hgrsd

Does anyone know if there is a zon-parsing CLI tool a la jq? I'd like to parse out some fields in a shell script without having to resort to sed (though it works OK). #zig #opensource

This is the kind of side project I can get behind https://github.com/auchenberg/volkswagen

One interesting thing of the way https://github.com/hgrsd/seminar works, is that it is not at all sycophantic. In fact, it very often tells me (in slightly pompous words), that I am a fucking idiot and my ideas are bad. Often with sources showing that "the industry" has roundly rejected my stupid ideas a few decades ago already.

I guess that's a pretty useful way of using AI? Don't worry, I'm still ignoring its warnings most of the time. #tech #ai #llm #softwareengineering

I recently built "seminar" as somewhat of an experiment (https://github.com/hgrsd/seminar).

I wanted to see if letting LLMs, with access to tools like web search, PDF reading, etc, could build a "body of studies" based on a given, fairly rough and ambiguous idea that I gave them. The idea being that those studies could build on top of each other, contradict each other, and end up with a rounded picture and new insights.

I also gave the worker processes the option of starting a new conversation thread with me for things it "wanted" to call my attention to, or suggestions for things I should read myself.

One design goal was to build something AI-based that rather than stop you from thinking, tries to make you think more. I think that it has succeeded in a fairly narrow sense: I've gained lots of interesting things to read as a result of these sourced studies. But I have not seen anything novel or interesting. In fact, the capability it has to come up with new research ideas based on "synthesising" studies from existing ideas generally leads to totally outlandish ideas that make very little sense, and which are presented extremely pompously as though they are worthy of a nobel prize - it's kind of funny.

Will I be using this seriously? Probably not. But it's definitely helped me understand the limits of LLMs in generating anything new, as well as their usefulness as a starting-off point for your own thinking.

#ai #research #llm #tech

Now, I have no experience using online dating or dating apps, so you can discount my view for a lack of expertise... but...

Do we really want AI-powered dating assistants in another move to extract more shareholder values out of love lives?

(This relates to "Bee", Bumble's LLM-based dating assistant / matchmaker).

All of the work on agent harnesses and harness engineering makes me think a bit about the older concept of genetic algorithms.

You define fitness functions (test suites, code quality analyses, architectural tests, linting, etc) and let the LLM iterate/evolve until the fitness function returns something above the threshold. #tech #llm #ai

A lot of the mystique of agentic #LLM use goes away if you think of it as general-purpose, slightly indeterministic orchestration layer. It makes the orchestration of lots of different workflows cheaper (in terms of not needing to build specialised tools, not in terms of token use), with quality being lower than a purpose-built orchestrator.

Random realisation while reading to my daughter: so many children's books have an arc where initially someone (child/monster/animal) is rejected by their group because of some mild "oddity" or difference, only to later be accepted and the bigotry to dissipate once they have somehow proven their value. At which point, the former bigots realise the error of their ways and become much more tolerant.

I guess I get what they're trying to do in terms of moralistic plot, but surely we can imagine a world where we don't need to first prove ourselves to the bigots until they think we're worthy?

#childrensLiterature #literature

Does anyone have a spare lobste.rs invite to hand out? :) #tech #softwareengineering #software #programming

A product manager from a different team, who doesn't know how to program, opened a PR in one of our team's repos.

My initial response is that our team cannot accept this, because it forces us to choose 1) doing the work ourselves to read it line-by-line knowing the person creating the PR does not understand it themselves, and amending it until we are happy with the standard and are willing to accept it as hour own code; or 2) taking a quick glance, merging it because it "looks fine", and incorporating functionality that neither us as reviewers, nor the proposer as nominal author, understands.

I feel like a total dinosaur but it also feels like a shift past developer-steered Claude output that somehow strikes me as even more problematic. Thoughts? #ai #llm #programming

Even if you don't take what LLMs say as gospel, and try to 'think critically' and stay in control, it is SO easy for AI-generated outputs to limit your considered design space. You'll be tricked into critiquing and refining what it comes up with, rather than thinking about your preferred approach from the bottom-up, which guarantees that you too will end up gilding a lowest-common-denominator approach while feeling like you're the one calling the shots.

You really have to guard against this and, if you use LLMs, make sure that you do your thinking first, by yourself and with your colleagues, until you have a very clear mental model of what you want to do and how. Don't let the tools dictate. Don't let yourself be bullshitted by plausible sounding "common sense" architectural decisions that the LLMs blurt out -- you'll lose a sense of overview of your own systems in no time.

#ai #llm #architecture

How is Hachyderm feeling about AI-assisted software engineering these days?

My position today is some uncomfortable blend between not wanting my skillset to be something I need to rent back from hyperscalers in order to make some people very rich, not wanting to delegate my thinking to a stochastic process, yet also recognising that there is a clear impetus to use these tools for "fast-and-somewhat-OK" implementation of rote work or well-specced, simple pieces of work. I.e., the work where doing it is not likely to make me a better engineer or to throw up interesting decision points.

That is the practical view that glosses over a lot of ethical concerns that might make me an AI tee-totaller.

What are other people's thoughts?

#ai #tech #ethics #software #programming