Hmm. The project uses BirdNET-Go, which does inference locally on a Raspberry Pi. Then it looks up the species against a set of already existing illustrations. So in this case it seems like the "ultra expensive, data-center-driven AI" is not really involved in the core loop of this project.
This is incorrect; the local model is a small dense neural network called BirdNET that would not have been referred to "AI" when it was published in 2021. The model outputs a bird species probability distribution and the web application displays an existing image file of the most probable bird on the screen. This is a lovely example of simple, offline, and fun project using a straightforward machine learning model.
Ok, so why didn't this small dense model come up, say, 10 years back? Oh it needed all the AI evolution and concepts that were powered and evolved inside the data centers. But we still want to claim that it has nothing to do with the Big AI.
Incorrect. As others have pointed out, BirdNet is a “traditional” neural network. The amount of compute needed to train something like this is many orders of magnitude less than an LLM. Something like this could be trained on local hardware with enough juice, or by renting a handful of GPU’s
What a shame that you use that excuse to offset all other evils of AI, without even having a hint of how many patients were saved by AI. Just like how oil salesmen and nukes makers say that they solve some great problem of the world.
Attacks have been distributed for quite some time if your service has any loot worth attacking. You have to handle the case where every request comes from a unique IP address.
You can /i-have-adhd when you get a wall of text and are exasperated. You don't always want the responses that it generates when you /i-have-adhd, and things in AGENTS.md tend to lose weight as context grows.
How do you differentiate between mastery of a topic and mastery of using an AI in an exam setting?
Open book tests and allowed cheat sheets already solve the memorization-is-not-required gap. It seems like allowing use of AI pushes that closer to thinking-is-not-required. And if no thinking is required, then how can you prove personal mastery?
If no thinking is required, why do you need thinking in the first place. Seems useless. If an ai can do it, why you need a human to be able to do, human should be doing something the ai can't.
First, cloning a repo doesn't bring the hooks with it - so it's not just as simple as saying "use git hooks".
Second, if the repo had hooks and instructions for the LLM or user to blindly install/enable the hooks, we'd instead be complaining about security risks and what might happen if the repo is compromised at some point in the future.
Third, sometimes you don't want to mechanically enforce things via git hooks because it impacts your use when what you're really trying to codify and enforce are the LLM's actions. In that case you can enforce mechanically via hooks at the harness level.
And finally, git hooks are a great solution for upstream repositories to enforce quality and protect branches. But it means that the upstream is the one running the checks. It makes the upstream a potential bottleneck - better to have the leaf nodes run the checks locally and fix any issues before pushing it upstream rather than push upstream, wait for results, make changes, push upstream, wait for results, make changes.
First, I'm SURE there are ways to send scripts from one computer to another over the internet.
Second, the point isn't about a specific repo, it's the general tendency to rely on fuzzy .md files scattered all over the place. And I really don't see how letting the output of a language model run a one time command is more secure than running a script.
Third, "nothing applies in all context"? Yeah, obviously. And harness hooks (at least with Claude code) are still more suggestions than anything else. The only way I've found is literally rejecting a tool use and forcing it to recall in the proper way, which of course makes for more token usage. I wonder who benefits from that.
Finally, no idea what you are arguing against. Use git hooks where they make sense, local or remote.
This is a silly analogy but it's like integer truncation has been applied everywhere blindly. Staffing model says you need 2.5 employees on average to handle the load? You get 2. Everywhere with everything.
And modern technology / weak labor market has made this worse. I worked retail in high school in 90s and we basically had a weekly somewhat regular schedule. It was very rare to get sent home early.
Now you hear about people being "on call" for basically minimum wage jobs, called in for half shifts, sent home early, etc. Stringing together 40 hours at a job like this is basically a job in itself.
Perhaps by some measure, but what I'm referring to is how everything is becoming a fly by night scam-adjacent "product" that isn't much better than Alibaba drop shipping.
What are the "trusted sources" that are better than Amazon (which is itself no better than Aliexpress)? Or Walmart, or target, or Newegg, which, ditto?
I buy straight from brands I trust whenever possible, not even trusting Amazon as an intermediary between us. Those brands aren't selling alibaba-dropshipper stuff.
I'd say my trust in stores that sell more than their own brand is at an all-time low. You have enshittifying mid-range department stores where you can't trust a thing, fake "outlet" stores, et c. I sort-of trust REI. I sort-of trust Costco. End of list, maybe?
Like if there's value being delivered in the "trusted source" sphere for a general retailer, I'm not seeing much of it. Quite the opposite, that's all gotten worse steadily this century.
reply