psadri

I agree. I recently started exposing my kids to programming and I chose a C64 emulator. The BASIC REPL is so simple/limited that it doesn’t overwhelm the kids with irrelevant syntax (those can come later, if they are interested). The fact that a 1..100 loop can introduce a noticeable delay. You can literally see the computer working. Primitive graphics and sound provide immediate feedback which makes learning engaging and fun.

I often spot the mosquito but lose track of it after a second (against a patterned background) or if it rapdily changes direction (which they seemed to have evolved to do).

What I really want is an iOS app that helps me locate insects in a room. Start video feed, analyze pixels for a small object flying around and draw a green box around each while flying and a red box once it lands somewhere. I can take care of the rest.

I don't disagree that code is far more efficient than inference. But what are those apps? A lot of apps are fetching some data, massaging it, rendering UI to let the user view / update data. Could the AI do some of that "work" for you, so that you don't need those dashboards / buttons / forms in the first place? Maybe you have an agent that does a db query (instead of a human viewing a dashboard) and takes some action (instead of the human clicking a button).

I love the idea of Phoenix and server side rendering (I happen to work on SkyMass, a related project).

This is a tangential comment and should not detract from what Chris and team have created. I think closing the loop between agent and the running output is a great/critical step forward.

However, I find using AI to build transitional Apps with a UI is a bit like improving the way automobile steering wheels are made. In a world that soon won't need steering wheels at all.

If the AI is so good to write the code for an App, how much longer before you won't need those Apps in the first place? And then the question is, what will fill the role that Apps play today.

Back at Polyvore we used real post-it notes on a board to track progress. We had a weekly cadence but the tasks were granular enough to be completed in a day or two. Each week we’d peel off and crumble the completed tasks into a bin. It was a really satisfying experience in a way that digital facsimiles can’t quite match.

Don’t forgot to realign the dilithium crystals once in a while.

Horizontal scaling. I love this out of the box thinking.

They should add a “cmd-enter” for ask, and “enter” to go.

Separately, if I were at cursor (or any other company for that matter), I’d have the AI scouring HN comments for “I wish x did y” suggestions.

I usually add “discuss first. Don’t modify code yet”. Then we do some back and forth. And finally, “apply”.

What do monkeys think about humans?

The root cause is that web browsers were designed for document delivery but are used for building application UIs. The browsers don’t offer a standard set of common application UI components, so every team builds their own leading to inconsistency and half baked implementations.

In contrast, when you build a native app, developers can draw on a standard set of OS provided UI widgets.

Dynamic languages like python should allow you to monkey patch calls so that instead of opening a regular socket, you are interacting with a wrapper that reopens the connection if it is lost. Could something like this work?

I think it depends on who is using the system. For a power user who is already familiar with the domain, a free form, open ended ui that can understand anything I very powerful and liberating.

For a novice user or someone who is not from the domain - it can be challenging because they may not know where to start.

There is so much that can be done in this space by fully leaning into the AI. It can for example figure out the user’s level and offer varying levels of guidance and help.

I’m a fan of this co-authoring-a-document model. It’s nice bridge between free form interaction and rigid UIs of the yore.

This has already been happening. The gap between wealthy and poor is increasing and the middle class is squeezed. Interestingly, simultaneously, the level of the poor has been rising from extreme poverty to something better so we can claim that the world is relatively better off even though it is also getting more unequal.

One more tool he could make is one to wrap that entire process so there is less copy/pasting needed.

I think an interesting avenue to explore is creating abstractions and analogies. If a model can take a novel situation and create an analogy to one that it is familiar with, it would expand its “reasoning” capabilities beyond its training data.

I think that is how human brains work. When we practice, at first we have to be deliberate (thinking slow). Then we “learn” from our own experience and it becomes muscle memory (thinking fast). Of course, it increases the odds we are wrong.

Yes, but we have the ability to reason logically and step by step when we have to. LLMs can’t do that yet. They can approximate it but it is not the same.

Are there any constraints around which functions can be turned into steps? I assume their state (arguments?) need to be serializable?

Also, what happens with versioning? What if I want to deploy new code?

Where is the state stored? In my own pg instance? Or is it stored somewhere in the cloud? Also, a small sample code snippet would be helpful.

One of my fondest memories is from when I was in university and one Xmas break my stipend was late for a few weeks and I completely ran out of money. All my dorm-mates were away for the break so I had no one to borrow money from. I scrounged around my dorm and found a fistful of loose change — it came to around $2.30 iirc. Bought a bag of rice and frozen peas and managed to last about a week. “Free” takeout soy sauce packets were a big plus. Good times.

If you are interested in Retool but with public facing apps, checkout SkyMass. It’s a code-first alternative to Retool that gives you the high level UI components, auth etc + flexibility of writing normal code.

https://skymass.dev

It doesn't need to be all or nothing. The traffic flow algo can ignore regions with known zero cell coverage. Perhaps we can fill the gaps with inertial navigation if there is at least some traffic.

Most of the schemas I have worked with used “id”. I agree it’s the default. But aliasing was inconsistent and made it hard to figure out which table id is being referred to later in the code.

I know it is a matter of discipline but any tool that encourages consistency helps. A database schema is definitely one of those tools.

One unrelated idea is to include the entity in the id itself. I have never done this but I’d imagine it would help with things like logging / observability. It would not play nice with indices though.

I have found that naming ids as <thing>_id helps downstream code when trying to figure out which thing's id you are dealing with. It also helps with avoiding renaming fields when a structure contains multiple ids.

I do agree it makes joins more verbose.

This happened to us in Minorca. It was a pain to backup 50 meters with only 20cm of clearance.

I’m surprised they can’t flag these misclassified roads by tracking traffic flow through them. A bad road should exhibit little traffic (or even more reroutes).

Let's say you have a running program on a computer, and you figure out a way to swap out parts of its instructions / state in RAM while it is running. What are the odds of your swap causing problems? Now, what if the program is 100 - 1000x more complex than anything you have ever managed to create.