Adam Hallett

@schgup27961 Is this an advertisement for Codex? lol

Single prompt → playable game. Built locally with: Qwen 3.6 (35B) Three.js ComfyUI (textures) No cloud. #threejs #gamedev #AI #localAI

@yacineMTB @SebAaltonen The harness maketh the engineer.

@SebAaltonen This Everything needs to be micromanaged The only way to keep this thing on rails is to force it to run an unshakable performance test every diff it makes Amusingly, when you do that, it knows it will get caught lying so it becomes a better programmer..

You have to review all LLM code! Codex 5.5 tried to push this awful hack to our Metal backend when it was coding font rendering. It decided to implement hacky "robust buffer access" style OOM check inside the shader and hacked our whole Metal binding architecture to add a special bind group slot 30 (hardcoded) to deliver sizes of all buffer bindings. This of course made the binding model super slow and required extra data for each buffer.

@Dimillian Too much stuff on the screen.

I've asked Codex for simple polygons but it seems like it cooked way beyond that goal

@JinjingLiang Free or Plus?

Somehow burned through my entire Codex limit in a single chat with GPT-5.5 xhigh fast.

@banteg i'm building an emulator and use 5.4xhigh but it takes days to make a small amount of progress. i asked codex when the project would be finished and it said 1-2 years.

i have a reverse engineering project i started with 5.3-codex. after a few days it hit a wall. then i had meaningful progress when 5.4 came out. but the original code seems quite convoluted and cursed, so it got bogged again. now 5.5 is breezing through it again. it's pretty interesting that there is a tangible wall in understanding for each model.

@mattforthelikes @OpenAI Yeah, I'm getting reconnects quite frequently.

Is @OpenAI Codex down or just me? First time having an issue.

@CtrlAltDwayne Pro users really need a way to differentiate ourselves from Plus users.

I'm a simple man. I see a free pen and I claim my prize.

@TheMaran Really skeptical they will spend time to walk to _your_ tree. Wouldn't they pull mangos from the closest tree? Because if one tree doesn't have a good harvest the customer will not return for the next year.

this Indian startup let you own a mango tree for $111.57 per season a kochi based startup, allows you to rent a mango tree and enjoy the entire harvest there are 3 types of trees you can rent base, standard, max you can get 30 kg to 60 kg mangoes from these trees this company is operating in 3 states in India, select your favourite tree, pay money you will get a dashboard with all the information about the tree you rented this company is literally connecting farmers with the direct customers who loves chemical free fruits

@stefnox @k1rallik It does sound so fake. Aren’t the opinions random? How can you get the right answer with randomness.

@k1rallik oh no... why this slop fake Polymarket things have started reappearing on my feed?

Someone trained a swarm model on 3 years of NBA data and let it loose on Polymarket The result: $1.49M. He didn't build a better model. He built a better crowd. Here's how the system actually works. He takes MiroFish - an open-source engine that simulates thousands of AI agents and feeds it raw sports data. The actual building blocks: > Player stat vectors: points, rebounds, assists, eFG%, usage rate across 3 full seasons > Team form tensors: last 10 games, home/away splits, pace-of-play, defensive rating > Matchup history: head-to-head records, positional mismatches, referee tendencies > Injury probability models: medical staff reports weighted by recovery timelines > Line movement tracking: where the sharp money flows before tip-off Check how this wallet works: @0p0jogggg?via=sales" target="_blank" rel="nofollow noopener">polymarket.com/@0p0jogggg?via… MiroFish generates 4,096 agents - analysts, bettors, insiders, oddsmakers - each with different reasoning. They argue, form clusters, shift opinions. Consensus emerges. He pipes that consensus into a 12-layer transformer trained on the full history of his 16,695 predictions. The model compares MiroFish output against live Polymarket odds. When the gap exceeds his Kelly criterion threshold, it enters. Lakers at 40 cents. MiroFish said 62%. One position: $190,823. The edge isn't data. Every sharp has data. The edge is simulating how 4,000 fake humans process it - before the real crowd does.

@boobs_scary @QuidditAI I have tried porting one graphics library to another. It is very difficult.

I'm locking in nobody speak to me for 72 hours I will have the wildest tech demo for you on Wednesday. I think I found the missing piece for this app.

The top post when I search for "Iran" on X.

@Gfilche @NoisetoS so true

@NoisetoS maybe. not like you will die. but like your business will die. that seems true

🚨 AI Is HyperChanging EVERYTHING. Adapt or die. the bigger the problem the bigger the opportunity. now is the time to learn AI and use it build the future.... or get left behind ⚡️⚡️⚡️⚡️

@tamrrat Does this violate street view TOS?

you can try it here! just paste any google street view url mint.gg/create

turn any google street view url into a 3D gaussian splat!

Okay, I hear you. But also: codex —yolo

@marcinthelab @sdamico I am pretty sure US companies are forbidden from supplying Russia’s army.

@sdamico Do we know why Starlink would agree to do this? Never got the impression that Musk was particularly pro-Ukraine.

a 35 year old tech guy became ukraine's minister of defense and did something incredibly obvious: got starlink to implement a whitelist to lock out russian users. the twist is, because the previous guy didn't do this, the russians became super dependent on starlink...

@theo @rfleury Shipping doesn’t mean anything anymore. Pure circlejerk.

>designs unique & innovative language, writes compiler for language from scratch, writes game engine in new language from scratch, designs state-of-the-art game in new engine Theo: “what a terrible developer” This guy is an embarrassment

@shrimpwtf @theo @AnthropicAI Let’s be honest… Your only option is Codex.

@theo @AnthropicAI this is the reason many such as myself are actively looking for an Anthropic replacement. Doesn’t matter how good your models are, your business model is trash

Seems like Anthropic lawyers sent some more love letters to OpenCode 🙃

@terezahurtova93 @asaio87 Before vibe coding all security vulnerabilities were missed by humans.

Vibe coding is amazing for prototyping. We use it too. But I would never rely on it for production - especially not with real client data. Today I spoke with a marketer hiring a “vibe coder” for client projects. They hadn’t even considered what happens to the data when you pipe it through these tools. That’s the real issue. Speed is great. Understanding the trade-offs is better.

People who say they can "vibe code" any Saas app in a week or less, they have never built software at scale or trying to sell you a nocode or vibe code tool. Trust me, its hard to build good software. Try it. I let you use AI

@spiffyman @HealthRanger They have been trying to do FSD for 10 years.

@atomical_on_git @HealthRanger What is Tesla's plan? Is it to build a bunch of new models and sell them to the public, like Rivian? Is that what they're trying to do? 🤔

So it looks like Kim Dotcom's leak was legit. There's a kind of civil war inside Palantir right now.

@spiffyman @HealthRanger Tesla isn’t growing anymore.

@HealthRanger I'm sure this has nothing to do with politics. Nothing at all. 🤔 It sounds like a combo of shorts and people on the Left that dislike Trump/Alex Karp and will do whatever they can to hurt the stock. Same story for Tesla for years. Enjoy.