Mooncast Productions

@KatDoes3D Finding people who ask the same question as you, then challenge them to take their favorite game, ask what specific mechanic makes it fun, then reimagine the same game with that mechanic changed in a fundamental way. Discuss. If you feel convicted, prototype it. Learn by doing.

Hello, actual game designers. Wannabe game designer, here. How do I pursue an actual understanding of game design? How do I learn more for real? The last thing on EARTH I want to do is form gamer opinions, and I'm super interested in design. Where do I start? 🩷

@thomas_fox28467 @ExtinguishedEng @BenjaminPDixon Studies show that humans hooked up to monitors continue producing language even during epileptic seizures and while under anesthesia, too. People seem to think the brain is one thing. It's many regions, working independently with interconnectivity.

Actually, the brain does something very similar. Brodmann areas 44 & 45 (Broca’s area) in the left inferior frontal gyrus handle the transformation of abstract intentions/thoughts into structured speech output — sequencing grammar, phonology, and articulation. Area 47 often joins in for semantic aspects. Transformers aren’t copying the brain exactly, but this is a clear biological parallel for turning internal representations into fluent language.

I'm not sure what to say to people who don't realize our brains are literally producing the wetware equivalent of the next-predicted token with every word we say and compose.

@ArrayEmpty @UnslothAI @no_stp_on_snek @_HermesAgent @MajorFAFO @ASUS fwiw, I am running that config in 6gb with a bunch spilled out to cpu so I can also run Qwen3.5 4B entirely on gpu in the remaining space. I have 8gb.

@MooncastOnline @UnslothAI @no_stp_on_snek @_HermesAgent @MajorFAFO @ASUS I've found any --n-cpu-moe adjustments < 40 crashes llama, so full cpu offload seems to be working. I think i will try that --reasoning on --chat-template-kwargs "{\"preserve_thinking\":true}" though....

I'm finally happy with my @UnslothAI unsloth/Qwen3.6-35B-A3B-MTP-GGUF, @no_stp_on_snek llama.cpp-turboquant, @_HermesAgent setup, @MajorFAFO params suggestions, and my own Rust/SQL/Hermes app. - @ASUS, 32GB RAM, 8GB VRAM 5070 laptop - MoE fully offloaded to CPU - I've fiddled around with --no-mmap and mlock. It pegs my RAM to 95% instead of 45% and after it spins up, I don't see a difference. --cache-type-k q8_0 \ --cache-type-v turbo3 \ -ot "\\.ffn_.*_exps\\..*=CPU" \ --spec-type draft-mtp \ --spec-draft-n-max 3 \ The video shows it has full hermes skills/toolset access easily, knows my system, can be creative, can print out a quick BASIC function, corrected itself when I told it that it was incorrect, and that it doesn't love me....

@ArrayEmpty @UnslothAI @no_stp_on_snek @_HermesAgent @MajorFAFO @ASUS You probably should try the proper llama.cpp distro. No turboquant yet, but stability and performance. I was using beellama which has great turboquant and mtp, but for whatever reason it is much slower than the main repo. AtomicLlama had massive memory leaks.

@kittykareninas You don't disprove them. You fire your teacher.

i hate ai detection programs so much. my fully HUMAN WRITTEN essay shows up as 87% ai written. what are you even supposed to do if your teacher brings it up? how can you disprove them?

@FlippyMeister Even if you are high level and have a good reputation, this is a brutal time. There's apparently a very large percentage of over 50s that never receive a job offer after a layoff. I'm not looking, but many friends of mine are.

I find it terrifying how a lot of the "older" people I've worked with in tech have really struggled to bounce back after a layoff. Extremely competent, tons of experience in their field, mentored me. I'd hire them in a heartbeat. But tech seems to have this invisible ceiling where after your 40s or 50s, it's a lot more difficult to get hired. Especially if you weren't really high-level at a very well-known company. Makes me feel like there's a clock ticking for me to "make it"

@loktar00 Pretty cool test. Nice.

Qwen 3.7 Max is the best and fastest by far, 0 Shot 1 try Deepseek I'd say is 2nd, has some better obstacle avoidance. Impressive results all around but Qwen 3.7 Max has proper turret orientation and proper track marks.

Qwen 3.6 35B-A3B on my potato RTX2080. Adjusted MoE to get same performance/vram comparison. Beellama K/V=turbo4/turbo3 + ngram speculative decoder (best in my tests), latest llama.cpp K/V=Q8/Q4 + MTP. MXFP4 has been considerably faster even using 3GB larger models. +41% perf!

@vboykis I find (very much unlike when I complete a project manually) I often complete a project agentically but I do not commit many details about it to memory, and thus do not learn much it. Making so much stuff that I know I have done something similar...somewhere. Too much, too fast.

Something happens to my brain after agentic coding that I can’t describe. It’s like cognitive offloading which folks have already written about, but even more. It feels like I can’t think through problems anymore. Like a fog. Using agentic but losing my hard-won agency.

@simplex_fx You asked for an example, I gave you one you would recognize. Calm down.

@MooncastOnline So you have a problem with Unity DOTS. I'm talking about DOD in general. Not that Unity shit.

Is it just me, or the whole DOD thing died around 2022? I mean, there's still some of us trying to build highly performant sw, but no one fucking cares (even in gamedev, not to mention outside of it).

@simplex_fx Reflowing data when you are the sole programmer interacting w/a system is much easier and cuts down cache misses, but the code (if written explicitly) starts off pretty fragile, like any other optimized system. It takes effort to generalize.

@simplex_fx Unity. DOTS. There are millions of loc out there that will never use DOTS, which bifurcates all existing codebases. It is literally not easier to refactor.

@ArkhamDreamer That is how doors were done in the late 90's. Valve did it in Half-Life (iirc) and we did a variant of it in Wing Commander: Prophecy when ships were going through jump gates. It was even on the box cover. Planar masking/clipping was a hot new feature then.

Just found an old 1998 E3 interview showing an early, never released build of Prey, and my mind is completely blown. They were showcasing seamless, real-time portal tech a full NINE YEARS before Portal came out. Looks like straight-up witchcraft.

Ever think... Seems like a lot was predictable from the 2000's until 2019. Just slow advances, small improvements. Then COVID hit and nothing ever was "normal" again. It's like our timeline had a Spiderman-multiverse psychotic break and we ended up on a losing one.

@haider1 Say the man responsible for the most complicated programming language ever devised. Kind of the Alfred Nobel of our time.

Creator of C++, Bjarne Stroustrup: AI-generated code isn't ready — it generates more bugs, more bloat, more security holes, and is nearly impossible to validate "senior developers are already retiring rather than deal with it" The problem is that even a small prompt change can shift the entire codebase in unpredictable ways

@RealtimeVFXMike I won't spoil the fun. But in Update, high fps jumps will never feel the same as someone with low fps, and they possibly may be unable to leave the ground.

@MooncastOnline I hadn't considered that kind of jumping. I just have it setting the velocity now. Probably would be something like checking the vertical velocity against the grounds vertical velocity.

Character controller update. Trying a new technique for stairs and more complex geometry. The blue box represents the estimated slope. Also interaction with rigid bodies and weight impulse on rigid bodies. Want to try a boat or a rope bridge next. #GameDev #IndieDev #Unity3d

@exQUIZitely Chris Roberts had aspirations to become a movie director, which he did for a number of years. WC3 was also released at a time when cdroms were new and video hadn't really been done before. The novelty wore off quickly.

@MooncastOnline Might be an unpopular opinion, but I didn't like the FMV parts. I prefer the pixel graphic style.

"Looks like it got a little hot out there, sir!" Wing Commander (1990) is often considered one of the earliest "AAA" games, even though the term itself didn’t exist yet. Chris Roberts’ game reportedly cost about five times more to develop than a typical hit title of its era, making it by far the most expensive game at the time, and raising eyebrows everywhere. It featured very advanced graphics, digitized speech, a cinematic story campaign with branching paths, countless cutscenes, and a “Star Wars in space” feel that made it stand out massively. At times, it genuinely felt like you were starring in an interactive movie rather than "just" playing a game. Its success raised the bar for the entire industry. Publishers were now forced to invest far more in production quality, bigger budgets, and marketing. Wing Commander sold millions of copies, and helped establish the space sim genre. The game not only paved the way for even larger budgets in its own sequels, but it also pressured other developers and publishers to keep up. Whether this was ultimately good for the industry could still debated today. On one hand, it started an era of cinematic, high-production games. On the other, it made it significantly harder for small teams and low-budget titles to compete, effectively throttling the kind of creativity that had been a hallmark of 1980s game development. With the birth of Wing Commander, the industry had entered a whole new territory.

@OfficialLoganK Please tell me you're planning to deploy Bayesian Neural Networks.

Why don’t LLM’s just tell you when you are asking a question / doing something that is out of distribution?