When I have kids, I’m gonna make sure the first AIs they get exposed to are really old dumb ones while I explain how they work. Then maybe they’ll always have that little bit of wariness in the back of their minds, like I got because I used those old dumb AIs when they first came out.

The apes settle their differences and none of them are forced out of the jungle into the plains.

And BAM, humans no longer exist.

Weird opinion: I like square brackets, angle brackets are neat, curly braces range from neat to ok.

But I just don’t like the look of parentheses compared to the others. They especially look ugly and clunky when you need a bunch of them in a row, in a way that’s not true for the others.

I can’t think of a reason why I feel this way apart from the fact they tend to be very clunky in math too. They’re also the ugliest one when handwritten despite being the arguably simplest shape.

All that said it annoys me how parentheses are the most used bracket type in programming.

But buying up the world’s silicon wafers that you don’t even have the ability to turn into chips is a much better use of resources, right Sammy?

Reminder that the transformer architecture, the thing all LLMs rely on, was originally built to translate between languages for the purpose of Google Translate.

And compilers are just translators, translating from a programming language (and/or an intermediary) into machine code. So the task of translating it back is very close to what LLMs specialize in.

So I always had it in the back of my mind that LLMs would be incredibly powerful for decompilation, I’m just surprised it took this long to be good enough to be a problem.

These certificate companies are responsible for keeping internet traffic encrypted.

They use a private key which you can imagine as a combination number lock of 116 digits.

This specific company seemed to only be generating numbers that fell in the range of 0 to 256, and at the time of the breach the smallest one was 4.

I want you to imagine your data being secured with a lock using a numpad of 116 digits. Now imagine the passcode turned out to be 115 zeros and then 4.

If you don’t prematurely optimize your code you’ll never optimize your code.

I know people are advised against premature optimization because it risks you wasting time on something that ends up unused.

But let’s be honest, once you get something into a working state will you ever be afforded the time to go back to it? Even if you are, are you really gonna choose to uproot code and do a refactor on it and potentially everything relying on it?

Rather than waiting until your hacks and workarounds run out and you absolutely have to do a refactor, I would prefer spending more time upfront to make sure it is efficient and flexible before moving on.

The README can only be out of date if the project is maintained without too much AI.

Because AI could fix a typo and then waste your tokens regenerating the entire README to tell new users that there was a typo that was removed.

Petition to have maps change the names of all cities, states, and landmarks to America or some variation of America.

You and I have a similar workflow. One thing AI solved for me is not having to hunt obscure specific documentation or wade through unhelpful forums. It also helps make a test script for when I want to see if something’s working without another 20 minute sidequest.

I’m mainly referring to the people who want the AI to do all the work, which has unfortunately become a common expectation lately. Some people want to feel like programmers without ever reading or writing any of the raw code.

I’d say I actually fall in that camp. Right now I’m trying to set up an open source AI assistant that’s like, a little dumb but can still help me while I’m coding.

My ideal relationship with AI is this image:

Really shows the difference between the “I like problem solving” crowd and the “I like shiny tech and everyone said this would make money” crowd.

If style == localtime(): belt.append(onion)

An AI agent making a frontend for a bot stopping tool in order to appease corporate drones.

It’s robots all the way down!

They might as well just die, after all god-grok will resurrect them right?

Turns out putting sawdust into our rice krispies made people stop buying our product.

Apparently they want the “rice premium”.

You wouldn’t even need to pop songs off the list, just shuffle, go through the playlist, shuffle then restart the playlist.

But the music app developers aren’t trying to give you true randomness. They’re using machine learning algorithms to try and guess what you really want.

Nothing does what it says it does anymore, everything’s trying to second guess you. I hate it.

I wonder if a parrot can be trained to do a podcast, a significant amount of them are just parroting each other anyways.

From what I understand, pain is generally defined as our neurons wasting energy.

If we assume the Free Energy Principle is fundamental, then updating information produces free energy, which is more intuitive to think of as wasted energy. Overfitted/overcomplicated beliefs take more energy more often to fix to be more accurate. So free energy is defined at complexity of belief minus accuracy of belief.

The brain over evolution has defined a very ingrained belief that “my pain sensors shouldn’t activate” so when your pain sensors activate it violates this belief.

Emotions are actually condensed representations of your bodily state and some of your mental state, and whether they are good or bad depends on if you can predict your emotions. For instance you feel fear because you are trying to predict your pain sensors will go off to reduce its pain, but you keep predicting it’ll happen when it hasn’t yet. It’s also why anger during a boxing fight could feel good and anger when you stub your toe does not, one was predictable the other a surprise.

Assuming all the links in the chain of theories are true, then current LLMs cannot feel pain or emotion. They care about prediction error and sometimes complexity during training, but during inference it doesn’t get any signal or consequence for delivering the wrong token. It has no internal state except for the context window, which is never surprising for the LLM. The attention layers literally guides the model for what is most expected from the text it’s handed.

Even if the LLM received a negative training signal during inference, acting wounded after pain is the expected response. A random sequence of gibberish would be more “painful” than any torture fantasy.

On top of all this, the Free Energy Principle works because on a cellular neuron level we are built to reduce wasted energy, free energy being one source of wasted energy. LLMs are built on matrix multiplication running through GPUs and neither are even aware of the energy wasted, nor the free energy produced from changing values.

I think a feeling machine is possible, but the foundation of neural networks right now don’t have the necessary components to drive any subjective sensation or emotion.

I feel like the big official sounding word of “impeachment” should be reserved for after the Senate convicts. Otherwise it’s just a meaningless gesture on behalf of the House.

I began searching to see how Bill Clinton was removed, then I realized the Senate acquitted him and his term naturally ended later that month. Andrew Johnson also wasn’t convicted.

This mechanism of removing the president from office has literally never worked before! The whole impeachment process is an overhyped nothing burger!

Edit: I should add that all three presidents had the majority of Senate vote to convict at least one of the impeachment charges, but they needed a 2/3rds supermajority because why not?