softwaredoug

That quote is the lead in to an example I talk about after this quote illustrates what I often see. > Agents bias to making the current change as safely as possible. I had a situation in a previous…

LLMs love to defensively wrap code instead of thinking holistically about the big picture. That creates a lot of bloat. A human coder might OTOH follow the Boy Scout rule and clean up as they go.

Pinky and the gains

It’s probably saner to look at large populations of different types of athletes / exercise patterns in a longitudinal study.

As this is a mice study it’s worth pointing out humans are particularly adapted for endurance exercise compared to other animals. Not gonna pretend to be a HN expert, but it seems relevant when compar…

Imagining adorable mice dumbbells

We already have organizations like Internet Watch Foundation advocating against e2e encryption in an environment where governments have demonstrated a willingness to collude with private companies to …

That’s a bit silly as we also could just write down a lot of exploits, sports bets, or stock market picks on paper and take over the world. I’d be more interested in “how quickly could you develop 200…

This is exactly why I advocate people write code, even in a world where agents code “better” It’s not enough to read code. You don’t internalize it. You need to experience it painfully :) Also agents …

Or defensively expect models to be stupid. Seems the smart thing to do is not assume an agent will do the right thing. But to create the scaffold / harness that enforces constraints to steer them…

Does this model do better ignoring side conversations? That's the biggest hindrance to using ChatGPT's carplay feature is someone will say something, stopping ChatGPT from speaking or taking…

I also think we’ll approach a point where increasing intelligence is not really going to suddenly improve most work tasks. I bet that’s already happened actually. We’re oohing and aahing about models,…

The "stochastic" is the key modifier to "parrot". > "Stochastic" means having a random pattern or variable that can be analyzed statistically but not predicted precise…

I don’t see a problem with the “stochastic parrot” label. It just turns out stochastic parrots are incredibly useful. At a minimum it’s probably more accurate than “AI”.

I’ve switched to a more precise use of agents where I give less autonomy but I have much higher trust the output will be what I expect. And I do more hand coding to guide the agent to useful patterns …

At best the conclusions are a best case scenario for degraded code quality. That things remain functionally OK with more costly token usage to get work done.

I can totally see doing this incrementally, but this seems extremely risky to do for the entire codebase in one shot on anything in production. Especially if you don’t have really thoughtful e2e tests…

Yes this is one reason IMO I think of AI code as instant legacy code. You take on a lot of tech debt. Then you need to do the same work you would do with any legacy app: finding where the brittle poin…

Even if the agent does everything, English is an imprecise description of what the code does. So I personally at a minimum will want to talk “in code” about what code does.

This is why I don’t get the perverse pleasure people get when they say they don’t look at code anymore It’s not like we’ve created a new abstraction layer with coding agents. It’s a leaky factory wher…

This has always caught up to me in a bad way. All it takes is some brittleness to need to take apart the app code and understand it. AI fixes one thing, breaks something completely unrelated. And us…

The study looks at junior developers unfamiliar with what they're implementing. Post-hoc they broke down the AI group and grouped some into tutoring->hand coding. Another group was AI maximali…