Video · 2026-07-20 · 1h 9m · 6 moments

Open Models Replace Big AI ⟡ Every New Browser Feature ⟡ Death of Stack Overflow ⌁ Syntax Weekly ⌁

✦ AI generated

timeline · colored by role

Mechanism

Vue 3.6's new opt-in vapor mode compiles components into bundles that skip the virtual DOM entirely, making apps faster with less runtime overhead.

CJ explains that Vue 3.6 introduces an opt-in 'vapor mode' that compiles components directly to fine-grained DOM updates instead of using a virtual DOM, borrowing an approach popularized by Svelte and SolidJS.

transcript

CJ Reynolds: the idea with vapor mode is you can essentially build your Vue components, your Vue apps into a bundle that does not include the virtual DOM, does not include any extra stuffs. It's just as little JavaScript as possible.

gives example · 1

Mechanism

Rift speeds up creating parallel dev workspaces by using copy-on-write filesystem primitives to clone files instead of physically duplicating an entire repository like git worktrees do.

CJ describes Rift, a new tool from the Opencode team, as an experimental alternative to git worktrees that uses copy-on-write file cloning (e.g. APFS clonefile) to make isolated writable workspaces much faster to create.

transcript

CJ Reynolds: Rift did an experimental alternative to get work trees. It creates isolated writable workspaces using copy on write file system primitives instead of copying your repository.

Claim◆

Kimmy K3 is the first open-weight AI model whose benchmarks put it on par with frontier closed models like Opus and GPT.

CJ reports that Kimmy K3, a new 2.8-trillion-parameter open model released by a Chinese lab, is the first open model to match frontier proprietary models like Opus and GPT-5.5 on coding benchmarks.

transcript

CJ Reynolds: up until now most of the open models have been decent but have never really been able to be on par with like Opus level models or GPT55 level models. But now with Kimmy, we're there and the benchmarks look good and the rumblings of people using it on Twitter look good as well.

explains mechanism · 1gives example · 1provides context · 1

Definition

"Open weight" only means a model's trained numerical parameters are published; it does not disclose the training process or data and is a separate concept from open-source licensing.

CJ clarifies that 'open weight' is distinct from 'open source' — releasing weights only exposes the billions of trained numbers, not how the model was trained or what data was used.

transcript

CJ Reynolds: That at the end of the day results in a file that has billions of numbers inside of it. And so when we say open weight, they mean you can look at the weights. But the thing about neural networks is just looking at the weights doesn't tell you how it was trained or what they did special with it. You just can see the numbers. So that's what we say when we mean open weight.

explains mechanism · 1

Data◆

Stack Overflow's daily new-question volume has collapsed from roughly 200,000 at its 2017 peak to under 1,000 today, coinciding with the rise of ChatGPT.

CJ shows Stack Overflow's own query data revealing new-question volume cratered from ~200,000/day at its 2017 peak to under 1,000/day now, tracking closely with ChatGPT's release and adoption since 2022.

transcript

CJ Reynolds: So in 2022 we're at about 115,000 questions per day and then slow decline. In 2024 we're at 40,000 questions per day. By 2025 it's only 15,000 questions per day and now less than a thousand questions per day.

Claim

AI coding agents tend to bolt on quick patches to a flawed approach rather than stepping back to reconsider it, unlike the pushback a developer might get from a colleague or Stack Overflow.

Wes observes that where a human debugging partner or Stack Overflow answer might say 'you're approaching this wrong,' AI agents more often patch around the problem with quick fixes like added if-statements, leading to messier code over time.

transcript

Wes Bos: Often the answer to your problem will be like you're approaching this wrong. And I don't know that the quick fix from a lot of these AI agents is often let's take a step back and redo this. Often the fix is like, all right, well here's how you can fix it in the hole that you're in. We can patch it here, we can add an if statement when this thing pops in, and that'll cover that use case. And you start to make a bit of a mess.

Highlight slides

Kimmy K3: Open-Weight Hits Frontier Parity✦ from: Kimmy K3 is the first open-weight AI model whose benchmarks put it on par with frontier closed models like Opus and GPT.Stack Overflow's New-Question Collapse✦ from: Stack Overflow's daily new-question volume has collapsed from roughly 200,000 at its 2017 peak to under 1,000 today, coinciding with the rise of ChatGPT.Year-by-Year Drop in New Questions✦ from: Stack Overflow's daily new-question volume has collapsed from roughly 200,000 at its 2017 peak to under 1,000 today, coinciding with the rise of ChatGPT.

Related episodes

Video · a16z Podcast

Ben Horowitz on the Global Race for Tech, Power, and Influence

→

Audio · All-In Podcast

SpaceX-Cursor Deal, SaaS Debt Bomb, New Apple CEO, SPLC Indictment, Colon Cancer Spike

→

Video · Lenny's Podcast

The rise of taste, human authenticity and judgment in an AI world | Adam Mosseri (Head of IG)

→