kevin42

Have you considered that maybe you aren't using it well? It's something that can and should be learned. It's a tool, and you can't expect to get the most out of a tool without really learning how to use it.

I've had this conversation with a few people so far, and I've offered to personally walk through a project of their choosing with them. Everyone who has done this has changed their perspective. You may not be convinced it will change the world, but if you approach it with an open mind and take the time to learn how to best use it, I'm 100% sure you will see that it has so much potential.

There are tons of youtube videos and online tutorials if you really want to learn.

Hi, Author here! I wrote this piece after some conversations with friends and realizing how we all had some levels of cognitive dissonance towards AI, IP, etc. I noticed I was moving my own goalposts both when criticizing AI or defending it.

I have to admit that I'm less of a skeptic than most, but there are some coherent skeptical arguments. I'm especially interested in what people think about their own skepticism on the technical side, as there seems to be a pivot towards the social lately.

You don't need to do much, the /agent command is the most useful, and it walks you through it. The main thing though is to give the agent something to work with before you create it. That's why I go through the steps of letting Claude analyze different components and document the design/architecture.

The major benefit of agents is that it keeps context clean for the main job. So the agent might have a huge context working through some specific code, but the main process can do something to the effect of "Hey UI library agent, where do I need to put code to change the color of widget xyz", then the agent does all the thinking and can reply with "that's in file 123.js, line 200". The cleaner you keep the main context, the better it works.

Not that I have seen, which is probably a big part of the disconnect. Mostly it's tribal knowledge. I learned through experimentation, but I've seen tips here and there. Here's my workflow (roughly)

> Create a CLAUDE.md for a c++ application that uses libraries x/y/z

[Then I edit it, adding general information about the architecture]

> Analyze the library in the xxx directory, and produce a xxx_architecture.md describing the major components and design

> /agent [let claude make the agent, but when it asks what you want it to do, explain that you want it to specialize in subsystem xxx, and refer to xxx_architecture.md

Then repeat until you have the major components covered. Then:

> Using the files named with architecture.md analyze the entire system and update CLAUDE.md to use refer to them and use the specialized agents.

Now, when you need to do something, put it in planning mode and say something like:

> There's a bug in the xxx part of the application, where when I do yyy, it does zzz, but it should do aaa. Analyze the problem and come up with a plan to fix it, and automated tests you can perform if possible.

Then, iterate on the plan with it if you need to, or just approve it.

One of the most important things you can do when dealing with something complex is let it come up with a test case so it can fix or implement something and then iterate until it's done. I had an image processing problem and I gave it some sample data, then it iterated (looking at the output image) until it fixed it. It spent at least an hour, but I didn't have to touch it while it worked.

Trying to one-shot large codebases is a exercise in futility. You need to let Claude figure out and document the architecture first, then setup agents for each major part of the project. Doing this keeps the context clean for the main agent, since it doesn't have to go read the code each time. So one agent can fill it's entire context understanding part of the code and then the main agent asks it how to do something and gets a shorter response.

It takes more work than one-shot, but not a lot, and it pays dividends.

This isn't meant as a criticism, or to doubt your experience, but I've talked to a few people who had experiences like this. But, I helped them get Claude code setup, analyze the codebase and document the architecture into markdown (edit as needed after), create an agent for the architecture, and prompt it in an incremental way. Maybe 15-30 minutes of prep. Everyone I helped with this responded with things like "This is amazing", "Wow!", etc.

For some things you can fire up Claude and have it generate great code from scratch. But for bigger code bases and more complex architecture, you need to break it down ahead of time so it can just read about the architecture rather than analyze it every time.

I think everything you said was true 1-2 years ago. But the current LLMs are very good about citing work, and hallucinations are exceedingly rare. Gemini for example frequently directs you to a website or video that backs up it's answer.

Compared to what though? I have ended up with needlessly convoluted solutions when learning something the old-fashioned way before. Then over time, as I learn more, I improve my approach.

Not everyone has access to an expert that will guide them to the most efficient way to do something.

With either form of learning though, critical thinking is required.

Is it just me, or does anyone else use AI not just to write code, but to learn. Since I've been using Claude I've learned a lot about Rust by having it build things for me, then working with that code. I've never been a front end guy, but I had it write a Chrome plugin for me, then I used that code to learn how it works. It's not a black box to me, but I don't need to look up some CSS stuff I've never used. I can prompt Claude to write it and then I can look at it then "Huh, that's how it works". Better than researching it myself, I can see an example of exactly how it's done, then I learn from that.

I'm doing a lot of new things I never would have done before. Yes, I could have googled APIs and read tutorials, but I learn best by doing, and AI helps me learn a lot faster.

My wife works for a small distributor of memory (not modules, just the chips), and their manufacturers won't even provide prices or take backorders.

I understand why they did it in the current market, but Micron's exit is going to cause retail prices to go even higher.

I used pcpartsbuilder to spec out a machine I was thinking of building a few months ago, and the DRAM that was $109 for 32GB is now $399. It's crazy.

That's a pretty condescending take.

To some extent I agree that the upgrade path is lacking. I recently helped a friend move out of the ino file model into building regular c++ applications because his design was getting pretty complicated. Once he realized that he knew more of c++ than he thought he did, it was a game changer for him.

At the same time, people have done some pretty amazing stuff using the Arduino platform without knowing how to use the things you mention. What you call eternal beginners have accomplished a lot. James Bruton does some pretty impressive robotics work using Arduino.

That's my experience as well. The LLM is great for what I consider scaffolding. I can describe the architecture I want, some guidelines in CLAUDE.md, then let it write a bunch of stubbed out code. It saves me a ton of time typing.

It's also great for things that aren't creative, like 'implement a unit test framework using google test and cmake, but don't actually write the tests yet'. That type of thing saves me hours and hours. It's something I rarely do, so it's not like I just start editing my cmake and test files, I'd be looking up documentation, and a lot of code that is necessary, but takes a lot of time.

With LLMs, I usually get what I want quickly. If it's not what I want, a bit of time reviewing what it did and where it went wrong usually tells me what I need to give it a better prompt.

You made a lot of bold assertions there. It's as if you have a complete and definitive theory of human intelligence to compare it against. Which if true, would be incredible, because there isn't a scientifically accepted theory, nor is there consensus from a philosophical standpoint.

I can't say that you are wrong, you might be right, especially about AGI. And I think it's unlikely that LLMs are the direct path to AGI. But, just looking at how human brains work, it seems unlikely that we would be intelligent either if we used your same reductionist logic.

An individual neuron doesn't "think" or "understand" anything. It's a biological cell that simply fires an electrochemical signal when its input threshold is met. It has no understanding of language or context. By your logic, since the fundamental components are just simple signal processors, the brain cannot possibly learn or be intelligent. Yet, from the complex interaction of ~86 billion of these simple biochemical machines, the emergent properties of thought, understanding, and consciousness arise.

Dismissing an LLM's capabilities because its underlying operations are basically just math operating on tokenized data is like dismissing human consciousness because it's "just electrochemistry" in a network of cells. Both arguments mistake the low-level mechanism for the high-level emergent phenomenon.

The same thing came up in the early 2000s when businesses tried offshoring software development. It turns out that translating business requirement to technical design and implementation is one of the hard parts of software development. In a pure waterfall model, this works ok, but you rarely know what you want until the end users see it. This is while agile is so important.

I hate to say it, but with over 30 years experience as a software developer in various roles, writing code has never been the hard part of delivering a successful project.

I think this is a really cool project, but the problem with putting so many peripherals on such a small processor is that it's really tough to have firmware that uses all of those things at once and fit in the memory footprint.

I can see this as a great platform for prototyping though.

There are more than one theories, as well as some evidence that consciousness may not exist in the way we'd like to think.

It may be a trick our mind plays on us. The Global Workspace Theory addresses this, and some of the predictions this theory made have been supported by multiple experiments. If GWT is correct, it's very plausible, likely even, that an artificial intelligence could have the same type of consciousness.

My issue is that from a scientific point of view, physicalism is all we have. Everything else is belief, or some form of faith.

Your example about relativity is good. It might have sounded insane at some point, but it turns out, it is physics, which nicely falls into the physicalism concept.

If there is a falsifiable scientific theory that there is something other than a physical mechanism behind consciousness and intelligence, I haven't seen it.

I like the distinction you made there. My observation that when it comes to AGI, there are those who are saying "Not possible with the current technology." and "Not possible at all, because humans have [insert some characteristic here about self awareness, true creativity, etc] and machines don't.

I can respect the first argument. I personally don't see any reason to believe AGI is impossible, but I also don't see evidence that it is possible with the current (very impressive) technology. We may never build an AGI in my lifetime, maybe not ever, but that doesn't mean it's not possible.

But the second argument, that humans do something machines aren't capable of always falls flat to me for lack of evidence. If we're going to dismiss the possibility of something, we shouldn't do it without evidence. We don't have a full model of human intelligence, so I think it's premature to assume we know what isn't possible. All the evidence we have is that humans are biological machines, everything follows the laws of physics, and yet, here we are. There isn't evidence that anything else is going on other than physical phenomenon, and there isn't any physical evidence that a biological machine can't be emulated.

>but obviously, there seems to be more than that.

I don't see how that's obvious. I'm not trying to be argumentative here, but it seems like these arguments always come down to a qualia, or the insistence that humans have some sort of 'spark' that machines don't have, therefore: AGI is not possible since machines don't have it.

I also don't understand the argument that "Your nuanced, subtly ironic and self referential way of formulating clearly suggests that you are not a purely algorithmic entity". How does that follow?

What scientific evidence is there that we are anything other than a biochemical machine? And if we are a biochemical machine, how is that inherently capable of more than a silicon based machine is capable of?

It's not a matter of whether it's taxed. It's a matter of how the costs are expensed. If I invest a $1m in software development now, I'm not making any profit yet. I may not make a profit. It's a question over whether I need to pay taxes on potential future profit now.

I was flying in and out of a private airport in uncontrolled airspace, so I didn't check. If was my first test flight so I was pretty much just circling around my airstrip.

If I'm flying cross country, in to controlled airspace, or an airport I'm not familiar with, I will. But this particular notam didn't show up when I looked at nearby airports.

I have an amateur built experimental airplane, and on my first flight, GPS wasn't working. It was working on the ground, so I figured a bad coax cable or something from the vibration. When I landed it started working again, but later in the day it wasn't. A few days later I found out that the nearby Air Force base was testing GPS jamming, and there was even a NOTAM about it.

There's a site that tracks GPS jamming: https://gpsjam.org/

I'm curious why not. I am running a few different models on my mac studio. I'm using llama.cpp, and it performs amazingly fast for the $7k I spent.

IMO, the real value of ROS is the data logging (bag/mcap files), visualization (FoxGlove) are the main value of ROS. Even then, I'm not sure it's worth the overhead and brittleness of building and running it. There is just so much complexity deploying and developing for it.

One decision is by the fourth circuit, the other is the fifth. I don't know enough about the details, but in situations like this, the Supreme Court takes the case to resolve discrepancies.

One of our cats was very old (18+) and sick. We made the choice to have her euthanized to end her suffering and the vet came to our house to do it. One of our other (male) cats loved her and always tried to cuddle her. My wife took the old kitty out of the room where they were cuddling, and after it was done she thought she would let him say goodbye. He got very upset at my wife and hissed at her for days. He's normally very sweet with my wife, but he was clearly distraught at losing his friend and seeing her dead. He didn't just think she was sleeping, and he blamed my wife.

I've always known our cats have emotions and understand a lot about the world they're in, but after that experience, I have no doubt that they are conscious.

I have a Vantage Pro2 with the WeatherLink Live. I'm really happy with it. It has a pretty simple API you can hit on the local network, and it can push to weatherlink.com. It's a bit on the pricey side, but I live in a fairly rural area, so it's nice to have my own data. Plus, I'm on a private airstrip, so other pilots can use see it.

My concern isn't about competition but that since the drones dial home and get over the air updates, there's little that would prevent China from using it to gather images and location data that that we wouldn't want them to (private/commercial/military).

Try not to read too much into that, the MIT license was in the original blank repo I started with. There are a few questions I need to get answered, so the lack of that file doesn't necessarily mean anything about the license. The previous Descent 1&2 releases were under the license seen here, I would expect that at minimum (minimum permissibility-wise):

https://github.com/OpenSourcedGames/Descent-2

I'm hoping the owners will be able to agree to MIT, but we're doing due diligence right now.

I was also one of the Kahn developers back in the day, before I worked on D3. :)