PS5 emulation has gone from barely working to running Dark Souls at 10+ FPS with virtually no graphical glitches… in 6 weeks. If you’re not familiar with normal emulator development time, this is… quite extraordinary. There have been insane progress on decompilations and many other things in the emulator space.
Everyone’s trying to figure out how to convert this speed to product features at scale, but enterprises are like container ships. Lots of might but slow to turn. The littler companies can actually take advantage of this and produce higher quality products at much faster speed. I think you’re expecting too much in the short term and too little in the long term. AI-native companies are gonna eat everyone’s lunch, once they figure out how to actually do it reliably.
Also, in 12 months, we went from seeing game dissassembly and decompilations projects be slow and annoying, to suddenly having enough mapower where people are taking the liberty to CHOOSE which decompilations to support. DK64's release was 80% anti-AI marketing, not because they were strictly anti-AI, but because they were telling the OTHER decompilation to fuck off and learn some goddamned standards. Completely unthinkable before.
There was a video by a layperson 2 years ago where they wanted to know the mechanics of some elusive pokemon pinball spawns where there was a whole bunch of missinformation about it online. The people behind that project had the correct addresses but they were completely unlabeled and the layman had to essentially figure out and do some of the work themselves to sort it out and figure it out. Nowadays that would never happen. An LLM will just do it for you within the day.
The fact that this is an account with a username specifically focused on this specific concern that was created 5 hours ago is real weird/suspicious.
Also, are these even actual decompiles? Sure, playing Wheel of Fortune with a LLM model that at some undetermined point spits out what seems to be a working example is sorta useful, but also…how exactly do you debug this nonsense? Don’t tell me we’ll just have the LLM thrash around a bunch again. What I’m asking is does anyone understand what is going on here other than “lol, it works, shut up”?
You must be new here. Lurkers posting under an apt username then disappearing is a feature, not a bug. He doesn't need to argue with anyone online - he said his piece and we're free to take it or leave it.
Usually, green username are a sign of "leave it", I'll agree. But this one seems very informative.
I mean, do you understand the systems you work on? Down to the ISA? The transistors? The electrons?
The stack gets taller and more complex, as it ever as. People who understand one part of the stack in great depth rarely understand the entire stack. Those who understand its full shape don’t have complete depth at any one part. This is how computing has worked for a few decades at least.
You don’t have to shut up, you’re free to dive in and learn. But you’ll have to balance breadth and depth like the rest of us.
I think most people with even a bachelor's in anything STEM know it down to the transistors. That's also been true for decades. Every time I hear this line of argument I cringe. Computers aren't that deep compared to other engineering topics.
LLMs work great at decompilation, I'm not sure what you are trying to suggest. Outside of game compilation disassembly is useful for debugging various crashes in 3p code where you don't have the source. And yes LLMs do the debugging here.
I’m suggesting that having a model wander around stochastically is not necessarily a great way to gain any insight or understanding. LLMs do not think, they cannot really debug anything. Attaching intelligence or intent to them is a non sequitur, it’s not happening. A cloud that looks like a toaster does not imply anything beyond the human obsession with pattern matching, why do we keep doing this with LLMs? They can randomly wander around and sometimes produce useful output.
I don’t doubt they help by randomly permuting around in a way that people cannot do at scale, but I question the economics of this approach and I think it needs to be used judiciously. Sometimes you just have to do that especially like you said, when you do not have the source or the code is intentionally obfuscated, this can be useful, but a lot of people seem like they’re just vibe coding crap and blowing money on tokens until they get a useful result and I have no idea what the value of that is.
I did the same thing with a old multiplayer game called Wulfram, we got it online and a mostly working server (most of the game logic was server side with no surviving code or binary) with the help of AI, Mostly opus and astra. https://wulfram3.com is our efforts in this.
> AI-native companies are gonna eat everyone’s lunch, once they figure out how to actually do it reliably.
Couldn't you say this about any company? What isn't just a matter of "figuring it out"?
Why can't they use AI to crack this nut? :D
> If you’re not familiar with normal emulator development time...
Emulators have historically been a shitshow until someone figures out missing pieces here and there. By their nature, that tide lifts all ships. It has very little to do with using AI. I'd argue the real reason emulators aren't what they used to be is the hardware is more complex now and the scene is no longer attracting the most talented devs. Video games used to be on the cutting edge and carried a lot more cultural weight. It's pretty underwhelming to get cred for working on lame x86 hardware that plays yet another franchise reboot.
If you don’t have style rules baked into the PR process, you should. Tabs vs. spaces is a team decision and needs to be matched by everyone to reduce code churn. It’s correct to reject a PR if it violates the team style guide (although it should be automated and have tooling locally to enforce it as well.)
this difference engine model has a lot of obvious flaws - the gears do not mesh with each other, the teeth directly overlap. the notches that are supposed to articulate to ratchet aren't actually engaging the teeth on the ratchet. the ratcheting mechanisms on the right are straight up just hovering in the air and are not attached to anything.
that's probably what people are talking about when they say it can't do CAD?
Complex assembly mating is hard to do even in SolidWorks. It's a demonstration model to explain how it works, not a kinematic simulation, turning three.js into a physics engine is beyond the scope of this project.
How do you actually do that? Is it all running locally? Cloud agents? Would love to hear about this. I see these deep agent loops mostly just burning tokens, but when I guide the AI I get very good results, so I’m not sure where the disconnect lies.
I use Zed (https://zed.dev) as my agent harness and either the $20 ChatGPT sub + Sol for personal/independent projects or an enterprise Claude account for sponsored/paid work. From the stats for my current work I use about $400/mo in tokens and a lot of that is non-coding work like pruning JIRA, managing business documentation, making dashboards - so my true coding agent cost is significantly less.
It's pretty simple, you could probably set up something like that by:
Configure some kind of CLI tool to talk to your ticketing system and git repo so you can programmatically interact with them. If you don't have a ticketing system, instruct the agent to use local text or markdown files to track issues and progress.
Ideally, make your code runnable in a way the agent can use. For my webapps I build a test harness so that I can run all the endpoints and workflows via reproducible tests against an embedded database. This is easier than it sounds, e.g. there are libraries out there to embed PostgreSQL or SQLite into source code, you can set up a test harness so you can run unit tests, integration tests and workflow tests that use your real frontend, server and database.
Paste this comment thread into the agent prompt and tell it to run a similar loop on your code base: a session that searches for vulns and writes up a report, some way for a human to do a review pass on the report, a session that indexes the reviewed findings into tickets, and sessions that fix the fixable issues and submit patches to your repo. The next search session should first read all the open issues so it doesn't duplicate work of earlier sessions.
LESS IS MORE - avoid fancy agent tooling and skills, don't cargo cult from others, build your own tools as you find your own needs. If something can be automated, use the agent to write tools and tests for it, don't just keep prodding the agent to do it.
I would love to use Voz and Ear, but I’d need a version that is competitive with other audio transcription LLMs for platform availability - meaning macOS, Windows and Linux, and supporting GPUs if available.
I love the idea! OKF 0.2 solves for the problem of the AIs generating massive amounts of documentation (far more than humans ever created) and giving it equal-weight over what a human actually approved and committed to. I've been using it to attribute my decisions with specific directions on how to ensure that it takes strong direction from my explicit decisions and clarifies implicit / AI-driven decisioning.
Adding on progressive disclosure to this is brilliant, and I love the idea of a fast, in-memory, single-binary tool. This is a great way to approach the solution to this problem.
One thing I will say though - I would never be able to use this in my enterprise. It would just be too much of an uphill battle to purchase something that is so niche in utility - this tool is not a ton different than just having the md files locally and having it use ripgrep to search over them, and telling CLAUDE to write the OKF files as well as an index when it makes changes, is it? is the index generated dynamically / is anything about the progressive disclosure different than just having the agent manage it while it documents?
If you're going for smaller teams that can buy tools without a ton of approval / procedural overhead, I think that might have some success. another possible solution would be to make the cross-repo search something that you can handle with OSS but you have to self-host, and then pay for support. if you got enough usage and penetration within an enterprise from the teams just using OSS and self-hosting, they might consider buying support after-the-fact.
Running an iOS app on Android is still gonna look like an iOS app. This seems worse than Flutter (reimplement the whole UI) or React Native (abstract on top of native UI.) If we’re ok picking a non-native UI why not just build a Cordova app?
Why don’t you have your actions run on pr or push instead of on commit? Why would you even want that? I’ve never seen actions set up that way. If I push a branch with 100 commits then it’ll only run CI once. It’ll show the rest of the commits in the UI, sure, but that doesn’t mean that it’s the same performance impact. It could very well be 1 db transaction with multiple rows written instead of 100. I think you’re reducing this problem too much without knowing their architecture.
Github runs CI per PR push (obviously), and per main-branch commit (click on commit history in any project with CI, see a build result check mark on each one - that's true if you push a dozen commits too (I've done that)), in nearly all setups I've seen. I'm not sure how much of that is required vs default though.
With enough effort, you can rather obviously run CI per PR commit (it's a programmable system), but I've never seen aUI-integrated way to track the results, aside from browsing custom job names, which is very far from what I'd call "integrated" when compared to PR-level build markers. Similarly, I'm not aware of (but would not be surprised by) any way to disable per-main-branch commit builds, aside from initial pushes.
But I haven't poked around deeply in the settings, and business-account settings are rather different anyway so those might be wildly different / more flexible / more obtuse in exciting ways. Github is a very large and complicated product at this point, darn near anything could exist if you dive through enough UI layers or use old URLs to find soft-deprecated features.
Also, honestly, 100 commits = 1 transaction? That's far more of an over-simplification than anything I've said. It's a massive product with thousands of engineers, there's no chance at all it's just one database.
Everyone’s trying to figure out how to convert this speed to product features at scale, but enterprises are like container ships. Lots of might but slow to turn. The littler companies can actually take advantage of this and produce higher quality products at much faster speed. I think you’re expecting too much in the short term and too little in the long term. AI-native companies are gonna eat everyone’s lunch, once they figure out how to actually do it reliably.
reply