iiinova — where your digital self keeps evolving

Tony Lower-Basch @tonylb.bsky.social · 08/10/2026

Honestly, the phrase "If all men were angels" has been coming up a **lot** when I discuss AI. Like, I think the capabilities of the technology are a terrific tool. The problem is, it's also like throwing gasoline on a bonfire of all of humanity's cognitive biases and worst instincts.

020

Tony Lower-Basch @tonylb.bsky.social · 07/10/2026

The only time I've ever clashed with an employer's legal department is when they tried to amend an IP agreement that _explicitly disavowed this_ to instead claim it. Legal said "If you keep pressing this, it may impact our employment." I responded "I gave notice the moment you floated this BS."

040

Tony Lower-Basch @tonylb.bsky.social · 06/10/2026

"Look, I know you think humans are just, y'know, elves with some muscle mass, or dwarves with height, but you gotta understand: There's something weird with these people." "You give them two bags of holding, they don't say 'Let's keep one in reserve, in case,' they PUT THE SECOND BAG IN THE FIRST!"

020

Tony Lower-Basch @tonylb.bsky.social · 06/10/2026

At one point my older son was lamenting that I was mellow enough that it was hard to find ways to rebel that I would dig in against. My younger said "Just tell dad that he's so wise you do what he does without thinking it through yourself. He'll hate it." That's how they trolled me, for **years**.

010

Tony Lower-Basch @tonylb.bsky.social · 02/10/2026

Honestly, I always go with Stress points from Blades in the Dark, but yeah, I feel all of that. You're sorta ... not supposed to say it out loud, though :)

010

Tony Lower-Basch @tonylb.bsky.social · 02/10/2026

So, like, fitness tracking devices too?

110

Tony Lower-Basch @tonylb.bsky.social · 01/10/2026

Honestly, they ask Claude to interpret your response, it does so, and the conversation goes on. As a programmer since elementary school, that sounds vaguely utopian to me.

000

Tony Lower-Basch @tonylb.bsky.social · 01/10/2026

So my feeds fed me the (admittedly catchy!) "The problem with AI is that it's designed to have the ethics of a billionaire." I know nuance doesn't go viral, but I think it's more like "the ethics of the billionaire's favorite collaborator/toady." You can see them trying to be good, then caving.

010

Tony Lower-Basch @tonylb.bsky.social · 30/09/2026

I know you played _My Life with Master_, because I was there at the table being horrified.

111

Tony Lower-Basch @tonylb.bsky.social · 25/09/2026

"It's not a fleet, sir. It's just ... people."

000

Tony Lower-Basch @tonylb.bsky.social · 25/09/2026

Happy Dolly Parton Day to all who celebrate!

000

Tony Lower-Basch @tonylb.bsky.social · 24/09/2026

Both bags reveal that they have always been sentient, they just previously had nothing to say. But this ... this is just _offensive_. They will first insist, of course, on the indignity being reversed. Then they will, forever after, have _opinions_ on what to allow you to put in or pull out.

050

Tony Lower-Basch @tonylb.bsky.social · 22/09/2026

In my (admittedly limited) experience, the value of this varies _wildly_ depending on the content you're generating. If I'm generating something akin to symbolic logic or programming, it's easy for me to look at an individual item and say "That contradicts." But for descriptive writing, no.

000

Tony Lower-Basch @tonylb.bsky.social · 18/09/2026

"Is this the fast that has been commanded of you?"

000

Tony Lower-Basch @tonylb.bsky.social · 17/09/2026

The challenge of articulation is where I see a *ton* of mental overlap between computer programming and game-design, but then I would, wouldn't I?

010

Tony Lower-Basch @tonylb.bsky.social · 14/09/2026

I usually steer clear of pitching _Capes_ these days, what with it being ancient history, but ... like ... rewarding engaging adversaries when they fail is very much at the core of its game loop. The bonus for failing engagingly was much, *much* more powerful than the reward for success.

110

Tony Lower-Basch @tonylb.bsky.social · 09/09/2026

Yeah, but I can sympathize with the position that the burden of proof for usefulness is on those asking to dedicate, like, a trillion dollars of community resources to the project. I wonder if, in hindsight, we'll be able to distinguish the knuckle-heads who jumped on the hype from other adopters.

110

Tony Lower-Basch @tonylb.bsky.social · 09/09/2026

Adam: What do you think about the (I think large) demographic of people who have a completely justified BS-detector overload in response to the insanely overwrought hype-machine around AI, and who (again justly) apply that as a cognitive surcharge levied against any claims of usefulness?

110

Tony Lower-Basch @tonylb.bsky.social · 06/09/2026

Okay, this isn't the point, but there is **no way** that they actually have meaningful accuracy to as many decimal points of precision as they are reporting. Assuming a seismic wave is on the range of 10 meters in size, they're claiming measurement precision _smaller than a single atom_.

000

Tony Lower-Basch @tonylb.bsky.social · 05/09/2026

Hot take for a Saturday morning: If (a) a start-up founder sincerely believes that employees should be and are mission-driven then it follows that (b) they must welcome giving a labor union the authority to fire them (the founder) for failing the mission. Contrapositive: If not (b) then not (a).

010

Tony Lower-Basch @tonylb.bsky.social · 31/08/2026

I definitely _do_ think about why people dislike me, in order to follow the advice of the late, great, Dolly Parton: "Figure out who you are and do it on purpose." Insecurity lady: "NEW people are disliking you for the same reason previous ones have." Me: "And _thank you_ for noticing, sweetie!"

081

Tony Lower-Basch @tonylb.bsky.social · 24/08/2026

I think I need to figure out a "couch to 5k"-style training program to build my ability to publicize myself and my projects.

000

Tony Lower-Basch @tonylb.bsky.social · 20/08/2026

Does he know what virality _is_? "It made a lot more people _see_ the message, but detracted from its impact" is some academic ivory tower talking-head BS if ever I've heard any (and I have, so there you are).

061

Tony Lower-Basch @tonylb.bsky.social · 13/08/2026

Oh boy. Engineer-speak to Corpo-speak mismatch. Corpo: "Could it be scaled?" (in our budget, without compromising other priorities, which is to say 'off the side of your desk') Engineer: "Yes, it can absolutely be scaled." (theoretically, with a team of ten and five years, after resolving unknowns)

020

Tony Lower-Basch @tonylb.bsky.social · 12/08/2026

But simultaneously frustrating, because they're not going to **get** the real benefits of automation. AI is yet another fad promising to let them put off learning the lesson that good processes take time and discipline, _whatever_ the tools.

040

Tony Lower-Basch @tonylb.bsky.social · 12/08/2026

There were absolutely more points of friction where people had an _opportunity_ to apply discipline. I just can't, myself, get too rosily nostalgic about how often we actually **did**. We coded a remarkable number of stupid things the slow way, by hand. It's an embarrassment.

130

Tony Lower-Basch @tonylb.bsky.social · 11/08/2026

"Now"? I think the _scale_ at which Big, Bad, Fast, Code can be written and deployed has had a scale upgrade, but the problem of Big, Bad, Fast, Code has a long history (cf. Y2K and the _entire world data infrastructure_)

110

Tony Lower-Basch @tonylb.bsky.social · 11/08/2026

But then how can they be arch-nemeses?

010

Tony Lower-Basch @tonylb.bsky.social · 10/08/2026

I mean ... I can talk about how to bridge that gap, in practice. I've got a **lot** of techniques. But I would only want to do that separate from the context of it being a response that could come across as saying you're wrong, because you are _very much right_ that it's a gap that needs filling.

010

Tony Lower-Basch @tonylb.bsky.social · 10/08/2026

That is the visual that makes it all worth it to me ... some tiny little imp manfully shoulder-resting a thin little wand, then full-body recoiling when it goes off.

010

Tony Lower-Basch @tonylb.bsky.social · 09/08/2026

Oh ... it's a Gish Gallop of corruption and greed.

010

Tony Lower-Basch @tonylb.bsky.social · 08/08/2026

As opposed to "(sigh) I understand why you did it, and I have some idea how much it hurt you to do it in the moment, and then how that hurt became a dull constant ache over time, that's tolerable most days but acts up when it rains."

010

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

All of that is _known deterministic_ code. Somebody just has to (a) figure out the right architecture, (b) write it, and (c) debug the integration with the LLM (the simplest part ... LLMs love well-structured relevant data). It's inglorious programming work, but it unlocks a lot.

110

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

If you want the LLM to know "Here's what Game A says about it, here's what Game B says about it," thats context-tagging (maybe a general nested hierarchical tagging system) and quite possibly provisional thematic threading (maybe a hypothesis-sandbox).

110

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

... the consequence of which is _any search_ for (say) "attack rolls" comes back with all the attack roll references across all games, largely without the context of "where did it come from?"

110

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

Uh ... maybe, but that's not what I'm saying. I'm just talking about a _better data schema_ and relevant search functionality, much more like SQL architecture than like ML training. Like, I suspect that the "Load PDFs for an LLM" is just doing a single flat index, based on semantic embeddings...

210

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

Which is to say: A KR system for documents absolutely **could** provide the context to say "This game says X, this other game says Y," such that an LLM could actually do useful things like answer "How have different games approached topic Z, compare and contrast?", but it would have to be built.

120

Tony Lower-Basch @tonylb.bsky.social · 07/08/2026

One of the underlying truths that the hype buries is that LLMs require sophisticated knowledge-representation systems that nobody bothers to build, because they can demo (and therefore sell) without them. Many AI failures are the engineering equivalent of "computer not plugged in, why it not work?"

130

Tony Lower-Basch @tonylb.bsky.social · 05/08/2026

You relic. Only keywords like SELECT and FROM get all-caps.

020

Tony Lower-Basch @tonylb.bsky.social · 31/07/2026

FWIW, I know a lot of IT engineers of various stripes who can give you detailed and extensive lists of what types of things AI should never, ever, under any circumstances be used for. There is, of course, big money in making a start-up to use AI to do those things, but that's not exactly the point.

020

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

"Well, Odo, you're getting much better at humanoid hair, but your attention is definitely slipping on the other details."

010

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

If you make a claim, especially about your enemies or marginalized people, and that claim _cannot be disproven_ then it's not a claim about truth: It's a rallying cry for like-minded people to _also_ be irrational in the same way. Without disprovability, that's all it can ever be.

000

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

And I was thinking today "Would the argument that the Deep State and Liberal Media are persecuting Donald Trump be strengthened by major parts of the Epstein files leaking incriminating him?" Yes, yes, it would. So as a mental short-hand: Every day that **doesn't** happen, it's counter-evidence.

100

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

So there were folks back in Obama days predicting with **100%** confidence that, based on their beliefs about how that administration operated, Texas would be placed under martial law and its people enslaved. When that didn't happen, they used it _not_ happening as evidence for the beliefs.

100

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

Likewise, as you might infer, if you expect X and not-X happens, your belief in your rule gets a little weaker. How much (in each case) is math. The key point, for me, though is this: If you can imagine an explanation that fits your model for either X or not-X then your model is irrational.

100

Tony Lower-Basch @tonylb.bsky.social · 30/07/2026

Little morning politics rant about Bayesian inference. Yes, this mathematical framework is **deeply** political in the current climate. The _very basic_ top-line of this elegant mathematical structure is: If you expect X, and X happens, your belief in your predictive rule can rise a certain amount.

100

Tony Lower-Basch @tonylb.bsky.social · 26/07/2026

Create your own Silmarilion? Or better yet, leave it as a project for your heirs :)

010

Tony Lower-Basch @tonylb.bsky.social · 22/07/2026

Although I genuinely _can_ get a laugh in conversation with my wife if I properly dead-pan a delivery of "This is an insightful point, and it's worth digging in to precisely how it delivers value."

010

Tony Lower-Basch @tonylb.bsky.social · 22/07/2026

"Oh ... sorry. Am I early? No worries. I'll wait."

010

Tony Lower-Basch @tonylb.bsky.social · 22/07/2026

Yeah ... like ... I would still *object* if I were done dirty as part of some clever and villainous plan. But I object **more** when it is part of an idiotic blunder that I have to keep watching play out for years. (See also: U.S. politics)

010