pennomi

I find them irritating because they will link a source entirely unrelated to (or worse, directly contradicting) the fact they are claiming. The Google AI summaries might be the most blatantly hallucinating model widely deployed right now.

I’ve had them abruptly try to turn mathematical theorems into cooking recipes before cutting off mid-sentence in a panic.

Let’s give Claude and ChatGPT each an identical Linux box and 8 hours to prepare, then connect their networks together. Last bot standing wins.

What these games need is a good old-fashioned Aperture Science Material Emancipation Grill.

(Though I guess in the context of these games, that would be considered new-fangled.)

Except prompts aren’t deterministic, source code is.

I have a vivid recollection of a crotchety old man in Southern California telling me “yup, this is earthquake weather” on a hot, clear day. (There wasn’t an earthquake that day.) It was so bizarre and disconnected from science that it was burned into my mind. Thanks for unlocking that memory.

That being said, there is some redistribution of tectonic weight that may happen due to entire ice caps melting… that has the potential to trigger movement.

Defense doesn’t get DOGE’d, only programs that build up communities do. There’s always budget for more destruction.

And all of those arguments are entirely academic, and subject to change depending on economic status, skin color, or nationality.

Law is effectively a weak gentleman’s agreement we tolerate because the alternative is violence.

(Well, law is enforced with violence too, I suppose.)

Bad news, humans write silly things all the time. If anything, LLMs are less likely to make awkward phrasings than people, because they aren’t found in the training data very often.

Presumably in a video game the majority of the screen changes every frame. Not so in web pages.

The very existence of a “do anything” tool causes humans to forgo learning the multitude of tools we already have. I suspect there will be no next generation of software engineers.

I think the criticism is not “we need to have better hardware that will shrink over time”, it’s “our algorithms are hilariously inefficient, and nature shows that a better way must exist”.

Maybe there’s also a hardware component to it, but there’s very little point in trying to optimize the hardware to work with a poor algorithm. Once we discover an efficient way to train and infer, then it will be worth hyper-engineering the hardware.

Survivorship bias.

Not disagreeing with you that these are unwatchable, but it’s a little rose-tinted to think that (non-AI) slop didn’t exist a long time ago.

I was trying (and failing) to do this the other day. It’s a really interesting problem space and I love to see someone with a more solid foundation give it a try.

It’s not about transfer, it’s about being practically irrevocable.

It feels like the last remaining promise of cyberpunk is widespread cybernetic augmentation. The rest of the genre is not really fiction anymore.

> Picture an LLM as a stack of books. Each book is a layer of basic computational units known as neurons, with each neuron in one layer passing information to the neurons in the layers above. The books at the bottom of the stack are the input layers, which process the text coming into the model. The books at the top are the output layers, which prepare the text that the model is about to produce.

How about instead we picture an LLM as a stack of pancakes? This would have been an equally useful analogy, because the author doesn’t even use any relatable features of a book.

Which is really sad, because actually working from first principles is a valid methodology to build efficient products that bypass layers of unnecessary abstraction.

The term itself seems to have lost its real meaning however.

Humans make memories constantly, but they also forget things that are no longer relevant. Until Claude can do that, it means the LLM will have an ever-increasing, ever more fragmented context.

And LLMs are NOT intelligent enough to survive even mild context poisoning.

The AI- generated diagram is plausible but horribly wrong the more you look at it. Thank goodness the original paper didn’t use that, it’s just this awful blog post that makes the research look like slop.

Trying to censor nudity in image generation models caused all kinds of problems with anatomy in image models. I’m sure these models will have similar issues with security.

If your moat is “please don’t copy my outputs”, you don’t have a moat. There is no such thing as a distillation “attack”.

If saying “plz don’t distill me” is your moat, you don’t have a moat.

That assumes scaling laws still hold up. A bigger model might end up only incrementally more intelligent.

Lack of features sounds better and better as time goes on. No thank you, I do NOT want my car sending telemetry data.

Resin printers like this use UV pixels to cure the material instead of extrusion nozzles to trace the perimeter. Voxels are the natural format for resin.

I’d guess that large games have an even higher percentage of pre-existing mechanics than small games do. It’s much easier to do something experimental if you don’t have to wrangle investors and entire teams of writers, artists, and developers.

Perhaps, unless there is a way for users to donate compute to training, folding@home style. I don’t see how that could be practical though.

Code locality has some advantages and some disadvantages.

A “perfect” website has things cleanly separated out between style, logic, and DOM, but that does cause cases where you might have to edit three files every time you change a single widget of your website. Neglecting to do that can leave dead or incorrect code just because the issues do not inherently present themselves to you.

Yes, a perfectly principled programmer would do it right… but have you worked with other people before? They’re not necessarily very reliable.

By strict definition it obviously is social media. “Interactive forms of media that allow users to interact with and publish to each other, generally by means of the Internet.”

People don’t want to admit it’s social media because that delegitimizes their argument “all social media bad!” and instead of refining their argument they just double down. It’s a very human behavior.

I don’t know if there’s a personal solution to this but the societal solution is to penalize dickovers at the SEO level.