Dad, Husband, CTO @ MSQ Intelligence. Coding since I was 8. Ultra-full-stack builder + designer (since pre-AI), AI wiz Coming fall: playhighroller.com
🇬🇧 United Kingdom
- Tweets 2,838
- Following 545
- Followers 1,093
- Likes 6,922
Pinned Tweet
Did you think this is possible in the browser?
A full FPS "horror" tech demo of my idTech 8 style GI implementation in @threejs
Sound on! 🔊 I spent all evening on the sound design, I hope you like it!
👇 I added further details in a thread for the curious bunch
53
23
369
51,495
LLM routers in almost all useful cases are to help non-tech people (which is the vast majority of LLM usage) get better results
If you ask, most of them have no clue about models or effort
Evaluating a router on DeepSWE is like getting Lance Armstrong to review a kids bike
I stayed up til 2am and spent $1,000 benchmarking Jev Router so you don't have to.
Performance on DeepSWE was roughly the same as GPT-6 Astra on low. It costs slightly more, and it took almost 5x longer to run.
1
177
Oh no, Dunning-Kruger has found an issue with how my character holds the sword! M
Because, AI is bad, apparently. I’ve asked him how to fix it.
I’ll contact the modeller I bought it from on @CG_Trader (who hand modelled this) and forward this numpty’s feedback 🤣🤣
2
10
685
Here is the same souls controller from scratch with Opus 5.5
It took 6 turns to get it to this point. The old one was a mix of Astra and Fable and at least 10x the turns + crap code
Hindsight is 20:20, but 5.5 still has a much better sense of how to do @threejs "the right way"
Hahahaha look at his goofy lock on diagonal walk
Straight out of Looney Tunes 🤣
Third person controllers are TOUGH to get right... Thank god @threejs has such a clean & lightweight animation API surface
10
31
2,374
Yep, Opus 5.5 is really good at following existing coding patterns
Bit useless considering I'd rather use .ts but @threejs source regularly puts scratch variables at the top of files just like this
In this case I would've also extended EventDispatcher to tie other systems in
There are still inefficiencies in a bunch of places - that I'm sure would compound over time - but it's like 9/10 on "Do this in a highly three-core compatible way including styling".
Other models pretend and end up loosely following while going on their own adventure. This is super important for production projects where code review happens on everything!!
Well except when it does those bloody 80 hyphen comments...
102
Torturing myself with a bit of OSRS doom while the clanker cobbles together the results of my 29 page prompt
Hopefully it doesn't start spinning up 12 instances and kill my GPU mid orb phase...
2
5
229
If this is temporal, this could very well be used for game QA.
Maintain same resolution; compare player movement cycle, observed rig size & motion to a game reference.
Check that going left actually goes left.
That sorta thing.
LLMs exceed with highly capable tools!!
I’m back, with a little something…
GPC 1, a new general-purpose classifier enabling: bounding boxes, poses, coordinates, direct mechanical control and so much more, all in milliseconds.
Post-trained on millions of datapoints and thousands of unique problem classes learned directly from you guys posting on X, GPC-1 is a great tool for many things. Early user feedback is also good :)
A new output class enables precise continuous range numerical outputs, so the model can output coordinates of bounding boxes, exact angles and distances, pose estimations, scores, and more. It goes a lot beyond typical classification which just pick a label.
Excited to see where this goes.
Available now!
1
1
360
I don't know if it's just me, but I culture an anticipative ambition for future models. Like, I know what we have today will be blown out the water in 6 months time. And even more so in 12 months
Opus 5.5 completely front run that. It writes code & comments how I ask it to. It interprets prompt intent like no other.
And I feel the benchmarks are lying - in the sense that they're underplaying how good it is.
- Astra wrote the worst code with no whitespaces. Refused to follow patterns. But forward-intelligent in some areas
- Fable 5.1 wrote ok code but did not wanna follow my patterns. It wasn't as smart as Astra in some areas - it was ahead in others (UI QA in particular)
Opus 5.5 feels like Fable 6 to me, and supersedes the two big ones. It wrote clean code that got the exact feel of a codebase I've been working on
I'm gonna see if it's the same case with Three.JS stuff tonight.
In particular, there's a three-way of writing things: when to extend Object3D, EventDispatcher, how constructors and scratch variables are used. And how that bubbles up to GPU upload
I wanna see if it performs here too - if it does - it'll be hard not to dig out the game engine project again!
2
8
301
Dan has set the standard.
Try it... It's unbelievable, and as a lifetime gamer - beats every single other Opus 5.5 demo in both playability AND visuals, by a long shot. Beautiful!
FYI there are demos from others that are spectacular also. This is an incredibly exciting time!!!
Ever wonder what $1874.40 of Opus 5.5 tokens looks like?
Wonder no more.
I highly recommend watching the whole video on 1080p on a large screen with sound on.
Pay attention to all the little details.
- Birds flying and diving into the ocean.
- Cloth, signs and lights swaying in the wind.
- Crabs scuttling along the shore and burrowing when you get close.
- Fish swimming alongside the whale.
- Lights illuminating the dock at night
Then zoom out and see the entire island. Never dipping below 60 FPS at 1440p resolution.
Yes, there's a few bugs and visual artifacts, some textures need improving, the shorelines waves sometimes look funny, but these are so trivial to fix at this point.
FYI I'm on the $200 subscription plan. This used 59% of my weekly usage and took 2h 7h of API time using multiple subagents, about 8h in real-time.
There was extremely little technical direction here. 99% of my prompts were "Add X and Y" or "This looks weird, make it better".
It's a great time for hobbyists, bad time for professionals. This experiment has further cemented my view that technical creatives are about to experience a massive disruption.
30
3,065
Cool experiment!
This was a fun little experiment i tried out today. I generated an animation with H3, then generated a 3D model per-frame at 12FPS with tripo3d.
I've had this idea for about a year now but agents can now easily automate the process so it only took a few hours to try it out.
This is horrible on memory and probably little to no practical application, but its fun and looks kinda cool. I might mess around with the technique a bit more and opus made a nice little video of it too :)
1
5
768
Matt retweeted
Things are changing, never doom.
Jobs will look different, never doom.
Software is being eaten, never doom.
Increase your ambition, care deeply about your (changing) craft, build things you're proud of.
Never doom!
43
80
944
24,475
