply bonus, which encourages it to go that extra mile when attacking ports.
To put it simply: the AI will perform very well if told to build the tools it needs to succeed. However, if it doesn't have the right tools, it will disappoint. Currently everything is done based on dice rolls instead of logic.
If all of you reading this only want an AI that does what it's told, then stick with SF. If you want an AI that will change its strategy based on global events, then either play the HPP mod or tell PI that you want an AI that can actually think for itself. The military AI is actually pretty good, but the vanilla production and research AIs are not very good, and don't do the military AI justice. In other words, if you tell me "I can take that hill with 2 more bombers and 12 more mortars", and I instead tell you "do that with 10 more butter-knives instead of what you want", that sums up the vanilla AI quite well.
