soultrees

Firefox till I die. (Or until they turn into a shit company)

Has anyone noticed that Copilot will shade out it’s answers more often when it’s writing code now? Usually I’ll paste in react components and ask it to fix the tailwind styling, but once it starts writing it gets filtered out by some secondary filter about half way through. I thought maybe the code it was outputting was too similar to copyrighted code and it triggered a liability filter of some sort.

In any case, super annoying to have that happen so consistently these days that I just use chatgpt to fix my tailwind styling now.

While not your typical 10x coder who's punching out 100k lines of code a year, I bring a lot of soft skills to the table from my previous work in sales, marketing, and design.

  Background:
  - Previously launched some of the largest cannabis brands in Canada in a highly-regulated market.
  - Currently working in the Vancouver Film Industry as a Key Grip (or what's left of it).
  - Using the strike downtime to brush up on my front-end and back-end skills.

Location: Vancouver

Remote: Preferred

Willing To Relocate: Yes

Email: hello@mattrichmond.ca

Resume: Available upon request

Technologies:

  Code:
   - JavaScript (generalist on my way to specializing)
   - TypeScript (generalist on my way to specializing)
   - Node.js (generalist on my way to specializing)
   - SQL (generalist)
   - LLM APIs (spent too many hours and too many dollars playing with OpenAI function calls this summer)
   - React (Next.js)

  Design:
   - Graphic design (specialist)
   - Data design (not quite a specialist)
   - Adobe CS (specialist in PS, AI, XD, ID, etc.)
   - DaVinci (Generalist)
   - Figma (not quite specialist)

  Field Skills:
   - Set Engineering (It's what we grips do all day)
   - Camera (understanding cinema and photography cameras on a technical level, and how certain params will affect the final shot, etc.)
   - Rigging (Rigging massive lights, massive diffs, over-top actors' heads)

Do you have any more information on this? Or any leads where to look.

And those bars aren’t going anywhere for those reasons you listed.

1 robot bartender =/= all bartenders becoming robotic.

Photography.

1. It forces you outside during daylight hours when the light is the best.

2. Taking a great shot will give you a shot of that dopamine you’re gunning for.

A good quote “ tripping with friends is bonding, tripping alone is therapy”. Both have their place and time and neither is better than the other.

You’re right, and I’m surprised the companies don’t have longer pay periods in the US actually. I also wonder if there’s an economic benefit to shorter pay periods? Because the faster that money is in the worker’s hands, then the faster it’s distributed again into the economy.

Monthly seems too long for me from a worker’s point of view though, you’re effectively giving the company a short-term, interest-free loan at that point. And when you factor in the fact that your purchasing power is reduced by the time you get paid, and any opportunity costs, the real cost to the worker can be quite high.

Unless the money is in to your hands or in your account, it’s not yours. In fact, every day in between the day you worked and the day you were paid could be considered theft - as the value of your cash has gone down slightly in that period due to official government policies to devalue cash at a certain percentage every year, and the company is earning interest (or investment) until it’s in your account.

Copilot is generally pretty awesome. Copilot chat gives me the most frustrating experience, and seems to be following the OpenAI path of “make the models worse, so we can get more reasoning data when the user has to clarify the question 5 times”. I find I have to ask it multiple times in a row now just to get the help I needed. I would be severely at a loss without these AI tools in my day to day but I’m tired of being a Guinea pig.

To be honest, that’s kind of what I already do. MacBook Air with a portable battery gives me about 12 hours. I live in BC so I’ll grab a coffee, drive up the mountain where I can get cell service (not needed in all cases) and just code away until I run out of battery, coffee or energy. The future is here.

How did you scrape that data? How do you store and retrieve it? Is it just a standard db or a vector db?

Sorry for the questions, but it seems like an interesting, yet probably common data set and as someone who is venturing down this path, I’d like to learn more about building my own dataset similar to this from scratch.

But does that include data already scraped? If not, then I don’t see how that will change. What will change however, is a dataset or api for sale, confirmed to be written by human authors which has value for anyone looking to train their model. Future data confirmed not to be written by AI will be invaluable.

To be fair to the op, prefacing as an LLM output was the right choice, and although I do have access to these same tools, it was nice to get a description and know possibly that it’s a little off the mark, but I’m further ahead than before.

To be fair to you, i agree. HN is probably the last place you want to use low-effort AI comments, regardless of how helpful they may be. Let’s leave the AI comments for Reddit.

“They say you die twice - first when your heart stops beating and the second after the last person on earth thinks about you”

Sun is looking pretty orange today in Vancouver.

You’ve given me something to think about and I think you’re right, there is an element of time missing. In fact there is a few parallel time elements missing from this too

For example, when we are shooting there is a rough formula for how long of a day we need to get those scenes. Usually it’s 1 hour per 1 page scene plus an extra 30 mins added on for each character in the scene. But that doesn’t translate to the final product as that information tells us nothing about how long or important a scene should be in the final product.

But it’s also possible I’m getting too ahead of myself here and maybe there’s another object that is created that includes the scene, production and final product objects instead of jamming it all into this object.

Open to anything suggestions you may have.

There are services that exist to hunt down any scene details like this and at this point is pretty standard. The thinking behind it is that even though most people won’t notice or care, the ones that dk notice will care, and if it’s wrong then it could be the single defining thing that pulls them out of the story. That and the pressure of the film industry in general and bringing bad press to a film. There’s always someone who notices every mistake and then they take to twitter over and on the other end of that message is some poor art director or continuity coordinator who’s losing their shit.

Perfect example, you seem to be someone specialized in a field that would exist somewhere in the same vector map as plumbing. Not plumbing but pipes, so therefore it’s in the studios interest to ensure the specialized feel recognized as well. Had it been so obviously fake CSI Miami style, you might not feel the same about the movie, or maybe you’d not care and still enjoy the movie but you sure wouldn’t be in HN singing it’s praise. So it’s all upside for the studio to ensure you don’t get pulled out of the story or at the very least don’t talk to your friends who also live in the same vector map as you. Not saying it’s wrong as it’s cool to see effort put into the creation process but it’s definitely thought of ahead of time.

I don’t think that’s a fair judgement. Sure, from the outside the decision to ignore medical advice seems crazy, but in the end it was his decision and he paid for it. Would a billion rounds of chemo have saved him? Maybe, but it’s also likely he would have ended up in the same spot anyways. We are all human, and doctors are just humans that are more well informed about the subject than others but are still human, and in the end it’s all just an experiment to see if this works better than that or that. So I think the lesson should be, if you get cancer then you should be able to finish your days, in your own way if you decide.

Here’s another project I worked on that I know will probably ruffle some feathers here. But to preface, I used it as way to learn Open AI function calls and - small concept of a zero to product launch full automation. This is a single page script that that generates an idea, creates an audience profile, then writes a book of short stories in a 5d array, looping over each level, then generates an image through midjourney through a hacked together api and creates a styled cover where there are a number of steps to ensure that aesthetically, the book cover is decent. It then uploads it to Amazon using puppeteer and uploads it to the woocommerce site as a new product.

https://github.com/mattrichmo/wisewriterv3

It’s my first complex app I’ve created that required multiple apis, and dealing with errors from multiple levels but I never actually uploaded a single book to Amazon so I can’t speak to the quality of the content. I was thinking if throwing this up on a website and letting people make a book for their child or something but I’d love the take the idea of zero-to-launch automation further. I’d love to go to bed and have a new well researched, highly desired product that actually serves a market need when I wake up sort of deal.

Timeline and frame based editing is the end result but this more about the elemental creation and from there, editing it into time based scene. Ive spent the last few years in the cinematography department, and most directors and director of photographers will write the scenes on flash cards and move them around and rearrange them because not every story is linear, even though we have to find a way to present every story in a linear form. And from there each scenes requires multiple angles, shots, motivations and things change so much on the fly, that a screenplay becomes a document that becomes quite dense with non-presented information.

So, I suppose the next step to this would be to parse a bunch of screenplays from different formats, into a single readable format and then train an image model on the frames of those movies we also trained the text model with screenplays on to get a cross reference of what is written down vs what is displayed visually. And we can break down the visual shots with camera movements, steadicam, dolly move etc as well as identify key props in the image model (maybe. Sounds expensive) and compare them to key props in the script. I don’t know, I’m spitballing now but a multi-modal Hollywood film producer would be kind of fun but this totally is just starting as a way to standardize the script in a granular form and to code since I’m not out on set.

Oh for sure, but since most screenplays are standardized to a point, the idea is to just use Regex to get there, then use a local version of llama 2 to finish up the classifications.

Where the fun with LLMs come in, is after all the screenplays have been parsed and turned into a dataset that is trained on not just the story but also the cinematography aspects, as well since we’ve broken it down to a granular detail of each element of each scene.

As someone who has been majorly affected by the Hollywood strikes, I finally dived in and spent the time needed to make mistakes to continually learn how to program.

That said, ive been working on a node.js PDF Text screenplay parser that will ultimately push out a JSON object according to an initiative to standardize all screenplay objects in screenJSON[0].

Here is the github for my project, it’s messy but I’m about 80% there and just need to spend more time on Regex to catch edge cases. Currently working on a next.js front end that maps out the script into a visual format but that’s not ready to be seen yet. Also don’t judge me on the name, it’s not going to stay as that. No AI as of yet, but could use text classifiers to be able to tell the difference between a key prop, and say, a camera movement but I’m trying to get as far as I can without LLM help of the classifications.

https://github.com/mattrichmo/ScreenBaby.

[0] https://screenjson.com.

I love draw.io but I’m surprised there isn’t a native iPad app. Diagraming apps are few and far in between on the iPad during my quick search last time I needed one. Ended up using lucidchart but I’d for sure use draw.io in the browser if it was more suited to my fat fingers.

Lately it’s even worse. And any AI agent that relies on open AI has gone down in quality too (phind). I find it’s more of a struggle to have the AI write code in the first pass, even when I explicitly say something like “I want you to write a function that does this {action}” and on average it takes about 3 messages back and forth before it actually writes it. Whereas before it would start writing it immediately, even though it was almost always wrong, I could tweak the result after. Now though, the stupid context memory is filled up with useless junk words not related to the actual function I want to write, since we spent multiple messages going back and forth.

That’s a business problem though, not a customer problem. And even more so now that this customer tried to move all their services, failed and has to move back to their old provider. This business will never ever get this customer back.

I went to use them today and their page was down. Crazy! First time I’ve seen them down in as many years as I’ve been using them.

From what I understand is that it’s not really close for anything involving creative reasoning but for basic instruct is on par.

I bet all the money will be returned. The people who pitched in to the bail fund are well connected and this trial isn’t meant to harm them (from the judges point of view).

But the fact that he’s in custody and isn’t on the run, means there’s more risk politically to forfeit the money than to give it back.

To prove that we are in a simulation.