SamInTheShell

Everything other than the isHuman part is absolutely detectable. It's just a matter of building a system for it.

The ability to detect isHuman just seems to be on chopping block as far as I can tell. There is little to no barrier remaining today to prevent people from leveraging AI to get around this entire problem. There's a few outlier solutions that work for now, but AI does two things really well given time: solving classification and regression problems.

What can still be done is maintaining short lived replication logs and pattern analysis in this space. There is no reason that a player's actions can't be logged, analyzed, and responded to. It doesn't even need to be real time, it can just be latent. The server doesn't need to rely on reported timestamps for the analysis part, because the server can use it's own timestamping upon arrival to at least see how far apart the order of operations occur.

I know it's not an easy problem space to be in, I don't envy it, but I can't agree with the sentiment that it's impossible knowing how these systems are already designed in larger engines.

Genuine question, why are people confused on what user input is here?

Suppose the game is a solo game primarily with additional multiplayer features. The agreement there is simply that they enter multiplayer with a valid game state (edit: you validate on join).

Upon entering multiplayer, monitoring begins. Player moves, is it valid? Player shoots gun. Is this valid? Ray cast reached target and does damage. Was that valid?

Did you track the player state during the replication streaming? If no, then there are user input validation gaps.

To be completely fair, most people aren't out writing their own game engine to account for this stuff and it's a lot of work to do that. None of the major game engines on the market do this.

In large database systems, log replication occurs all over the place and validation takes place (some systems better than others for sure). Difficult to implement, sure, but when you know, track, and monitor the state; you can validate and respond.

Moderation is a fine suggestion. This idea could actually be implemented in such a way that other player clients do detection and reporting behind the scenes. Suppose a ton of clients see one user violating rules of the physics engine. Automated reporting on that can be observed as a trend and a ban would make sense.

Having argued about this specific topic in the past, I agree that going so far as to do checks on player movement is difficult and even expensive if you're validating physics for every user.

What concerns me more is the increased risk of RCEs when developers skip fundamental security practices because "the anti-cheat handles it" for them.

> ok grandpa, time to get you a chamomile and time to go to bed

So you resign to insults because you're either incompetent in code or think you know something I don't. You would be better served to make a case for a counter argument, but can't because you lack the skills required.

Edit: Good developers figure out how to make better code, while some developers are just doomed to mediocrity. ¯\_(ツ)_/¯

It's just a fad. It'll die down eventually like everything else does. Don't see much talk about cryptocurrency lately (not that I care to see more, the technology choices are cool though).

Might take a long while for everyone to get on the same page about where these inference engines really work and don't work. People are still testing stuff out, haven't been in the know for long, and some fear the failure of job markets.

There is a lot of FUD to sift through.

I wont play anything that doesn't work through proton or wine. This is how I vote with my dollars these days.

This is a hill I'm willing to die on: Failure to do proper server side checking is carelessness in the face of making a quick buck.

I distinctly remember being a kid bypassing this type of software back when it was just Game Guard at the top. There's nothing that's going to stop kids from doing it today.

It's an arms race doomed to failure until these game devs stop being lazy and just put in the sweat required to get their houses in order.

Server side checks. That's it. User input is tainted evil and you can't trust it. Anything less than server side checking is insecure.

Vibe coding isn’t going to net you horizontal scalability out of the box if you know nothing about scalable system architecture. These models have more examples of single node wonder codebases than they do for real world application of algorithms like Paxos or RAFT. That being said I let Claude Sonnet 4 eat my GHC premium creds a few weeks back to see if it could figure out multiraft in Go. The short answer is just no, without doing an entire refresher on where it went wrong.

It doesn’t really matter though. If you target a DB like Yugabyte and use distributed storage like an s3 bucket or minio, you can get pretty far as long as you keep the core components memory efficient and stateless.

Having come from traditional HTML/css/js back in the day and recently picking up react js in the last 5 months.

I agree with the sentiment.

Looks like it was vibe coded by an amateur too.

Don’t know if their chat demo works, but if it does it could be a great way to just waste their compute.

The way I interact with Claude involves referencing pop culture stuff that I enjoy so that it responds with something that I find mildly interesting to read.

At the end of the prompt:

Startrek stuff - Per Jean Luc Picard's orders, engage. - In the immortal words of a great science officer, the needs of this codebase outweighs the needs of few. - Code hard and prosper - You are borg. We are borg. (followed by a simple "status report" prompt after it's done with it's glazing)

Starwars - May the code be with you. - Code or code not, there is no try.

You probably get the idea.

As for the botched code problem. I surprisingly have less issues with Claude than I do GPT. I've gotta ask, do you have any deep knowledge in coding? How are your instructions?

My first thought was “sounds like communism”. Not that I’m either for or against the idea. Just seems to line up with the literal definition as I know it.

This is one of the more concerning LLM use cases I would say is not a good idea. Most LLMs I’ve tested have the DSM-5 and other psych texts baked in. The problem I see with the idea is that they can’t proof their results. These inference engines can do sentiment analysis, which opens up a really good tooling opportunity for aiding a trained professional. What I see being problem though is that it’s too easy for a lay person to taint the context in ways the LLM can’t recover from. Also an LLM will have a really hard time calling someone out if they are blatantly lying to it.

This is pretty cool dude. I was able to prompt it through VS Code Copilot. Gotta do some work to fix my custom CLI tool to work with it to see what some of the ollama models do.

My bets are hedged on being replaced one day, followed by a few years roughing it, to be eventually be met with something along the lines "Well damn, we really couldn't complete the entire loop on automating the automation" because frankly autocomplete will always be just that. Autocomplete.

Till then, I just learn the tools with the deepest understanding that I can muster and so far the deeper I go, the less impressed with "automated everything" I become, because it isn't really going to be capable of doing anything people are going to find interesting when the creativity well dries up.

Accurate. It's autocomplete on steroids.

You should make a bunch of functional demos that were churned out with the tool.

I’m pretty skeptical of the one shot claims for anything intensive. Considering much of the things I glossed over were written by AI too, I wouldn’t put it past GPT to glaze things thick, while having no clue what it’s talking about.

I think you’ll probably have a hard time with this because your idea is literally stuff everyone else is working towards for a few years now. Even myself, I’ve thought about wiring whisper v3 together with some other LLMs to vibe code conversationally with agents (it just doesn’t drive my main project goals).

Frankly I wouldn’t put competing code on GitHub if it was a proprietary tool and if I did, the reality is there are surely tons of other people doing similar that they can say they drew from if they lack proof to just being on the product they released for years.

I just stopped by here to point out your domain is still parked, like you just bought it.

You sir, have delivered upon me an entire "ha ha" that I could not hold in. Thanks.

I'm just skipping over a lot of this and only going to address the sentience thing.

AI sentience will never be a thing. The transformer model never changed the fact that models like GPT are still just solving a classification problem, no different than any other classification model before it. It doesn't take a PhD student to understand it's just clever math.

I like how this is written in a way that an LLM doing planning can probably infer what to do. Let me know if I hit the nail on the head with what you’re thinking @antirez

Probably worth knowing, there’s already a popular clipboard manager in the wild for MacOS called Clipy.

https://github.com/Clipy/Clipy

I think a lot of what you’re pointing this thread towards boils down to philosophical beliefs. Objectively throughout history there have been people resistive to technological advancement and those people have more often than not been the idea losers in history.

I’ll throw you something I believe that we might agree on though. I don’t think the colossus data center Musk setup in Tennessee is good for anyone. Those generators he’s been running are abhorrent and the guy needs realignment of his neurons through some percussive maintenance, but alas that’s probably illegal because he’s too much of a chump to accept a boxing match.

I’m not special and anyone can do what I’m doing. Civilization has been advancing technology since people stood upright. To stand still and not expect change is just ignorance. I can’t fix flawed people, I can only march forward.

Market displacement is nothing new. Happened in the 2000's (dotcom bubble), happened in the 2010's (cloud infra), happened again in 2020 (services workers being funneled into tech), and it's happening again now (AI is replacing those that fail to adapt with the market changes in tech).

The one thing that has kept me viable as an employee over my 15 years in tech is that I literally don't want to do the same thing I did yesterday 1000 times. I want to do it as few times as possible before I automate the problem away, so I can move on to something new. There will always be something new. There will always be someone with a dream and no skills; for me to step in and help out.

I fail to see the problem.

Nah, anyone who thinks they aren't going to need a compute expert to deal with their tech stack is huffing copium.

> Are those internal documents in the room with us right now?

I have no clue what you're on about here.

If you have a legacy knowledge base, like maybe using mediawiki for corp knowledge, what you do is maintain a vector database that gets updated when it sees changes. Using embeddings enables lookup through sentiment.

In a control loop with well maintained vector embeddings, these LLMs are absolutely better than a human at finding, citing, and summarizing information needed by the user.

Tools like glean already exist for this if you doubt it.

These LLMs are really good at digging up internal docs if you give them access to your knowledge sources with tooling to search and reason in a loop before responding.

If you're in Google Cloud, the risk of using Gemini seems not much different than a using their cloud storage imo. Then you also have redundancy if you have OpenAI, Llama, and Claude as drop-in replacements.

The only thing you need to maintain is context.