realaleris149

Not sure we need any kind of framework any more. I think for web we can get by with bespoke. I’ve made (put agents to make it) a small but not necessarily simple project[1] recently and decided to go raw dom … and it worked quite well.

No special issues with the UI itself. Took some care at the beginning to structure it with reusable components and that’s all.

[1] https://tinker3d.app/

Maybe the idea is not to have the whole model but the executer so … by copying an API key?

I can create the programming part of the game. I have done a couple. Where it stops working for me is at the visual assets. And especially sewing the code to the art.

Also tried it with AI generated art. Is not there yet. I can generate a beautiful image but then cut it to pieces so I ca use it in a game breaks apart.

It can be done but it looks obviously underwhelming.

Do they? It might have been true month ago, not sure right now.

Or some solar panel arrays closer to the sun and it does not matter?

We would have heard if there was any result?

> agents work in public channels, never DMs.

Recently I’ve been looking into why LLMs tend to write in some off putting ways. This one apparently is for LLMs to get better results.

It puts the cases to avoid closer to the main requirement so is easier for a LLM to follow without losing track with more ambiguous language.

Is very interesting because is harder to read by a human but it gets better results for an LLM.

I am using a very similar approach and seeing about the same. It works well and fast with some attention to details.

I also do not understand the obsession with “software factories”. Like “JIRA to product” nonsense. Is like we can hand wave the details that matter and somehow the end product would still be what we want.

Not even talking about the degenerative codebase you mention, which is also an issue I am seeing after some time in any project.

Looks quite useful. Especially like the caffeinate integration with "until agents idle". Is there a way to see more details about the tools activity from claude? Sometime I would like to see what was executed and the output, would be nice to have a detailed view from "Activity by tool" section.

> "PC LOAD LETTER? The fuck does that mean!?" — Michael Bolton, Office Space (1999)

The error message is mocked in the 1999 comedy film Office Space, when employee Michael Bolton becomes frustrated as he doesn't know what it means.Later in the film he and his coworkers are shown destroying the printer with a baseball bat.

> from one of Germany's large insurers

Is this voluntary insurance? It changes meaning to the statistic to % from those that made an insurance.

Take a look at a sufficiently old random internal repo which was not written with LLMs and compare.

My observation is that they are equally bad and hard to maintain or even more so than the new ones.

One thing I’ve noticed is that the LLM assisted ones have a lot more comments which is nice but take more time to read.

The agents will read them

From an agent perspective is very well documented but from a human perspective is not very clear - how you would use it?

> When I say "thanks, that worked," that memory gets promoted. When I say "no, that's wrong," it gets demoted. … > No manual tagging.

I think this is also a kind of tagging.

I always had problems understanding algorithms from such descriptions. Even from pseudocode I find it difficult to understand. What I usually do is search for an implementation, even in a language I am not familiar with is still better. When you have code you can run it, test it, debug it - not so much with descriptions and pseudocode.

> 6 digit years

Totally insufficient for capturing important future events like the dead of the sun.

Not op, but I think "doesn't play nicely" means does not work so you have to do it in other ways. This has been my experience as well, albeit it was couple of years ago.

There are other commands?

They just launched Junie https://www.jetbrains.com/junie/

As LLM benchmarks go, this is not a bad take at all. One interesting point about this approach is that is self balancing, so when more powerful models come up, there is no need to change it.

Still can't get past "Gleam doesn't have loops". I did try to stick with it as the doc suggests, but I guess is a steep transition for me. I like everything else about the language.

https://tour.gleam.run/flow-control/recursion/

In America!

The intro paragraph in the original URL https://openai.com/index/announcing-the-stargate-project/ mentions US/America for 5 times!

The number of people is irrelevant. What is relevant is what each one did. If they did something illegal that is punished with prison time, they go to prison.

Or https://wafflegame.net which looks a bit more polished.

I used https://github.com/jetify-com/typeid for generating api keys for a project. They are nice because you can use prefixes for different type of keys and know immediately what they represent. No dashes also makes them easier to copy paste.