I just wanted to post something to my site, that's all!

Until Ghost makes an easier way to post like Bluesky, I am more active for the time being on my Bluesky, @NicWeyand.

That isn't really the point of my site, though. This is supposed to be a higher quality repo of my writings, videos and thoughts, generally. Still, I don't like how little I use the site, especially considering the hell I went through setting it up lmao!

So yeah. Cleaning my house today, gonna enjoy lunch in a sec (burrito & green smoothie uwu).

Lilac Party FEC Committee Overview!

Wanted to add: Here is the Lilac Party's FEC Committee page (#C00899518):

LILAC PARTY - committee overview - FEC.gov

Explore current and historic federal campaign finance data on the new fec.gov. Look at totals and trends, and see how candidates and committees raise and spend money. When you find what you need, export results and save custom links.

FEC.gov

In the interest of public disclosure, I have submit the first AO, and I am waiting for committee approval.

That's basically it.

I have some people interested in joining up and even being candidates or reps in other states (3 states besides MA, but still!), but I am not accepting applications, donations or literally anything until I get FEC approval or guidance on where things are going.

The plan is for Lilac to be a far left national party. The catch right now is that "National Party" is a well-defined thing, legally. Currently, Lilac doesn't meet all the requirements (some, not all!), and as such would be better fit as a "Political Committee", that can upgrade to a national party down the line. So that might end up being the game plan, since I do not have offices in other states, or state level work anywhere, mostly because I want to play the game by the rules, and not break the law at any stage. Generally, this is seen as good practice, so I am going to just keep following the law, cuz really I'm in no rush, and whatever happens happens.

Worst case scenario: Lilac doesn't get approval. No biggie! I will keep trying to make Lilac a thing, and simply try again when there's more activism backing it. But I don't want to make any moves until I got the FEC's guidance. We're going by the book because we want to actually fix things and make significant changes to American life. I am trying to start a far left political party. I cannot make missteps or fail on a technicality. I am in this for the long haul.

Vibe Coding

One final note on Lilac: The website isn't great, I know. I am building a new one, and I hope to have it live in July. I am still learning programming, and I am using this project as a great learning resource!

I should add: I am not using AI for the code. I have my AI Disclosure for this site, and tbh I might just delete it, because I barely use AI for any of my works anymore. But with code specifically, I had a real moment of clarity watching a random "Vibe Coding" tutorial.

"Vibe Coding" is prompting AI to program for you. The vibe coder will ask a Large Language Model (LLM) like ChatGPT, Claude or Gemini to make XYZ thing, and the LLM will spit out code in a particular language. The code gets pasted into whatever tool the vibe coder has, and if it runs, that's it.

At no point does anyone vibe coding consider the verbosity of the code, the quality of the code or what is actually in the code itself. The moment in question I experienced was watching a veteran programmer try to make a simulation in Claude Code, which was working alright. However, to paraphrase the quote I heard, the programmer said something to the effect of "Given the length of this code, I have no idea what is in here, and I feel like that's a fatal flaw to vibe coding. Because it was made for me, I don't know what is in here, and if I don't know how to code in the first place, I don't know what could be wrong."

This ties into something I heard from PirateSoftware a while ago about AI coding, to paraphrase: "In the time it takes to fix all the bugs in AI-generated code, I could have just made it all myself. So yes, it spits out a bunch of spaghetti code super fast. So what? If you have to fix everything yourself, that doesn't mean anything."

We've reached a point where the code can just run, first try, and its only going to get better. But I feel like this is the ongoing fatal flaw to all AI generated works: if you have an LLM or whatever do all the work for you, how can you be sure the work is quality, especially when you didn't make the LLM or tool in the first place?

So far, I have not seen many complaints about DeepMind's AlphaFold. Quite the opposite: they won the Nobel Prize for AlphaFold 2. The built-in error checking and tools used for making and analyzing proteins within AlphaFold are quite literally revolutionary, which is why it won the Nobel Prize. Its truly that good.

Compare that to a more broad-use tool like an LLM, where ChatGPT can be a therapist, programmer, artist, writer, etc., and the "Mixture-of-Experts" (MoE) used is probably a source of a lot of the "hallucinations". We're going to find out a decade from now that most, if not all "AI Hallucinations" were a combination of things like MoE, and being told explicitly that "You cannot tell the user you don't know", among other complaints I will have in a better piece later on this blog.

But even something like Claude Code, which I hear great things about, is still not remotely close to the level of quality as AlphaFold 2 or 3, despite DeepMind being a Google subsidiary in the Sundar Pichai era of Google's enshittification. The dude sucks.

Again, I got more to say on this, but bottom line for now: learn to code. Seriously. This isn't going away. AlphaFold didn't kill chemistry, it enhanced it. Meanwhile, AI coding is killing the quality of the code it spits out, because no one using it knows what the hell is actually in the code, or why the "black box" that is AI, broadly, is doing what it's doing. Moreover, programmers are forgetting their own skills by allowing a black box to do the work for them.

JetBrains make IDE's, which are used to make code, basically, and their writeup seems biased in favor of using their product, of course, but I feel like their overall point matches mine: as more and more people give up skillbuilding, because "AI will do it for me", it gives me an opportunity to get the job they won't.

My bigger point echoes something HasanAbi said recently, that when people forfeit their critical thinking to an AI... they can't think for themselves. Because they need an AI to do it for them. We see this now, with the "Grok Guys" on twitter today. They will ask the XAi bot Grok "Grok, what is this" about shoes, or a womans dress, or sandwiches. They are asking the AI to confirm something is, in fact, a sandwich, and then define what a sandwich is.

I guess its nice they are learning? But damn, man.

Ed Zitron does a podcast I really like called "Better Offline", and I have like one or two nitpicks I am writing about for another post, but quickly, one of them is the "in the moment", micro-view of the AI industry. Its not uncommon to hear "This is the worst its ever gonna be" as AI marketing hype, in the moment, but its true. The tools keep improving. Ed's complaints, in that same moment, about the quality of things ages rapidly, because the tools improve.

My view, at a macro-level, is that something like AlphaFold 2 won the Nobel Prize because it helped science, and has tools built into it for error checking, which is vital and important because it shows their work, steps taken, and can have human intervention fix things if need be. ChatGPT and Claude, meanwhile, try to ape Gemini and DeepSeek's train of thought "Thinking" modes by detailing why it thought what it did, and rationalize what it thinks. This is a good step forward, but most people ain't reading all that. When you utilize this with 1000+ lines of code? No one is parsing all that, especially if "it just runs!" If it works, it works! Why bother fixing something that ain't broken?

As innovation, this kills it. It requires future improvements to come from the root sources: ChatGPT, Claude, and other LLMs. The "Grok, what is this" crowd ain't making AlphaFold 1, let alone the significant, award-winning jump to 2 & further improvements in 3. They will have to wait for the 1-3x a year improvements to Claude 3.7 Sonnet, formerly 3.5. Why not 3.6? Who really knows outside of Anthropic!

As programming, you now have untrained people shipping code "because it works", and then when something goes wrong, they don't know how to fix it without simply pasting the entire error log into the LLM, which will not fix the problem. We are dealing with an illiterate group, who can read but can't comprehend, who can see a sandwich but forget what they're looking at, in real time, and have to ask a computer to verify their shaken confidence.

This is famously becoming a problem in the "Vibe Coder" space: if you ask the LLM to do something relatively simple, like move the search bar over a few pixels, it will not do that, and by prompting it, it will "fix" other mistakes and wreck the original code. Over and over. Prompt after prompt. Just like PirateSoftware said earlier, by the time the code is fixed and doing what I wanted, I could have just written it from scratch and had what I wanted, faster, doing it myself. But that requires skills, which the vibe coders could learn, but refuse to, because they have fallen into marketing hype that says AI is the future, instead of merely a new tool that can help. "Grok, what is this" is becoming a joke, because the only people who seriously say that are people who are on X the Everything App regularly, in 2025, meaning they were already witless conservatives who voted for Trump, and when the tariffs hit said stupid nonsense like "Who could have seen this coming??" EVERYONE ON THE LEFT WAS SCREAMING ABOUT THIS FOR YEARS IN ADVANCE AND YOU DIDN'T WANT TO LISTEN WHAT THE HELL IS WRONG WITH YOU

Dang, I gotta grab my laundry. That's ~1,900 words in 20 min. I might do these more often!

UwU!