Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A few random notes from Claude coding quite a bit last few weeks

twitter.com

Post reply on HN
Print view View original

761–770 of 870 posts

Re: A few random notes from Claude coding quite a bit last few weeks

#761

Post by chrisjj »

jjfoooo4 wrote: post #702
chrisjj wrote: post #535

Earlier quoted context omitted.

Contract? These docs are information answering user queries. So if you use a chatbot to generate them, I'd like to be reasonably sure they aren't laden with the fabricated misinformation for which these chatbots are famous.

It's a very reasonable concern. My solution is to have the bot classify what the message is talking about as a first pass, and have a relatively strict filtering about what it responds to. For example, I have it ignore messages about code freezes, because that's a policy question that probably changes over time, and I have it ignore urgent oncall messages, because the asker there probably wants a quick response from…

OK, but little of that applies to this use case, to "then tell it to update the documentation accordingly."

Re: A few random notes from Claude coding quite a bit last few weeks

#762

Post by vardalab »

kejaed wrote: post #103
vardalab wrote: post #93

Earlier quoted context omitted.

I'm almost a boomer and I agree. THis dichotomy is weird. I am retired EE and I love the ability to just have AI do whatever I want for me. I have it manage a 10 node proxmox cluster in my basement via ansible and terraform. I can finally do stuff I always wanted but had no time. I got sick of editing my kids sports videos for highlights in Davinci Resolve so just asked claude to write a simple app for me and then us…

So what is your workflow now with this app for kids sports highlights?

Well, it's not really a full-blown app yet. Claude wrote a plugin for MPV. So now when I watch video I just push a button to mark in and out of highlights similar to how it works in DaVinci Resolve. Then I have a command line tool that takes those timestamps in a video file and cuts it up into individual clips and then re-renders those clips and creates a highlight reel. Another command line tool takes three or four large MP4 files that the camera generates and downloads them and combines them in the actual game video on my desktop and also uploads it to my archive and transcodes into a bunch of different formats and uploads to YouTube. And for transcoding, again, it divvies it out to the video cards, which works pretty well. I think I have five or six encoders available so it chunks it up and then reassembles. All in all, it's nothing fancy, but it reduced quite a bit the friction of coming home after games and getting a video up on YouTube for grandparents.

Re: A few random notes from Claude coding quite a bit last few weeks

#763

Post by fragmede »

lelanthran wrote: post #757
sailfast wrote: post #204

Earlier quoted context omitted.

That’s far-fetched. It’s in the interest of the model builders to solve your problem as efficiently as possible token-wise. High value to user + lower compute costs = better pricing power and better margins overall.

> It’s in the interest of the model builders to solve your problem as efficiently as possible token-wise. High value to user + lower compute costs = better pricing power and better margins overall. It's only in the interests of the model builders to do that IFF the user can actually tell that the model is giving them the best value for a single dollar. Right now you can't tell.

Why not? Seems like you'd just build the same app on each of the models you want to test and judge how they did.

Re: A few random notes from Claude coding quite a bit last few weeks

#764

Post by charcircuit »

direwolf20 wrote: post #751
charcircuit wrote: post #267

Earlier quoted context omitted.

>"Reasoning", however, is a feature that has been bolted on with a hacksaw and duct tape. What do you mean by this? Especially for tasks like coding where there is a deterministic correct or incorrect signal it should be possible to train.

it's meant in the literal sense but with metaphorical hacksaws and duct tape. Early on, some advanced LLM users noticed they could get better results by forcing insertion of a word like "Wait," or "Hang on," or "Actually," and then running the model for a few more paragraphs. This would increase the chance of a model noticing a mistake it made. Reasoning is basically this.

It's not just force inserting a word. Reasoning is integrated into the training process of the model.

Re: A few random notes from Claude coding quite a bit last few weeks

#765

Post by jrowen »

jatari wrote: post #717
direwolf20 wrote: post #706

Earlier quoted context omitted.

Because of the collapsing empire, mind you, not because of the LLMs.

Creation of the internet, social media, everyone on the planet getting a pocket sized supercomputer, beginning of the AI boom, Trump/beginning of the end of the US, are all reasons people will study this period of time.

This is really interesting because I wholeheartedly believe the original sentiment that everyone thinks their generation is special, and that "now this time they've really screwed it all up" is quite myopic -- and that human nature and the human experience are relatively constant throughout history while the world changes around us.

But, it is really hard to escape the feeling that digital technology and AI are a huge inflection point. In some ways this couple generations might be the singularity. Trump and contemporary geopolitics in general is a footnote, a silly blip that will pale in comparison over time.

Re: A few random notes from Claude coding quite a bit last few weeks

#766

Post by lelanthran »

fragmede wrote: post #763
lelanthran wrote: post #757

Earlier quoted context omitted.

> It’s in the interest of the model builders to solve your problem as efficiently as possible token-wise. High value to user + lower compute costs = better pricing power and better margins overall. It's only in the interests of the model builders to do that IFF the user can actually tell that the model is giving them the best value for a single dollar. Right now you can't tell.

Why not? Seems like you'd just build the same app on each of the models you want to test and judge how they did.

> Why not? Seems like you'd just build the same app on each of the models you want to test and judge how they did.

I tried that on a few problems; even on the same model the results have too much variation.

When comparing different models, repeating the experiment gives you different results.

Re: A few random notes from Claude coding quite a bit last few weeks

#767

Post by Applejinx »

AnimalMuppet wrote: post #661
rubzah wrote: post #639

The Slopocalypse - an unexpected variant of Gray Goo: https://en.wikipedia.org/wiki/Gray_goo

Well, it may consume the AI environment. Maybe even the internet. It's not going to consume a PC with g++, though (at least if the PC doesn't update g++ any more once g++ starts accepting AI contributions). There may come a point where having a "survivor machine" with auto-update turned off may be a really good idea.

I already do this, in the form of survivor machines made to do initial coding on a retro platform so the result will translate across all possible platforms. Got to, as I'm an Apple coder primarily, so if I want to target older machines I can only do it through a survivor machine: support is always pruned out of Xcode and it would be insane to try and patch it to keep everything in scope.

Re: A few random notes from Claude coding quite a bit last few weeks

#768

Post by Philpax »

chrisjj wrote: post #746
chickensong wrote: post #719

Earlier quoted context omitted.

That split has always existed. LLMs can be used on either side of the divide.

We see a ton of "AI let me code a program X faster than ever before." We see almost no "AI let me code a program X better than ever before."

See this episode of Oxide and Friends, where they discuss just that: https://oxide-and-friends.transistor.fm/episodes/engineering...

Re: A few random notes from Claude coding quite a bit last few weeks

#769

Post by oblio »

runarberg wrote: post #662
oblio wrote: post #655

Earlier quoted context omitted.

I'm not a big fan of LLMs, but while using it for day to day tasks, I get the same feeling I had when I first started the internet (I was lucky to start with broadband internet). That feeling was one of empowerment: I was able to satisfy my curiosity about a lot of topics. LLMs can do the same thing and save me a lot of time. It's basically a super charged Google. For programming it's a super charged auto complete co…

When I first started using the internet, I was able to instant text message (IRC) random strangers, using a fake name, and lie about my age. My teacher had us send an email to our ex-classmate who had move to Australia, and she replied the next day, I was able to download the song I just heard on the radio and play it as many times as I wanted on my winamp. These capabilities simply didn’t exist before the Internet.…

Before LLMs it was incredibly tedious or expensive or both to get legal guidance for stuff like taxes, where I live. Now I can orient myself much better before I ask an actual tax expert pointed questions, saving a lot of time and money.

The list of things they can provide is endless.

They're not a creator, they're an accelerator.

And time matters. My interests are myriad but my capacity to pass the entry bar manually is low because I can only invest so much time.

Re: A few random notes from Claude coding quite a bit last few weeks

#770

Post by gregjor »

akoboldfrying wrote: post #552
gregjor wrote: post #516

Earlier quoted context omitted.

I don't understand why anyone finds it interesting that a machine, or chatbot, never tires or gets demoralized. You have to anthromorphize the LLM before you can even think of those possibilities. A tractor never tires or gets demoralized either, because it can't. Chatbots don't "dive into a rabbit hole ... and then keep digging" because they have superhuman tenacity, they do it because that's what software does. If…

You're a machine. You're literally a wet, analog device converting some forms of energy into other forms just like any other machine as you work, rest, type out HN comments, etc. There is nothing special about the carbon atoms in your body -- there's no metadata attached to them marking them out as belonging to a Living Person. Other living-person-machines treat "you" differently than other clusters of atoms only bec…

Wrong level of abstraction. And not the definition of machine.

I might feel awe or amazement at what human-made machines can do -- the reason I got into programming. But I don't attribute human qualities to computers or software, a category error. No computer ever looked at me as interesting or tenacious.

Post reply on HN