Usage Rules: Leveling the Playing Field for AI-Assisted Development

Nice!

Can you point to API docs for ‘search.hexdocs.pm’?

I don’t know if they exist yet

Thank you Zach.
This is very interesting to me.

I am trying getting up to speed with AI assistance software development.

I installed ollama on my Mac mini M4 pro and would like to use it.

Could you give me some pointers on how to proceed best?

Cheers from Chrichton

As a point of feedback, starting with (emphasis mine):

We have a new paradigm available to us in software engineering, whether we like it or not.

is tremendously off-putting. Yeah, just shove it down our throats because whether we like it not it is happening.

Personally, I find most of the LLM efforts going on in the industry right now to be borderline abominations, if only due to the resource consumption and the often misleading rhetoric around purported benefits. But I also know that people will do as they will do, and don’t feel there is much point in crusading against it in places like this.

But if your hope is to broach the topic and find a place for it here despite a lack of general consensus on the benefits and costs, maybe don’t start with what amounts to ultimatums such as “whether we like it or not”, as if we are without agency (excuse the pun?) and technology is something that happens to us rather than due to us, that those who disagree must acquiesce and jump on your hype trains, or at the very least not impede them because, again, there’s no real choice.

Having to write that in response to a post that opines about how open source is about taking back the power and leveling playingfields should not be lost …

Personally I think you skipped an important part of that sentence. What I said was “available to us”. As an open source maintainer, I cannot stop people from using LLMs. So I have a vested interested in improving their usage of it.

Before the introduction of usage rules, I would get questions and issues daily about LLM hallucinations. Now, this happens far less often. I’ll update the introduction today to make it more clear. My point is not that we must use LLMs, but that they are now in use by many. OSS maintainers know that more than anyone

Local models are not very good right now AFAIK. I can’t advise much on how to use them as I’ve not really done it myself. I have another article on my substack about my general workflow.

@zachdaniel can you please share a link to that article? As for local LLM I see that maybe some industries will force it later either due to security or compliance issues plus cost as an example Banking in case it will make use of AI channel to serve customer, think of Ash AI prompts instead of regular chat banking

I like >90% of what I read from you but in this case I’m with @aseigo; the words “whether we like it or not” are aggressive and off-putting.

A bit less thorny framing would be: “These tools are a fact of life and are heavily used” or some such.

I’m comfortable with the framing. AI tools are a part of the industry. I can’t say that they are a net good. But pretending that they don’t exist, or deciding not to be exposed to them just isn’t an option at this point, at least for a large portion of software engineers. “Wether we like it or not” is in fact exactly the point I’m trying to make.