This topic is for the purpose of discussing #10655: adelikat & eien86's NES Lode Runner in 13:40.249

Nice improvement, yes vote!

Nice improvement, yes vote from me!

Sorry, what's the point of all the AI agent stuff in the submission text? Is any of it actually relevant to the submission? What are Judges supposed to glean from it? ...Also, are you aware that we've been discussing banning that sort of AI usage in submission texts and across the site?

Hey Samsara,

Sorry, what's the point of all the AI agent stuff in the submission text? Is any of it actually relevant to the submission? What are Judges supposed to glean from it?

It is a voluntary disclosure that I used an AI agent to help me with this work; I thought voters should be informed of this when giving their opinion. The help was mostly connecting the game to jaffar, finding RAM addresses, and implementing my "attack" ideas. The two links take you to my work log, which I store in Jaffar's corresponding game examples folder, in case anybody (or myself) want to resume the work. They detail my day-by-day work log, ideas that failed or worked, and some important game mechanics. I explained this in the (now cancelled) #10639: eien86's NES Bomberman in 29:32.581 Since these are written by AI, I did not include them in the sub notes -- simply put them as links. Feel free to remove them if these violate the current rules.

...Also, are you aware that we've been discussing banning that sort of AI usage in submission texts and across the site?

Yeah, I was aware of this thread but haven't followed the conversation too closely. Tbh, I'm fine with whatever you decide. I'll of course abide by the site's rules, as always.

I dislike the current state of AI industry and I support the TASVideos AI policy. I love the creative usage of AI, as exhibited here, the end result being an improvement to an outstanding TAS. A conundrum. Is it now an uncomfortable fact that by removing info about AI tools used for tool-assisted speedrunning: TASVideos is no longer a website that has the latest info about tool-assisted speedrunning? Instead, I have to ask eien86? Another conundrum. I believe most or all online communities face this same conundrum, sooner or later. Conundrum is not even the correct word for it, but I will not ask an LLM to fix it. You get the gist. Anyway, I vote yes! This TAS has the hyper speed element that makes a TAS a wonder. A spiritual successor to the hyper-speed Gradius TAS. As a manual taser, if I tried to replicate eien86's result, then I might use more electricity than what eien86 used here. This is a random guess because nobody has posted any real research data about it. If manual tasing is discovered to waste more electricity compared to some future AI-assisted tasing method, should we then discourage manual tasing? I don't know. Edit: More problems: Since information was removed from the submission text, I now get the false impression that Claude Code (Opus 4.8) may have generated some part of the TAS input. A conundrum again.

What was the improvement you found in level 1?

What was the improvement you found in level 1?

3 frames towards the end, but it wasn't exhaustive as the stage50 work

alex can you please not write scrambled messages please.

What was the improvement you found in level 1?

3 frames towards the end, but it wasn't exhaustive as the stage50 work

Its likely not a real improvement. Most levels delay frames in order to manipulate the optimal spawn timer in the next level or to get a more optimal select glitch I just want to say, this submission has me annoyed. I don't think this process shows an in depth understanding of how to optimize this game. There's some big potential improvements that would be better suited for a more exhaustive search, 4 levels reduce speed in order to do the select glitch, maybe this isn't necessary? Every level might have frames to shave, if they can be kept after spawn timer manipulations and select glitch optimizing. All this did was shave a few frames on the last level, the one you don't need to understand much about in order to shave some frames. You have a nice bot that can probably shave frames off everyone's last level. This rubs me the wrong way, and surely I wouldn't be the only author to feel this way. I would have been happy to collaborate on a lode runner effort but I didn't even know about this work until just now.

You can collaborate on it. The game is complex enough and will not run out of improvements any time soon, or ever. The game is kind of trolling tasers with the select glitch. You would save a lot of trouble by doing a 'no select glitch' category. Although, I understand the appeal of the challenge that the game presents. As for improving the last level of each published TAS with a bot, I think it would take one person at least 100 years of full-time work. It would be a very difficult task to decide when to stop botting any given game. We could allocate 10 days of botting per TAS and still fail to find an improvement in all games. No need to worry about that. Also, some games are computationally irreducible to the point that they can't be fully optimized at all, the Doom uv-max category being one example of that. (By the way, please collaborate on that and blow the minds of everyone.) The last level is usually more complex compared to earlier levels, so when resources for botting are limited it makes sense to try to improve the last level with a bot. The likelihood of finding a new glitch in the last level is also higher compared to botting the first level, if the game has progression towards complexity. This tendency may not apply to Lode Runner, but even then eien86's improvement seems like a genuinely effortful contribution, not comparable to something like removing 1 idle frame from the end of the run, which would be a TAS out of spite.

I’m with adelikat on this. This just feels disrespectful.

as eien said, fully optimizing the lode runner tas could take a year and its probably not worth their time. It makes sense to only improve the last stage to shave off some frame for now. The tas was improved by eien and he puts himself as co-author. I do get that adelikat is annoyed about somebody else improving their tas by frames while potentially ignoring other improvements. I would be annoyed too but improvement is improvement i guess... Other people have done the same and worse more than a handful of times from what i have seen. Take an existing tas and implement 1 "strat" in the middle or at the end of the tas just to shave off frames for the sake of publication instead of reanalyzing the game again.

I've shared similar sentiments as adelikat here when a TAS is either just shaving or porting input. At the same time if Jafar is being used in conjunction with AI I would expect much better than what's showcased, but we still saw an error from this combination previously, and knowing that Jafar is a brute forcer is it wrong to expect this to spend more time in the oven? This is probably apart of continued experiments which still needs to settle properly and understand what it's doing.

I’m with adelikat on this. This just feels disrespectful.

Thank you for your input, Walgrey. Could you please explain who or what was being disrespected here and how?

This is just a small improvement to adelikat's masterpiece

Calling something a masterpiece is generally regarded as a high form of praise, which is the opposite of disrespect.

om, nom, nom... crunchy!

Well, that's unfortunate. He was great at TASing, accepting runs, and overall a great guy despite having different views than the majority of the community.