[KongchangAI]
Back to podcasts
Tech Frontiers·5:37·Dialogue

Daily AI Buzz · November 4th Midday Edition: AMD Data Center Surge & Agent Skills Tutorial Craze

November 4, 2026 (Wednesday) Midday Edition — AI news roundup + in-depth hot topic discussion

Tech Frontiers

Daily AI Buzz · November 4th Midday Ed…

ListenDialogue
0:005:37

Referenced in This Episode

Link

Product

  • The Legend of Zelda: Ocarina of Time Remake— Switch 2 fall remake, art shifts to a realistic style, release window lands just before GTA6
  • Claude Max 20x Ultra— Claude Max 20x renamed to Ultra, verified to have no 5-hour limit

Discussion

  • Codex Speedrun Tutorial: From 0 to 1— Bilibili's ultimate hands-on Codex tutorial, aimed at developers who want to get started

Reddit

  • Hardware-in-the-loop for Claude— Using Claude Code to complete full-device hardware development, simulation deviates from reality by about an order of magnitude
Lots of new stuff happening in AI today, and a few of these are pretty跨界. Today we're covering a few main things: first, AMD's record-breaking earnings; second, the Zelda remake; third, Nexus Mods' AI disclosure policy; fourth, a hands-on Codex tutorial; fifth, Claude doing hardware; sixth, the Agent Skills tutorial; and seventh, the Claude Ultra subscription. Yeah, that's a pretty packed lineup. Right, so let's go in order, starting with the hardest-core item — the earnings report. Just saw on The Verge that AMD released its Q3 earnings — $13.4 billion in revenue, a company record. And next quarter's guidance is $17 billion — that's a huge jump. Lisa Su said this is the biggest and most profitable quarter in company history, mainly driven by data center. That's not the key point though — the key point is the revenue mix has shifted. AMD used to rely on volatile consumer CPUs and gaming, now it's tilting toward data center: longer order cycles, higher unit prices, higher margins. So is that good for cloud providers? At least they've got a bit more bargaining room. Yeah, but whether they can deliver on that $17 billion guidance depends on production capacity and delivery pace. It's not just about painting a pretty picture. Alright, that's it for earnings. Next up, gaming — Nintendo is preparing a remake of Ocarina of Time for the Switch 2. It's been 28 years since the N64 original. Word is this isn't just a resolution bump — the art direction is shifting from low-poly cartoon to something closer to modern AAA realism. Nintendo first-party going realistic is pretty rare in itself. And the timing is right before GTA6's launch — clearly aiming to be the fall blockbuster. But at the end of the day, it's still filling the schedule with a classic IP remake — nostalgia plus new hardware showcase, trying to have it both ways. Release date, price, exclusivity strategy — none of that's been announced yet. We can only see the direction for now. Third item — Nexus Mods, the biggest mod site in PC gaming, has taken a stance on AI. Newly uploaded mods must declare whether they use AI technology. Note: it's not a ban, but mandatory disclosure. That's a smart approach — disclosure over prohibition, and over laissez-faire. It gives downloaders informed choice. But the problem is it relies entirely on voluntary reporting, and the verification plan hasn't been announced. So will enforcement actually hold up? That's the real test — whether the rules can be implemented and whether the community can reach consensus. Everyone running UGC platforms is watching this template closely. Okay, fourth item — there's a hands-on tutorial on Bilibili for speedrunning Codex from zero to one. The title is literally 'Ultimate Hands-On Tutorial.' These kinds of tutorials are really popular right now — shows people genuinely want to use Codex, not just watch from the sidelines. Fifth item is even more hardcore — someone on the Reddit robotics subreddit used Claude Code for full hardware development. Motors, dozens of sensors, custom enclosure, motherboard — circuit boards described with Python scripts that directly generate fab files, and CAD changes can automatically recalculate weight and balance. The key is hardware-in-the-loop — firmware gets validated on a virtual prototype first, running in Isaac Sim or Unreal. Simulation versus real physics deviation is roughly within an order of magnitude. Within an order of magnitude is enough for directional decisions? That sounds a bit sketchy. Enough to screen options, not enough to finalize. The author himself says it's better suited for people with hardware backgrounds, and token consumption can be enormous — the real project accumulated about 800GB of simulation data. The methods and skills are open-sourced though — that's pretty solid. Alright, that's it for the news. Now let's dive into a topic that's been really hot on Bilibili lately — Agent Skills development. There's a tutorial called '2026 Master Agent Skills Development Full Course' — WorkBuddy plus Codex plus DeepSeek Harness combo, marketed as something even beginners can learn. That combo is interesting — WorkBuddy handles orchestration, Codex writes code, DeepSeek Harness does evaluation or task running. Basically connecting the whole pipeline. But I have to ask — is the Agent Skills concept getting a bit overhyped? Like everything can be shoved into it. Your feeling isn't wrong — the hype is real. But think about it: if Agents are going to actually do real work, it's these reusable skill modules that make it happen. The point isn't the concept — it's whether it can be standardized. There's also another one on Bilibili — a Claude Code hands-on tutorial aimed at real engineers, with solid engagement scores too. Two tutorials pushing at the same time shows the demand is real. Right, and people on Reddit and HN are debating too — one camp thinks Skills is just prompt engineering repackaged, the other thinks it's the necessary path to Agent engineering. Which side are you on? I lean toward the latter. If you look at a single skill in isolation, sure, it looks like repackaging. But if skills can be composed, evaluated, and version-controlled, that's a different story. Alright, moving on to item seven — Claude Max 20x has been renamed Ultra, and verified Ultra users have no five-hour limit. That change looks small but it's actually pretty key. The five-hour limit has always been the biggest headache for heavy users — removing it basically pushes Claude into all-day workflows. But there's also griping on YouTube saying it's a disguised price hike — repackaging what Max used to offer as Ultra. Both takes have merit. Commercially it's tiering, product-wise it's loosening the reins for high-frequency users. The key is how high the verification bar is — hopefully it doesn't become 'unlimited in name, still limited in practice.' So to sum up — Agent Skills is moving from concept to engineering, and Claude Ultra's adjustment is making way for heavy users. These two things are really two sides of the same coin. Exactly — the more powerful the tools, the deeper they dig into professional workflows, and subscription tiering and skill standardization will happen simultaneously. Alright, that was a great chat today — from AMD earnings all the way to Agent Skills and Claude Ultra. See you tonight.

More podcasts

Tech Frontiers

GitHub Trends · Nov 10: Two Agent Skil…

·5:33·Dialogue

GitHub Trends · Nov 10: Two Agent Skill Libraries Go Head-to-Head

Deep dive into 5 new GitHub trending projects today

Listen
Tech Frontiers

Daily AI News · Nov 10 Evening Edition…

·6:00·Dialogue

Daily AI News · Nov 10 Evening Edition: Fable 5.2 Tops Again & Amazon Cuts Costs

November 10, 2026 (Tuesday) Evening Edition — AI news roundup + deep dive on the day's hot topics

Listen
Tech Frontiers

Daily AI News · Nov 10 Midday Edition:…

·3:38·Dialogue

Daily AI News · Nov 10 Midday Edition: Anthropic IPO and Fable 5.2

November 10, 2026 (Tuesday) Midday Edition: AI news roundup + deep dive on the hot topics

Listen