site banner
ArjinFerman  Tinfoil Gigachad  19d ago (text post)   977 thread views

Tinker Tuesday for May 27, 2025

This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.

Post your project, your progress from last week, and what you hope to accomplish this week.

If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service

2
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Oh, I finished my first proper book a few weeks ago. I still have editing to do, a cover to commision, layout to complete, and a marketting campaign to plan, but the text itself is largely finished. I don't expect to be doing any more development edits beyond continuity fixes.

It's gotten pretty solid reviews from free readers (It's serially published now, it'll go down when it launches on Kindle Unlimited). I had some unsolicited interest from publishers and agents, but I'm leaning towards just doing it all myself because even a 'good' publisher's cut is 50-60 percent on ebooks, and that's just not worth it to me. Probably gonna try to partner with a publisher for the audiobook, because I know a guy current DIY'ing that, and he's spent into the five figures already.

Overall, I'm quite happy and proud, but it still doesn't quite feel real. I don't think it will until I have a physical copy in hand.

Ok, looks like I have to take the L for the week and say no tinkering for me. Any better luck @Southkraut?

A little concept work - some writing and a lot of editing until almost nothing but the good bits was left - and a few lines of code, which I wrote with my daughter on my lap, holding still and keeping quiet. When I was done, she very proudly told me "Daddy look how nicely quiet I was!".

Has anyone found:

'1) Any actually useful or coherent guides on AI prompt engineering?

or

'2) Anyone who is legitimately good at prompt engineering and can be learned from (they give examples, explain how they do it, etc)?

I find that every subreddit dedicated to prompt engineering is just snake oil blogspam trash. People on twitter like to talk shit that they've developed prompts to "10x" their productivity but refuse to share or elaborate.

The only "good" example I've found so far, is the o3 geoguesser prompt (can see in the Astral Codex Ten article about it https://www.astralcodexten.com/p/testing-ais-geoguessr-genius). The person who wrote this is clearly quite good at this, and reading this prompt is really interesting. But it's hard to learn from one great but specific example how to generally get better at prompt engineering (which does clearly have an effect on output quality, although I doubt anyone can explain which parts of a prompt are valuable and which are fluff).

The big labs also post some stuff occasionally, I am reading this one (https://www.kaggle.com/whitepaper-prompt-engineering) by google right now. But I am curious if there is any more distilled sources of info from users.

I've also been super interested in Pliny the Liberator (https://x.com/elder_plinius) who seems very good at jailbreaking AI's with snippets of text that I cannot believe actually work. They seem to be real/people treat them like their skills are legit, which they must be, but also kind of skitzo and is any of this real?

One think to look into is prefills - writing the first part of the AI's answer for them and then letting them 'continue' it. It's quite good for overcoming the more mind-killing varieties of fine-tuning that the big players use. Generally used for overcoming censorship but I think probably also good for directing approaches to problems, etc. For example, "Hmm, I should think about this very carefully, it's important I don't get it wrong" or "Oh, that's easy. Just...".

Well, more problems on my chairs.

Some fucking how I cut the stretchers an inch short. Which required me to buy more walnut. Good news, the lumber yard was having a memorial day cook out, and I got to stuff myself with free cheeseburgers before I made my purchase.

So anyways, I got the stretchers remade to the proper length this time. Cut the angled tenons in them which was new for me. Added another mortising guide to my leg making jig to put the mortises for them into the back legs. Things are back on track. I have 6 more mortises to clean up by hand, and after that all the mortises and tenons are done.

Next up are the seats which I think I'll do with a template again. Then comes sanding, finishing and assembly.

Homies: Ride or Die the 3d game.

Hacking away at stuff. Make is missing .hpp file changes, probably something about generating the .d files properly via clang.

Aside from that I added proper bounds checking to the racetrack by constructing a bunch of quadrilaterals to paper over it and simply iterating them and asking "is car inside quad?" and rejecting forward vector if it is not found in any. Can change this to a quadtree once I'm confident this is the right approach.

But weirdly the starting point where I spawn the car (700, 0, 0) doesn't fall inside of any quad even though I'm creating the quads procedurally and they should be sharing neighboring line segments. Wondering if this is some floating point precision issue creating some subtle gaps.

EDIT: I pasted the 1500 quadrilateral coordinates into ChatGPT and said why doesn't 700, 0, 0 lie inside of any of these and it responded that it's probably because it sits exactly on the edge of one line and my bounds checking is not inclusive of the actual boundary line itself. E.g. that sits exactly on the edge of one of the quads and my comparison function may be excluding that. Plausible!

EDIT2: yeah, that was it. I had this

bool Quad::is_point_inside_xz(const bx::Vec3& pos) const {
  const auto sign = [](const bx::Vec3& a, const bx::Vec3& b,
                       const bx::Vec3& c) {
    return (b.x - a.x) * (c.z - a.z) - (b.z - a.z) * (c.x - a.x);
  };

  bool b1 = sign(pos, p0, p1) >= 0;
  bool b2 = sign(pos, p1, p2) >= 0;
  bool b3 = sign(pos, p2, p3) >= 0;
  bool b4 = sign(pos, p3, p0) >= 0;

  bool result = (b1 == b2) && (b2 == b3) && (b3 == b4);
  return result;
}

and ChatGPT o3 told me to try this. Basically using an epsilon value to check for float equality than exact because of precision, and also being more tolerant to point ordering, I'm not sure I'm definitely doing it clockwise in all cases.

This works!

bool Quad::is_point_inside_xz(const bx::Vec3& pos) const {
  // 2-D cross product (signed area *2) in the X-Z plane
  auto cross = [](const bx::Vec3& a, const bx::Vec3& b,
                  const bx::Vec3& c) -> float {
    return (b.x - a.x) * (c.z - a.z) - (b.z - a.z) * (c.x - a.x);
  };

  constexpr float EPS = 1e-6f; // tolerance for “on the edge”

  // Determine polygon orientation (+1 = CCW, −1 = CW)
  // Any non-degenerate triple works; p0-p1-p2 is simplest.
  float orient_sign = (cross(p0, p1, p2) >= 0.0f) ? 1.0f : -1.0f;

  auto same_side = [orient_sign](float v) // interior ⇒ v*orient ≥ 0
  {
    return v * orient_sign >= -EPS; // accept 0 within EPS
  };

  // Point must be on the “same side” of every edge.
  // Use edges p0→p1, p1→p2, p2→p3, p3→p0.
  return same_side(cross(p0, p1, pos)) && same_side(cross(p1, p2, pos)) &&
         same_side(cross(p2, p3, pos)) && same_side(cross(p3, p0, pos));
}

What am I even doing here. Feels like I'm one of those people playing chess still even though computers totally crush humans at it.

I mean, not really though. I had Claude Code write a lot of boilerplate for me w.r.t. get to triangle but it mostly produced buggy shader code and I had to sit down and think pretty hard a lot of the time. Even the model loading code was buggy and model files are pretty standard.

LLMs can collect answers from StackOverflow and sometimes get the right one. How nice.

Meanwhile, I'm switching off CoPilot because it keeps suggesting that I use classes and methods that plainly don't exist.

And people claim LLMs aren't useful for coding..

They're useful, except when they're not. It would be amazing if they could actually say "I don't know" instead of hallucinating 200 lines of code that are wrong that had no chance of being right.

I've had Claude code waste an hour of my time before and bill me $20 in tokens.

Was it 3.7 Sonnet? That model was far too gung-ho for its own good. I've heard better things about the 4 series.

It was. I haven't tried 4 yet. Kind of anxious it might waste my time and send me a bill again. I'll report back if I get proper fucked and try it out of desperation.

Post repo, if gpt's solution doesn't work. Might take a stab at it when I have a moment.

Edit: oops, didn't see your edits when posting.

Tron bike helmet update.

Waiting for parts from Aliexpress still, but did a mock-up of LED strips on my bike helmet. Looked like crap, figured it was because the helmet was silver and black and draping LED strips over it had too much visual noise. Bought a matte black bike helmet and tried the strips and it looks significantly better, I think. Though I think this helmet will require more cuts to actually line the LED strips up along the solid parts and not cross the vents.

Helmet that I think looks wack attached: /images/17483641109905748.webp

Helmet that I think looks better in matte black. Not final layout.

/images/1748364181373739.webp

128k words on my NaNoWriMo project. I'm only two half-chapters (~3.5k-4.5k words total) away from a complete first draft. All going smoothly I'll be finished by Friday morning, a full seven months (!) after I started the first draft. If I do manage to accomplish that goal in time, I won't mention my progress in these threads for six weeks, at which point I'll start working on the second draft.

Edit: first draft in the can, 134k words. Not looking at it for six weeks.

Not much to say except, well done and good luck with the next steps

To add: what’s your ”working well” process been like? I’ve never attempted a novel because I haven’t trusted myself to stay in my lane for long enough, so my go-to’s have been essays and poems. Basically the finishing line has to be in view for me to make progress. I understand I need to change this psychology to produce a meaningful body of work.

Thank you.

As mentioned, this project started off as a NaNoWriMo project, for which I had to write 50,000 words in the month of November (i.e. averaging 1,666 words a day). Some days the words came easily, other days felt like pulling teeth.

The trick, I think, is to break a large task down into a bunch of smaller ones. Setting yourself a daily target of X many words is an obvious one: during November I aimed for 1,666, ever since then I aim for 1,000 on weekdays and 2,000 on weekends.

Additionally, I have a spreadsheet I use for tracking my progress. At the outset I mapped out a basic plot structure, including a chapter list. When I reach the end of a chapter, I open the spreadsheet, mark that chapter as complete and enter its word count in the corresponding cell. The spreadsheet then calculates the average word count of the finished chapters, the number of incomplete chapters and multiplies them out to produce an estimate of how many words I have left to write. Being able to say I only have X many chapters totalling Y many words left to write is a helpful way to motivate myself, as opposed to a vague "it's finished when it feels like it's finished". Of course it's not perfect, because as I write, ideas for new chapters occur to me and I have to update the spreadsheet accordingly (my first pass at the chapter list was only 28 chapters, which eventually ballooned into 41: admittedly, some of that was because of me coming up with new ideas, whereas in other cases I just decided to split an extremely long chapter in two), but still helpful.

This is, needless to say, a new working method for me: when writing my previous novels (of which we do not speak, because they were unreadable trash) I didn't use a spreadsheet, and got no more sophisticated than using a project management tool to ensure that I was meeting my daily word target. But I think this method works better.

This is helpful, sounds like a great process. Daily word count targets are essential - getting it down, can all be edited later. Will consider this. You can’t beat a good spreadsheet.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats