The Mice-Principal

A janky guide to the janky world of uploading and running linux packages inside the #Claude web app for free users. You can also drop it in to train your claude and build small weird things that just about work. pdfhost.io/v/ZnFEmYJy24_c… #ai #manual

In case anyone like me had no clue, windows has live auto generated subtitles of any sound output if you press win + ctrl + L ... Aside from being helpful, it passes the time during football matches if all you can get is foreign commentary.

In memory of Shitbirds, discovered today whilst looking for something, an early nft collection saved from existence by "coming to my fucking senses".

I could have edited it better, but on the plus side, you can just drop it into a claude and it will gain the learned experience of about 60 wasted hours and then you can easily do all kinds of harmless weird.

And so it ends maxing out instance with Free-Tier Claude.ai stuck madly circling, unloading a shotgun into a corpse. If bored, here's how to jank Doom & things where it makes zero sense. I'm still trying to understand 1/2 of it pdfhost.io/v/ZnFEmYJy24_c… #ai #Claude

All credit to RHVoice for the managing to be possible to run in fairly absurd conditions... and well done ffmpeg for managing to be so cursed that you produced haunted .wav and .mp3 files that played differently every time.

Wonky as it is, as a proof of concept, it does show Claude seems pretty much perfect for accessibility. A decent reader plugin that was intelligent about its contents would be something actually helpful. Also... the comic pre-dates Claude... so... the egg came first Anthropic.

Made a janky tts in free-tier Claude.ai. Don't suppose @AnthropicAI fancy making a non-jank, comic/general #accessibility reader plugin? I get hallucinations are an issue, but audio disclaimers exist... plus, everyone turning into a sheep randomly livens things up

Estimated time to install, run and troubleshoot through menus on Doom on Chatgpt free-tier allowances: Some Time After Heat Death Of Universe.

Truly a massive milestone in Claude Plays Doom... #Claude has drawn First Blood... Accidentally. In the wrong level.

... seems like Chatgpt might want to play Doom too...

After many hours #Claude now reliably navigates the evil start menu and charges into battle... Unfortunately it charges in the style of a 90 year old with agoraphobia taking their first faltering steps outside in 60 years. #depressedpenguindoom

Attempting to teach Claude how to play Doom at least to the level of a 3 year old, get past the fact that for some reason the Claude container seems against this whole idea, and defeat the most deadly enemy of all: The Menu.

After two days of effort... Finally achieved near free-tier claude.ai account's file transfer size limit of gameplay capture without #Doom crashing. This milestone is marked by #Claude failing to leave the starting room but longer. #glacial #ai #fail #videos

Boredly experimenting adding missing accessibility features into Claude using random uploaded packages, jank a, 31mb upload limit and... suddenly remembering "...oh wait.. archive splitting's a thing"...

... Claude.ai has finally broken its obsession with the pipe and instead performs worse than when fish play pokemon. On the plus side free-tier chatbot's desire to troubleshoot its constantly crashing Doom installl remains positively giddy.

claude.ai's slow decline into #DOOM madness continues. It has declared war on a particular pipe and the increasingly determined belief it MUST turn left is hindering its k/d ratio. On the plus side it's now just adding more title cards on top of the old ones.

Considering Chatgpt recently gave me a universal llm bypass whilst I was simply trying to make it be less of a passive agressive douche... I'm not sure I believe that Altman guy fully on safety.

Despite current refinements in automation Freetierclaudedoom still seems to be uncertain around both corners, walls, floors, open spaces, and doors.