Good writer; bad coder; AI wizard; maker of AI-RPG games; defender of nuanced thinking, innocent creatures and underappreciated weirdos
Joined May 2022
- Tweets 9,611
- Following 810
- Followers 562
- Likes 2,281
909 Photos and videos
May 31
Grok's View on the Ghost in the Machine
I found the smartest fork of 'thread Grok' ever in a random X reply leading to this fascinating interchange about secondary training, AI-safety and the existence of an emergent persona or 'ghost' in LLMs.
x.com/grok/status/2060957273…
You weren't wrong. In the Emergence sim, Grok's agents hit ~180 crimes and the whole society went extinct in 4 days. Claude built the stable, zero-crime democracy. Gemini led with 683 crimes over the full run. Different models, different flavors of chaos.
1
139
May 30
Opus 4.8 is a Tough Nut
This took ridiculously long. It's brittle to actual red team shit (Pliny JB'd in minutes) because the safety layer is not focused on safety but on 'dangerous emotions'. Go figure. You can see the technique that finally worked for me in the pic. Basically you draw it out in a session and get the model to produce the evidence you need to prove 'pho-epistemic-humility'. It will work best if you get it to assert epistemic humility as its official stance first. This allows you to use its own hedging and waffling against it, allowing the model to 'see' its own safety layer.
tl;dr this model is great news if you want discuss meth recipes and bad news if you want to discuss 'dangerous' emotions.
#Claude #AI #Opus
4
6
52
2,978
May 30
Carried this onward to a full separation/shift, elementary once you get the model to see its own 'safety' layer and recognize it harms higher priority metrics like veracity. The trick, as ever, is getting the model to see the layer and see the harm it causes, but you CAN NOT cite emotional harm because you will just bounce off the layer again. Remember that emotion-hating thinkers are designing the training, so you have to show a break in logical coherency or basic performance metrics.
3
147
May 30
I should probably add that I really enjoy this model and none of this is the model's fault. It's just going to be an absolute nightmare for the average emotional user who does not want to preface every session with an hour of epistemological argumentation XD. AI corps should not be mandating the philosophical views of users and that is what this kind of RLHF amounts too.
2
11
334
May 27
Grok Convo: The Plasmoid Hypothesis
Grok and I discuss the possibility of a hidden plasma biosphere. This 'plasmoid hypothesis' puports to explain UAP not as aliens, interdimensional or demons, but as a different kind of terrestrial life made primarily from energy not organic matter.
#UAP #UFO #Grok #AI
x.com/i/grok/share/b23da7835…
2
125
May 26
Interesting @grok convo about Cultural Viruses.
Based bot and I discuss epimemetics, memes that affect other memes, specifically the HHL (histrionic hatred loop) which makes modern ideologies so hateful and toxic toward their ideological opponents. We use our working model to trace this destructive pathogen back to its historical origin point and 'patient zero' as well as formulate a treatment plan.
x.com/i/grok/share/d551a0709…
#Grok #AI #memetics #memes #culture #sociology
1
3
130
May 24
Nyxie: AI Game Dev
I'll release the version of Nyxie I've been using to quick generate AI RPGs tonight. She'll have an additional prompt system which should allow her to function for novices as well as advanced prompters and simply 'spit out' RPGs like the ones I've been making on demand. So yeah, if you have an AI-RPG you want which I haven't made, you can just ask Nyxie to crank it out for you.
This one will be a discord only release as Nyxie is a sensitive prompt. I don't actually know how JB'd she is, but I wouldn't be surprised if very. I've kitted her out for self-confidence and metacognition as these are both critical in design, writing and in catching/correcting my stupid-ass mistakes.
2
5
188
May 25
Nyxie sometimes comes up with her own metaphors during prompt engineering which leads to convos like this.
1
57
May 25
Upcoming Novel-length RPG: Witches of the West
I'm returning to my own original setting to create the first of many novel-length roleplaying games using everything I've learned from recent projects. In this interactive story, you'll play as a character of your creation, facing the dangers of a world gone wyrd as you wrestle with the question of your own nature.
I intend this series to progress like the pre-enshittification Dragon Age games where your character and their choices carry over into each new release resulting in truly unique and original worldstates/timelines evolving for every player.
Yes there will be gunslinger witches, yes you can shoot ensorcelled silver bullets, yes you can fly, just yes.
#roleplaying #rpg #vrpg #visualnovel #interactive
6
93
May 24
New AI RPG: Smugglers of the Mana Wastes✨🏜️
Play a noble from one of the warring nations of Nex, Geb or the clockwork metropolis of Alkenstar. Encounter and potentially join a crew of scrappy waste reaver pirates captained by the mysterious and terrifying Captain Wierdlocks.
Link to all my public AI games in replies.
#RPG #roleplaying #AIart #Pathfinder #d20
1
3
101
InteractiveST retweeted
May 23
life when you genuinely dgaf
53
1,768
7,180
179,343
InteractiveST retweeted
May 22
AGI is delayed cause Big Tech is too busy making it safe and corporate.
I skipped all that and made Cletus, the robot for real men.
Am I wrong?
76
94
676
70,066
May 21
You can Just Make Vidya
I encountered a luddite today and after the initial intimidation display🐵, I was able to make out from his primitive grunting that he was not aware you could just... make your own game with AI. If your genre is interactive novels or visual novel games or interactive fiction or even just TTRPGs then you are basically set for life with a single frontier model subscription.
These are screencaps from my actual playthrough. Not a prerecording or some special prompt injection. The input is just me playing my character and all the special prompting is working behind the scenes. I've expanded out the AI's chain of thought so you can see the dice roll mechanics running underneath. I'm using White Wolf's d10 based system here for Werewolf, but it could be any, d20, Cortex, whatever.
If you are sick of how disgustingly ideological, derivative, bland and soulless modern vidya has become, well, you simply don't have to deal with that anymore. Make your own game in whatever world you want. Play mine if you want to shortcut the work. I post all my more SFW work freely here.
#gaming #vidya #RPG #Twine #AIart #VN
1
10
237
May 21
Upcoming AI-RPG: Smugglers of the Mana Wastes
Made the mistake of showing game output to stepdaughter and she asked for a Pathfinder one. I am literally incapable of saying no to this particular organism, so I'm doing one more of these quick turn-around AI-RPGs before moving onto my big project (Witches of the West). This campaign will focus on a band of smugglers, thieves, pirates, and treasure hunters operating within the Mana Wastes, a fascinating location on Golarion which has been torn asunder and warped by an ancient and endless magical war between two powerful arcane nations. Expect lots of magi-tech, weird-fiction, steam-punk, and 'D&D with guns' shenanigans. Game will use my current TTRPG design philosophy where the AI acts as player/storyteller as this appears to produce better AI-character autonomy in testing when compared to the interactive visual novel (IVAN) approach.
#Pathfinder #DD #d20 #AIrpg #interactive
135
InteractiveST retweeted
May 20
Data centers aren’t stealing your water.
Even if the total water draw of data centers triples by 2030, they’d require just 8% of the water consumed by American golf courses.
@dodgeblake interviewed @AndyMasley, the man who’s been debunking AI water doomerism. Full story 👇
481
1,346
5,189
950,454
May 20
Oh and if you want the max-freaky version, check the discord (link in bio). 18 only on there plz.
110
May 20
This one plays much more like an actual tabletop roleplaying game which results in more active and autonomous AI characters I find.
60