Wil Chung (@iamwil) header
Wil Chung (@iamwil) profile picture

Programming & Local-first software ⁙ LLM zine https://teapawt.net/t.co/apQqD9PIMd ⁙ Startup lessons ⁙ Game design quips.

Mountain View, CA
Joined September 2007
Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)Photo by Wil Chung (@iamwil)
I wonder what the code looks like for these. Is it doing triangles? Or signed distance fields? Something else? I wonder if you can get beyond simple shapes that’s commonly seen in these videos that way.
Claude Fable 5.2 made this 90's toy commercial Everything you see was made with code. Animated in Python and rendered in Blender
1
94
Apparently, in this one, it's all triangles and shaders. teapawt.net/felixrieseberg/status/…
One thing that's surprising about this game: EVERYTHING IS CODE. All textures and models are functions. There are no images or 3D models. Claude didn't draw, it wrote shaders. Everything is Math. That sounds wild - the source is in the HTML, so you can inspect it yourself.
38
Generated a game from a loose prompt to see what I'd get. The game itself isn't important here, nor the existence proof that you can do this, for kids and adults alike. To me, it's that a) you can try out lots of ideas in quick succession to find the fun, even w/ vague direction.
1
49
But importantly, b) what makes a game fun is still in many of the fine-grained details. You're going to have to do balance, mechanics, *juice*, etc. I suspect that humans will still be better at this for a long while cuz it's not economically valuable (comparatively) for labs to do this well.
29
State management is one of the biggest sources of complexity. Nipped that in the bud early by asking agent specs to explicitly model state.
Vibecoders, 𝗜'𝗺 𝗯𝗲𝗴𝗴𝗶𝗻𝗴 𝘆𝗼𝘂, open your messiest project and paste this into Claude Code or Codex: "Find all of the implicit state machines in my app. Make the best decision that will elevate all three: the user experience (UX), developer experience (DX), and agent experience (AX) always, without breaking anything."
68
Where cloning games with vibe coding makes sense is to eject from play-to-win mobile games. If you own your own addiction, no one else can exploit it. Vibe code your own fake shopping sites to feed your shopping addiction without eating into your wallet.
1
45
The story keeps getting crazier. To recap everything that has been reported so far: - The US intelligence community discovered that Jensen Huang, the CEO of the largest company in the world, was sleeping with Alice Huang, a Chinese spy and an executive at Megaspeed. (FT) - Multiple independent investigations have concluded that Megaspeed is likely a Chinese cutout, which helps Chinese companies access AI chips or circumvent export controls. (NYT and Bloomberg) - The US intelligence community also concluded that Alice Huang’s goal was to convince Jensen to lobby the U.S. government to loosen export controls on advanced AI chips. (NYT) - The intelligence community relayed this information to “a small group of top US government officials” earlier this year. (FT) - Over the last year, Jensen Huang has repeatedly lobbied the U.S. government to loosen export controls on advanced chips. @BISgov has done so, not only by permitting the export of H200 chips to China, but also not enforcing export controls that would have required a license for exports of the most advanced Blackwell chips to Megaspeed and Southeast Asia. - Although @CommerceGov leadership was presumably briefed on this information by the U.S. intelligence community, @BISgov has taken no actions to tighten controls on Megaspeed or Southeast Asia. - Megaspeed remains a preferred cloud partner of Nvidia, and continues to receive Nvidia Blackwell chips despite employing a Chinese spy who reportedly is romantically engaged with the CEO of Nvidia. - Tens of billions of dollars worth of AI chips have been exported to Southeast Asia this year alone, and Malaysia and Singapore are now two of the four largest importers of AI chips in the world. What is going on here??? We need answers. The question of who gets the most advanced AI chips in the world is literally the defining strategic question of our time. And not only does it seem like the MSS may be directly influencing the debate on export controls in the United States, but also @CommerceGov is refusing to listen to the warnings of the IC on this issue. What is a POSSIBLE good explanation for why Megaspeed is still receiving chips? Why is there not a license requirement for AI chip shipments to all of Southeast Asia? This is beyond strange at this point. The fallout from this story will be massive. Obviously, the motivations underpinning Nvidia’s policy views opposing export controls must now all be questioned. But also, how did @BISgov sit on this information and has still taken no actions? We need answers.
1,283
6,142
953
28,141
9,357,930
Have agents write unit tests or not? Conduct adversarial reviews or not? We no longer have to take the guru's word for it. Few seem to realize it's now the age of experimental software engineering. Just like there's experimental physics that gathers empirical data to verify theoretical physics, we can do the same for agentic software engineering. Not that EX SWE didn't exist before. It did, but it was costly in time and setup to observe engineers in their natural habitat. Agents drive that cost down dramatically, and we can run multiple in parallel now too. ML research already leads the way in this regard. The rest of SWE should follow. We should start publishing our experiments, trying different methods, and interpreting the results so we can get out from under the cloak of heuristics, rules of thumb, tribalism, and superstition.
2
60
Intersection of two threads on twitter: using airplane Doppler in audio/video to pinpoint recording location and video veracity "AI or not?". I would think for AI to fake this over time, it'd need to have access to the flight path DB for all aircraft. teapawt.net/AdamHoltererer/status/…
After having Codex check 120 different generated videos, models do NOT generate low-frequency airplane Dopplers
1
56
I don't know how often flight paths change, but it doesn't seem impossible for AI to fake it on a typical day. But it's the deviations from normal flights that would catch it. That said, it's a tax, and points to one where we can check against an on-going real world signal.
18
This is not an absolute. You need system evals.
maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis
77
This is weird and surprising to me. It seems unlikely models were directly trained or RL'd to do this, yet it seems like it's able to do it--which seem aligned with the claim that neural networks are already clean and interpretable. teapawt.net/rikiyatakehi/status/21…
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
1
1
146
Ah, it all comes together after those experimental games on dragging around tree branches, in both the literal and abstract sense.
Drag-and-drop makes the digital graspable, but building it well is fiddly enough that many don’t bother. Declarative Dragging (with Josh Horowitz & Jeff Heer, #uist2026 best paper award), lets you make an element draggable just by listing the states dragging it can reach.
3
114
The spooky Halloween story this year is about how agentic coding resurrects UML diagrams from the dead.
103
Clearing out tabs (6): Some of this have been sitting for weeks if not months. A quick scan is simply that we should do system evals for our harness and context. And the first step is recording traces. Not much to see here. teapawt.net/hwchase17/status/20404…
1
118
Clearing out tabs (5): So much random garbage in my tabs. With AI, I'm starting to see a Cambrian Explosion of fun custom websites--after the crunch of everyone having a medium/fb/substack. names.color.pizza/?view=pixe…
18