AI 行业热点
🎙️ 播客精选
Inside Nemotron & NVIDIA’s AI Lab | Bryan Catanzaro
The MAD Podcast with Matt Turck · 2026-07-02
Speaker 1 | 00:00 - 00:21
If you accept as the truth that we’re gonna be running at the limit, then what that means is that the way to get more intelligence is to be more efficient. We can’t get more intelligence by applying more force if we’re already at the limit. We have to be more thoughtful about how we use what we have. We build tools. We build external organs that help us solve problems.
Speaker 1 | 00:21 - 00:30
You know, we we have an external stomach. We call it kitchen. Now we’re creating an external brain. What is the implications of an external brain? Pretty profound.
Speaker 1 …
🐦 X/Twitter 热点
Swyx (@swyx)
@mattpocockuk @trq212 what other people are recommending to use [8 ❤️ 1 🔄]
where i’m currently at for Big Boy projects:
sol ultra to plan
fable 5 to critique
sonnet 5/terra ultra/swe 1.7 to ultracode/slop cannon
devin review to review (using kakuna)
~always use a variant of @mattpocockuk’s grill-me or @trq212’s interview-me to elicit decisions upfront [364 ❤️ 7 🔄]
- @resend AGAIN. TODAY.
at this point if i get a text from my mom i expect codex to ask for a resend api key in order to read it [1 ❤️]
Thibault Sottiaux (@thsottiaux)
- Tomorrow might be 8M active user celebration day. Just saying [3531 ❤️ 123 🔄]
- Build your dreams [243 ❤️ 1 🔄]
- ChatGPT Work presents [1700 ❤️ 34 🔄]
Peter Yang (@petergyang)
- .@nikitabier is it possible to find accounts that just instantly reply to posts from bigger accounts?
Those are usually AI bots worth banning? I block a bunch every week. [21 ❤️ 1 🔄]
- Love the man but this isn’t a small tweak I only see neutrals now 😂
Not sure this will last tbh better enjoy it while we can! [10 ❤️]
- At the risk of losing friends:
Singlethreads: Flowers on sushi?
Benu: Thousand year egg I can buy from Chinese grocery store
Californios: Ok this was good but I had better food in Mexico
Birdsong: I remember this one fondly when they had 1 star
Nan Yu (@thenanyu)
- This becomes much more apparent once you examine the jobs of people with “designer” in their title outside of software! [35 ❤️]
- Chinese room btw [7 ❤️]
Cat Wu (@_catwu)
- Artifacts just got an upgrade! [212 ❤️ 6 🔄]
Thariq (@trq212)
- This makes artifacts much more expressive and can be combined in creative ways.
One of my favorites is to create a dashboard for a project in Claude Tag that can be edited by others or by my local Claude Code sessions. [845 ❤️ 45 🔄]
Amjad Masad (@amasad)
- Our code [125 ❤️ 4 🔄]
- Getting realtime progress updates on my model training runs.
This feels like early vibe coding except it’s making personal models. [180 ❤️ 4 🔄]
Guillermo Rauch (@rauchg)
- The two most popular features so far have been:
1️⃣ The ease of use / filesystem API
2️⃣ Observability
We’re doubling down on both! [236 ❤️ 13 🔄]
- Powerful building block for autonomous, self-optimizing websites and applications.
Give agents the ability to set up and tune experiments with feature flags. [145 ❤️ 3 🔄]
- “Open-weight models ran 29% of gateway tokens, up from 11% in April” [247 ❤️ 12 🔄]
Aaron Levie (@levie)
- A few thoughts on what we will see in AI structurally for the foreseeable future:
Frontier intelligence continues unabated and pushes the industry forward continuously. The top labs will continue to buy the best and the most data, build the most compute, be at the forefront of improved training breakthroughs, and so on. A few different approaches stratify the market on pricing and capability, but overall competitive pressure brings down pricing on a per task basis. That said, we just ask more from the models over time - as one thing gets cheaper, we just use more - so frontier spend and use remains robust.
Open weights rapidly absorbs frontier breakthroughs (and drives other breakthrough directions given the constraints), offering both lower cost intelligence and the ability to be post trained for specific workflows and domains. This creates a healthy counter balance to the frontier as you can run models “at cost” on a hyperscaler at any time, and tune models just for your tasks.
The Applied AI layer has a huge opportunity to combine frontier intelligence with open or cheap closed models to orchestrate workflows in any given domain. Due to evals, deep domain context, being trusted with enterprise data and workflows, this layer can maximize performance and cost combination. The applied AI layer will also often have their own RLed models especially for high volume, predictable tasks in their systems.
Individual enterprises will generally focus on their enterprise context, making sure they can get any AI system the right data and information to work with, in a continuously improving way. Some will go off and train their own models for specific areas of work (large banks, pharma, etc.) where they can get real alpha from doing so given the many tradeoffs, but most will spend energy on making sure they can get all of the gains from AI breakthroughs on their data and workflows.
Net net: even though some of this gets framed as zero sum, there’s just a ton of opportunity for all layers of the stack and approaches. [142 ❤️ 16 🔄]
- Here’s a great post on driving down costs, while maintaining high performance, with frontier intelligence as a manager and lower cost models for the workhorse tasks. This will be the template for what model routing looks like in the future.
“We started this experiment expecting to measure how much Fable’s 2x premium would increase cost. We were surprised to find that Fable’s effective delegation actually decreased cost overall. It specified constraints and outcomes instead of spelling out the implementation, gave feedback instead of making fixes itself, and in most cases never touched the code at all. These are the habits of a good manager.”
The industry is increasingly figuring out what it looks like to mix models together to be able to get targeted performance levels and optimal cost structures.
Of course, the only way to get this is to have a deep understanding of the business problem you’re trying to solve and how to effectively route work to different models. If you’re in the applied layer - whether it’s customer support, legal, finance, or coding - this is how your harness will become a core area of differentiation. [101 ❤️ 15 🔄]
- The biggest challenge right now with the topic of every enterprise having their own model is that your most valuable information and insights are not only always changing, but they’re often your most sensitive information.
Your most sensitive information can’t be packed into a model usually because it contains data that not everyone gets to have access to, and you can’t keep your security layer inside the model or an agent.
I think there are going to be 100X more use cases for custom trained models, especially inside of domain-focused products, but training a model per enterprise is going to be a lot harder than it looks. [187 ❤️ 23 🔄]
Ryo Lu (@ryolu_)
- works on Xteink X3 + X4
ask Cursor to flash it for you [48 ❤️] - built a custom e-reader firmware with Cursor, hardware hacking is fun!
made for:
– beautiful latin + CJK typography
– vertical layout, line breaking, large character sets (縱書．禁則)
– syncs books + progress with ryOS
– speedy rendering + caching [947 ❤️ 25 🔄]
- super stoked for jenny to join and lead our team 💛
makes me really happy for the team, and personally excited to regain the focus and mental space to dream big again [1937 ❤️ 17 🔄]
Garry Tan (@garrytan)
- Era of the Gentleman Scientist is so back [1009 ❤️ 53 🔄]
Zara Zhang (@zarazhangrui)
- The 3 levels of AI adoption for organizations
Most companies are at level 2 [62 ❤️ 5 🔄]
- @ashebytes [4 ❤️ 1 🔄]
- Back in March I recorded a 45-min conversation (outdoors!) with @ashebytes on building in public, growing my audience on X (without the slop), and my thoughts on vibe coding
Watch it here ⬇️ [27 ❤️ 2 🔄]
Nikunj Kothari (@nikunj)
- Btw this was one-shot by fable with a voice prompt while driving on the way to SF with some steering and edits obviously [2 ❤️]
- Built using @tryramp CLI (thx @zack_field) & @claudeai Fable.
Repo link here:
Fully open source so feel free to fork away or make improvements! [4 ❤️]
- So @tryramp’s core mission is “save money AND time”, but I was still spending time categorizing & adding receipts to my expenses.. at least till I discovered their CLI.
Introducing the Ramp-Autofill skill ✍️
finds receipts automatically from your iMessage & Gmail. if it’s in a link, it’ll use playwright to convert the web page into a pdf and attach to the receipt automatically.
fills in memos on who you were meeting using your google calendar events.
understands your style of writing memos and categorization required by your organization by reviewing past transactions. it will then auto categorize all the missing transactions.
verifies it’s work, flags any discrepancies and makes it easy to run as a scheduled job so you never have to look at expenses again.
Simple drop in for Claude Code (read what Fable thinks) - used it over the weekend to get through the last 60 days of expenses - fully open source! [25 ❤️ 1 🔄]
Peter Steinberger (@steipete)
- I moved our maintainer agent to the cloud and they are fighting already. [26 ❤️ 1 🔄]
- We shipped! iOS and Android apps also got updates and they look great.
Had to bump Node to keep things smooth, if the autoupdater doesn’t work, run the web installer and it’ll take care of that for you. [72 ❤️ 3 🔄]
- “stress test” is a good prompt. [164 ❤️ 2 🔄]
Aditya Agarwal (@adityaag)
- I don’t know if I am using Codex or ChatGPT (wtf)
But I do know that I am asking a AGI LEVEL coding agent:
“What necklaces does Benson Boone wear? (for my daughter) [11 ❤️]
Sam Altman (@sama)
- “hard questions are great but only if we deem you worthy enough to not silently downgrade you, or even get access at all” [4678 ❤️ 158 🔄]
- i thought this was satire, kept looking for the handle to be spelled c1audeai or something [8294 ❤️ 231 🔄]
- still sorta breaks my brain to see our models be good at design finally [6817 ❤️ 128 🔄]
📝 博客文章
由 Follow Builders 自动生成 · 2026-07-14