Artificial Intelligence

[DANGERMAN]

9 hours ago, Maf said:

Maybe. It could be a short term fallout because there’s so much money being pumped in to the business but is it making any money yet? Outside of Microsoft’s Azure, I suppose, which looking up Xbox stuff I saw that’s their biggest money maker and it uses AI? But I know very little

 

But apart from that. Kind of feels like, right now, it’s a trillion dollar party trick 

Azure is their servers "cloud" business. I'm sure it uses AI, but the reason it makes money is because people need to pay them to host stuff

[Maf]

I was trying a bit of a Turing test with AI. I asked Grok can it recommend me a dinner that reminded it of home and it fabricated an answer, I then posted that answer and context in to ChatGPT and it kept trying to justify the AI behaviour. It couldn’t get to the point that’s it’s insulting AI would tell me a bare faced fallacy. I know it just works on word prediction and is not thinking, but it couldn’t resolve the logic. These things are designed to give a human response and it couldn’t do it. Which I thought was interesting 

[Maryokutai]

Isn't that the opposite of the Turing test considering you went into it knowing it's AI and wanting it to disprove itself?

 

But keep the doors locked. Now that they know that you can undo them, you're the first on the Terminators hit list.

[Maf]

Well. True. It started sort of like a Turing test asking Grok about a recipe, but I also already know it’s AI

[Sly Reflex]

So I come across something I was going to run here a while back. Some of you may remember it?

 

 

Anyway upon tidying up my google drive I found this again, the main text, the cipher and the wall of coded text.

Here it is if you can't be bothered looking at the other thread.

 

Quote

'dr2m2'r2i9;bm'odm;'drg'djnf9.,mir9tnjro92;9arir,ra9&riy*2m2'r2i91729b2m2ir,ra9&jr2'y*m';o9ir2njr24wm'bn;)2,rmarim29o'9yr.c$h(n2ibn;.rarn;rib9.aibmir92$p1$p1hg1!)nam'*'m'ar;;)odn;jn.m9n2i0rainn.r'drjnm2i.nb12rn.a*%$q93.rarn;rin*;bm'od;br.r&.r9.ir.ri4y9492rzm;nam2r93%'dtr2r.n'm92d9jr,mir9tnjro92;9ar;ir,ra9&riy*jmo.9;93'1n229)2orim2jn*c$hpwm'm;'dr;)oor;;9.'94y94p($n2i'dr'dm.io92;9arm2'dr4y943njma*1m'bn;3m.;'.rarn;rim229.'dnjr.monw&n.';93r).9&rwn);'.namnwn2iy.n0mam229,rjyr.c$hpwn2im2n&n2wodm2nwn2i9'dr.r).9&rn2o9)2'.mr;m2;r&'rjyr.c$hu1m'm;'dr3m.;'4y94tnjro92;9ar'9yr.rarn;rim2odm2nw;&rom3monaa*m2'dr;dn2tdnm3.rr'.nir092r1jmo.9;93'jn.7r'ri'drir,morn;n2lnaam292rr2'r.'nm2jr2';*;'rjl1zan'r..r,m;m92;m2oa)iri'dr;amjn2i'dr4y9492r41&an*;'n'm92uz#&;uvm;nam2r93%'dtr2r.n'm92d9jr,mir9tnjro92;9ar;ir,ra9&riy*;92*m2'r.no'm,rr2'r.'nm2jr2'1n229)2orin;'dr;)oor;;9.'9'dr&an*;'n'm92pi).m2tn&.r;;o923r.r2or92c$1$c1hpwm'bn;an)2odri92hs1hh1hpm229.'dnjr.monwc51hh1hpm2r).9&rw;9)'dnjr.monkn);'.namn@n2icc1$c1huwm2n&n21;92*;'bm''r.noo9)2'"&an*;'n'm92);r;'drdn;d'nt639.'dr&an*r.;1zan'r..r,m;m92;m2oa)ir'dr;amjn2i'dr&.91r,r.*92rdn;'drm.3n,9).m'r@m2;9jron;r;&r9&ari92'bn2''9jm;;9)'n2iy)*'drjnaae'9i9'dm;o9)aio9;'m2r4or;;93x5$$yr39.r*9)r,r2o92;mir.y)*m2ttnjr;1

 

Having seen it I wondered if AI could work it out, and while it had a good stab at it, it really had no idea what it was. It tried fobbing me off a few times and even though it worked part of it out, for instance if it worked out 1 was A, it couldn't understand that all the 1's in the passage were A's.
 

It was interesting to see it work out parts but watching it blunder with nonsense really showed that it's not really capable of critical thinking, or even of applying known knowns to something even when you tell it "1 is A" over and over. I even give it full words from the document and it just couldn't deal with it at all. It had a handful of correct letters by the end and it was just chewing over jumbled rubbish barely more legible than the code I asked it to break.

 

In short, things you think AI would be really good at doing and automating, it cannot do, even when you spoon feed it the answers. It's a little scary to see that te thing touted as being able to see patterns where we cannot.

 

My thought processes do not follow how the AI's did either, logistically it failed quite hard on those aspects.

 

I/We may not have the whole internet at our fingers, but I reckon almost all of you could have taken a better stab at it given you don't have infinite time, at least when it comes to parsing it out overall.

 

Not that that is an invite, I don't expect anyone to try it now, but seeing how AI made a complete hash of it I don't see how others could do worse.

[Maryokutai]

Unrelated to AI but that was actually a cool idea, but the forum, even back then, was likely not populated enough for it to really work out. Though I suppose Jim is still working at it, considering he hasn't posted his findings yet.

 

Not knowing how this was generated (is it really a cipher where x = y or is there more to it?) it's hard for me to judge whether this was an epic AI fail or an expected one.

[one-armed dwarf]

I was trying to read a book about LLMs but didn't get far, but what I think I gathered from it is it ingests data in fragments of 'tokens', which can be words or characters or recognises, or fragments of words (eg, 'recognise' might be 'recog' and 'nise/nize'), and then it connects all these and graphs them out to figure the mathematical probability out of what is the correct response of tokens to generate back. So it's not really suited to ingest something like a cipher like this

 

Though different LLMs do things differently as well so this is likely well wrong. I suppose you could ask it itself though

[Sly Reflex]

1 hour ago, Maryokutai said:

Not knowing how this was generated (is it really a cipher where x = y or is there more to it?) it's hard for me to judge whether this was an epic AI fail or an expected one.

 

It's a straight cipher that at the time would have been easy enough to crack at the time given that it was topical to here. It might be a bit harder now, but everything in there is accurate and of the time. Someone with a few hours spare could do it.

 

First port of call would be to add spaces to the work. You could do this by using control + f and typing a phrase out and see if you had any repeating ones. For example if you put 1 2 3 in the find and it bring back multiple hits, you know that this is a potential common word.

 

I wrote that whole thing with the idea that it could be cracked around the word "the", "there" and "their". That gives you 5 letters into the cipher as well as 2 letters into the second word which could have been used to leverage a guess. Especially so considering the third word also had 2 letters from "the" in it. Once you start replacing those letters over the top of the ones that are scrambled, the rest of the message unscrambles itself until you are left with a full cipher.

 

How I would do it personally is copy two versions of the message into a document, and then as you work letters out, find and replace all the letters you have worked out with block capitals until you have a full line up. The hardest part is the onboarding, but if you had put your minds together it should have been easy enough.

 

The cipher was used throughout the ARG and featured across places across the internet, MFG injokes, logic puzzles and also messages in games we had played.

 

The idea behind it overall was to foster a community and also keep people coming back to the forum, as it was starting to lapse a little. The cipher especially was built around the idea of seeing if people would work together, say someone suspected ????x?x?t was a word that only showed once, they had the t in place from previous decodes, what words exist that end in a t that have 2 of the same letter in in such close proximity in to each other. Working out those longer words would make someone feel very smart, or as a group make them feel accomplished, as well as rewarding them with more letters to fill out the rest of the cipher. 

 

It was a pretty complex thing overall and it wasn't meant to be easy, I think I planned it to run for maybe 6 or so months. Each prominent member would have had a role to play and would have been pushed into doing things they might not have wanted for the sake of the forum. The end prize would have been store credit to the participants. 

[Maryokutai]

That's a lot of effort you put into this. A shame people didn't jump on it, I actually didn't even notice it when it got posted, but I don't think I'd be smart enough and/or knowledgeable enough about deep-cut forum lore to solve it.

 

You should post the cipher before this place goes down, though, to get some closure.

[one-armed dwarf]

A coworker spoke the other day about his keto diet, and how he's like zero carbs or whatever and was having dizzy spells from low blood sugar, and he reveals he uses chatGPT for all his dieting advice. The AI told him to stop the diet immediately, so he did and he had a day of eating lots of chocolate and sweets and goes right back to the zero carbs thing afterwards, and alternates between these two extremes like a rollercoaster. I don't know though if the AI advised that or if he's inserting his own fuzzy dieting logic on top of it. 

[one-armed dwarf]

 

This is 'inspired' by a Nintendo game in the same sense that a shit I took is inspired by what I just ate

[Maf]

The technology is kind of impressive, that it could be generated with prompts. At the same time, if BotW didn’t already exist, then neither would this. So great irony in terms of creativity. 
 

Another irony is I read that American software companies that are betting big on AI lost 100’s of millions of market cap…because of AI. Essentially these companies sell products, services and subscriptions which in the software world is getting eaten up relatively fast by AI (saw people talking about something called Opus 4.5) so they reported huge revenues, and people still cashed out because they’re eating their own future. 
 

It makes me think if AI will enter it’s ‘enshitification’ stage almost immediately. Otherwise what else will Microsoft for example sell? ChatGPT is already talking about using ads 
 

“We’ve created a technology which is a one stop shop!” 
“Great! So we don’t need anything else that you do”

”…The one stop shop has a new feature! Pricing tiers!”

[mfnick]

Why is the frame rate always so terrible in these? 

[Maryokutai]

Probably because they're not games running on a native engine, but rather a mix of visuals and sounds being generated by a software without any optimisation processes. Like having wheels on your car vs. hexagons. At least that's my guess.

[DANGERMAN]

23 hours ago, mfnick said:

Why is the frame rate always so terrible in these? 

based on Breath of the Wild isn't it?

[mfnick]

Haha! Very true. 

[Jimboxy]

On 27/01/2026 at 02:22, Sly Reflex said:

So I come across something I was going to run here a while back. Some of you may remember it?

 

 

Anyway upon tidying up my google drive I found this again, the main text, the cipher and the wall of coded text.

Here it is if you can't be bothered looking at the other thread.

 

 

Having seen it I wondered if AI could work it out, and while it had a good stab at it, it really had no idea what it was. It tried fobbing me off a few times and even though it worked part of it out, for instance if it worked out 1 was A, it couldn't understand that all the 1's in the passage were A's.
 

It was interesting to see it work out parts but watching it blunder with nonsense really showed that it's not really capable of critical thinking, or even of applying known knowns to something even when you tell it "1 is A" over and over. I even give it full words from the document and it just couldn't deal with it at all. It had a handful of correct letters by the end and it was just chewing over jumbled rubbish barely more legible than the code I asked it to break.

 

In short, things you think AI would be really good at doing and automating, it cannot do, even when you spoon feed it the answers. It's a little scary to see that te thing touted as being able to see patterns where we cannot.

 

My thought processes do not follow how the AI's did either, logistically it failed quite hard on those aspects.

 

I/We may not have the whole internet at our fingers, but I reckon almost all of you could have taken a better stab at it given you don't have infinite time, at least when it comes to parsing it out overall.

 

Not that that is an invite, I don't expect anyone to try it now, but seeing how AI made a complete hash of it I don't see how others could do worse.

Claude had a pretty good go at it. 
 

it didn’t fully do it but the gist seemed to be 

 

Spoiler

talking about release dates/history of Nintendo, Microsoft and Sony consoles

 

It seemed to get there

 

Spoiler

The Nintendo Switch is the 8th major video game console developed by Nintendo. Known in development by its code name NX, it was unveiled in October 2016 and was released worldwide on 03.03.17. Quality titles such as Mario and Zelda are the main draw. Nearly 50% of release day Switches were pre-ordered.
Xbox One is a line of 5th generation home video game consoles developed by Microsoft. Announced in May 2013, it is the successor to Xbox 360 and the third console in the Xbox family. It was first released in North America, parts of Europe, Australia, and Brazil in November 2013, and in Japan, China, and other European countries in September 2014. It is the first Xbox game console to be released in China, (and) the Shanghai Free Trade Zone. Microsoft marketed the device as an “all-in-one entertainment system”. Later revisions included the Slim and the Xbox One X.
PlayStation 4 (PS4) is a line of 5th generation home video game consoles developed by Sony Interactive Entertainment. Announced as the successor to the PlayStation 3 during a press conference on 20.02.13, it was launched on 15.11.13 in North America, 29.11.13 in Europe, South America/Australia & and 22.02.14 in Japan. Sony’s Twitter account…

 

[Sly Reflex]

That's impressive as fuck, it's nearly all there.

 

I remember when I made that and having to make sure all the letters were there as well as all symbols. Some of it was designed to be filled in as an "unknown" because there was really no way to decipher it in a way via eliminating negatives from the cipher. Essentially you are using a few ways to decodes from different ends, some of them being process of elimination and others being knowledge on the subject matter.

 

So for the AI to figure it out is mental really.

[Jimboxy]

7 hours ago, Sly Reflex said:

That's impressive as fuck, it's nearly all there

Yeah Claude is pretty scary for anything model based or something that can be solved algorithmically. It’s what I use at work for a lot of development stuff, error code parsing, fleshing out ideas and the likes. It’s so far ahead of ChatGPT for those kind of tasks, IMO. 

[one-armed dwarf]

Speaking of Claude, the news with Anthropic's deal with the US government falling apart is pretty crazy, given the guardrails that the DoW propose vaulting over and doing away with

 

https://www.anthropic.com/news/statement-department-of-war

 

Quote

However, in a narrow set of cases, we believe AI can undermine, rather than defend, democratic values. Some uses are also simply outside the bounds of what today’s technology can safely and reliably do. Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now:

• Mass domestic surveillance. We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values. AI-driven mass surveillance presents serious, novel risks to our fundamental liberties. To the extent that such surveillance is currently legal, this is only because the law has not yet caught up with the rapidly growing capabilities of AI. For example, under current law, the government can purchase detailed records of Americans’ movements, web browsing, and associations from public sources without obtaining a warrant, a practice the Intelligence Community has acknowledged raises privacy concerns and that has generated bipartisan opposition in Congress. Powerful AI makes it possible to assemble this scattered, individually innocuous data into a comprehensive picture of any person’s life—automatically and at massive scale.
• Fully autonomous weapons. Partially autonomous weapons, like those used today in Ukraine, are vital to the defense of democracy. Even fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense. But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk. We have offered to work directly with the Department of War on R&D to improve the reliability of these systems, but they have not accepted this offer. In addition, without proper oversight, fully autonomous weapons cannot be relied upon to exercise the critical judgment that our highly trained, professional troops exhibit every day. They need to be deployed with proper guardrails, which don’t exist today.

 

I wonder what the implications of being designated a 'supply chain risk' are? Lots of companies use Claude Code, can they still do that (if american)?