iiinova — where your digital self keeps evolving

I finally blocked her because this is the latest in a long string of engagement farming and band-wagoning behavior that I am just not interested in dealing with it in my feed.

0120

Now that I think of it, I guess that's also how FATE rolls go?

000

Weirdly I think it *does* work in something like Battletech, because the bell curve nature means you are making each move to stack up as many beneficial modifiers as you can from different sources each turn. It's almost like building combos. There, the Sigmoid curve feels like it works.

100

Oh and as a die system: I think 2d6+Mod works best for something where challenges encompass larger scenes rather than individual actions, and where consequences are going to be detailed. I wouldn't use it for something like a DnD skill system because the "interesting" window of modifiers is so small

100

I feel like the Ethos of PbtA is less "overcome the adventure" and more "surviving and defining who my character is by what gets left behind".

But again, odds are you will get something like what you want out of each scene, you're just rolling to determine who dictates the next scene and/or how much the encounter will leave a mark on you.

Also, there's an implicit set of "fronts" the GM is tracking that are various off screen threats that tick towards affecting the world over time (even as no rolled encounters happen).

100

And indeed, that's often a good strategy (just like choosing paths that give you options that match your best attributes). You *could* play PbtA as a chain of skill challenges back to back, but its original vision was more as a way of adjudicating scenes where the outcome was chancey or unclear.

100

Ultimately it works best as a collaborative system across the table (when the GM gets a move it's "what's most true to the story") rather than a gamist adversarial one (where the GM choice would usually be "what's the optimal antagonist outcome") .

010

So it's not really about success or failure, it's about the terms on which you move forward, how much you control the next set of options, and how much it's going to cost you by the end. Of course, enough bad things can still lead to a total failure via GM moves.

- Bad outcome means the roller might still get what they want, but at a cost *they don't choose* (you move the story but it might be phyrric) and lose control of the narrative (GM makes a move)

- Neutral outcome means the roller gets what they want at a cost (you move the story but make it complicated) which they can choose.

-Good outcome means the roller gets what they wanted and more, or takes a win and retains narrative control.

Nathan Hoobler @hoobler.net · 27/09/2026

The thing about PbtA is that every roll moves the story forward, it's usually a question of on whose terms. The rolls are usually forced by by circumstances or decisions so not acting is just not an option (and who gets to decide what's happening and control the narrative is a prize to be won).

220

As someone pointed out elsewhere, the deaths associated with GPT-o4 coaching people into self harm was not the model being a "super-persuader", but rather being "super-persuadable" and readily reinforcing whatever preferences the user already expressed. Arguably "alignment to user" was the danger!

010

Nathan Hoobler @hoobler.net · 27/09/2026

I had never considered that and you're blowing my mind right now

Nathan Hoobler @hoobler.net · 26/09/2026

I think this is basically a restating of the Mean Value Theorem applied to outcomes over a set of people. O1 cannot be worse than O2 if all O1(x) >= O2(x) for {x|people}

000

Reposted by Nathan Hoobler

Arseny Kapoulkine @zeux.io · 25/09/2026

meshoptimizer 1.3 is out! Featuring a new voxel remesher, support for normal generation, several simplification improvements and various optimizations and fixes. GitHub stars and reposts appreciated! meshoptimizer was started 10 years ago; excited for the years to come! github.com/zeux/meshopt...

29427

We model uncertainty with random variables which rely on certain assumptions and use the math of statistics to propagate that uncertainty through the system. However, that doesn't make either of those actually real on a fundamental level.

000

"these are the faces of the die showing the set of possible outcomes. *Assuming a random selection from the set* and *identical trials* what's the expected value after N rolls?"

100

The bigger thing is that probability is abstract mathematics -- "let's assume a random variable exists and is fed into a CDF as input -- what's the output?" And the real world it can only be *predictive* by extrapolating from a *description* of a set of possible outcomes.

So I think it's fair to say that by the "perfect knowledge" standard there would be virtually no random events outside of (quantum phenomenon). I could predict a die roll with perfect information/skill. So this line of thinking might be reductive/misleading.

210

Again, our own human principles of "morality" and "ethics" *emerged from introspection enabled by our evolving intelligence*. Now maybe such a system could accumulate disproportionate influence relative to its own introspection, but viewing this as obvious and inevitable seems flat wrong.

040

But are you telling me that, at no point during the process of becoming more intelligent than all of accumulated humanity by turning the solar system into one supercomputing cluster, the Galactic Paperclip Elemental doesn't use its intellect to think "oh, hey, maybe this isn't what they meant?"

140

rev. howard arson @theophite.bsky.social · 25/09/2026

My hot take is that the Orthogonality Thesis is flat out wrong in the limit, and flawed in its fundamental assumptions. By its construction it assumes "intelligence" is defined as something that can be kept orthogonal to an "intention" even as it increases unboundedly.

220

Reposted by Nathan Hoobler

i think that the orthogonality thesis is somewhat confounded by the fact that the only obvious example of intelligence we have access to has morality as a result of evolutionary mesa-optimization substituting an instrumental objective for the terminal objective!

4451

I feel like this matters because it recognizes that the future is, in fact, contingent upon the present, and also on the actions you might take to affect the path as information reveals itself.

060

And more importantly, the exact same thing would happen if you somehow re-ran the election/event. There's no "multiple outcomes", there's one outcome that you were unable to predict based on your data and model (which your probabilities expressed).

250

Like if we have an election with polling that says Candidate A is favored 70/30 and you take the modal outcome as candidate A wins and then they lose, you aren't 30% right! The uncertain elements your probabilities modeled broke in ways that conspired to make the less favored outcome occur.

160

This is something that's always bothered me because it is confusing prediction likelihood measuring uncertainty with a genuine random variable. We are uncertain about the future because we lack perfect knowledge of the present, not because it's unknowable!

180

Steve Klabnik @steveklabnik.com · 24/09/2026

The tort offense here is negligence, and the "hack" is a crime of recklessness rather than intent (manslaughter vs murder).

000

Reposted by Nathan Hoobler

“I asked ChatGPT to explain this compiler error, and it helped me fix my code" "uhh actually you entered the compiler error into ChatGPT and used its output to fix your code, stop anthropomorphizing" they're gonna go "uhh yeah okay". it's just not an effective way to communicate

14917

Thinking about that more, did we just reinvent the Turing Test?

000

One might argue that we *are* approaching The Singularity, but it's epistemic rather than functional

2114

But I do think that's at least partially a product of me molding my brain to the technology of the time, rather than a particularly universal truth.

010

Similar to how it's sometimes hard to use coding assistants because turning a vague idea into code is how I've learned to puzzle through what I really want. Jumping to a plain-text description specific enough to work actually feels like skipping intermediate steps!

I've tried to use whisperflow for the past week and failing. I think it really comes down to me spending the prior 30 years getting really good at putting ideas into transmissible format via a keyboard and not really knowing how to use the same thought patterns while speaking.

To put it another way: wet biology is definitely very information-dense in a physical sense relative to, say, a combination of crystal lattices and etched patterns, but it's not clear to me how much of that is "signal" vs "noise". And seems like a valid question!

This is all definitely hand-wavey, since it's not clear how you count the more continuous states of neuron biology (beyond counting quantum states which would way advantage it over binary data on a drive)

100

So if you subtract out the "matter" the information on the model weights is likely closer to "white noise", whereas my intuition is the brain has much more redundancy/repeating patterns (and does so in a denser way).

100