AI 行业热点
🎙️ 播客精选
Ep 87: Gemini Co-Lead on World Models, RL's Next Domains & Continual Learning
Unsupervised Learning · 2026-05-22
Speaker 1 | 00:00 - 00:28
Oriol Vinyals is the co lead of Gemini alongside Noam Shazir and Jeff Dean. He’s had an incredible career in AI, pioneering many of the breakthroughs in deep learning in the last decade, and it was a ton of fun to get to sit down with him after Google IO. If you’ve been following Google IO, they basically shipped a bunch of products across a a ton of interesting surface areas throughout AI. And so Oriel and I hit all of them. We talked about what’s required for further advances in multimodal models and what’s going to make these world models actually usable.
Speaker …
🐦 X/Twitter 热点
Swyx (@swyx)
- co-sign. a very handy mental framework for what kinds of learning transformers do well today, and why it runs into limitations. when @ankit2119 and i wrote about the need for adversarial world models earlier this year, we were describing a couple of the functions of these rungs of thinking that bring us ever closer to the kolmogorov-limit generator of reality. throwing more params, more power, more everything at a demonstrably inefficient paradigm will be outclassed by the simple solution that can hypothesize and seek truth rather than backfit a house of cards - although the bitter lesson is it is simpler to scale and we may hit agi anyway because human intelligence just isn’t that smart nor plentiful [13 ❤️ 1 🔄]
- [11 ❤️ 1 🔄]
- Kakuna: skills with checklists that only know how to harden your codebase
/plan with it then let it /goal for a day, it comes back with same functionality but all the boring stuff done for you + an audit of its own work.
focus on subagent parallelism and encodes strong opinions on how AI engineers should design apps for human and agent access/devops/product mgmt.
instead of dark factory, go “mullet factory” - party in front (ship unique lovable features), dark in the back (timeless production principles). basically its the antientropy/antislop part of symphony broken out as its own thing
not gonna go psychosis further than this but extend as you will. link below [153 ❤️ 11 🔄]
Kevin Weil (@kevinweil)
- A friend shared this with me, and I love it so much. Make no little plans. [32 ❤️ 3 🔄]
Peter Yang (@petergyang)
- Lately, I’ve been obsessed with how the best solo founders and engineers use agents to 10x their output. I ask them questions like:
- What’s your AI stack?
- Show me how you build end to end.
- How do you manage multiple agents?
📌 Subscribe to get all these episodes soon (episode with Ryan dropping this Sunday): [26 ❤️ 2 🔄]
- Read more here: [1 ❤️ 1 🔄]
- I hate seeing all the mass layoffs. Here are 6 things you can do as an employee to take back control:
- Read the signals
If business growth is stagnant and leadership is suddenly obsessed with “flatter orgs” or “restructuring for the agentic era,” you already know what’s coming.
- Learn Codex or Claude Code
These apps are the best training grounds for working with AI agents. Tell them about your job and ask them what you can build. If your company doesn’t let you use these tools, use them for your personal projects.
- Build side projects
Ship a small product or make an open-source contribution. If you work at a large company, your builder skills have probably atrophied. It’s time to bring them back.
- Develop a GitHub history
If you do step 3 consistently, you’ll develop a body of work on GitHub over time. My friend @zarazhangrui is a great example — she built a frontend-slides skill that now has 16K GitHub stars and has been shipping non-stop since.
- Become top 10% at your craft
AI gets people to average really fast. But that just means customers are willing to pay more for human craft and taste. Pick one skill you genuinely enjoy working on and put in the reps until you’re in the top 10% at it.
- Let the market determine your value
Build in public, solve real user problems, and craft great products. That’s how you’ll get noticed by great employers, not by submitting 100s of resumes. And if all else fails, consider becoming a founder. I think entrepreneurship is the safest job in the AI era anyway. [125 ❤️ 11 🔄]
Google Labs (@GoogleLabs)
- Take a quick break from scrolling and check out 👀 We had a bit of a refresh! Our goal was simply to make sure you can easily find the latest and greatest innovations from the Lab, including those we just announced at I/O.
Explore our portfolio, test a few experiments out, learn more about Google Labs, and help us shape what’s next. 🚀 [301 ❤️ 29 🔄]
- Check out these experiments features at [12 ❤️ 3 🔄]
- We asked the team: What’s the most underrated or surprising feature from their product?
Here’s what they said. 👇 [82 ❤️ 6 🔄]
Aaron Levie (@levie)
- Here’s a key line in this mythos update. This is precisely an example of why engineers don’t go away, ever.
We’ve made it far easier to create and find security issues, which means the new bottleneck is our ability to actually review, respond to, and fix the issues.
Far from AI magically solving all of this, there still is major triage work and human judgment required to do the follow on work to actually protect systems. As a result, we’re about to enter a security engineer boom.
Jevons paradox all over again. [237 ❤️ 21 🔄]
Garry Tan (@garrytan)
- GBrain is my gift to you so you can have the same personal AI that I do. It’s experimental but getting better every day. MIT License. [35 ❤️ 2 🔄]
- GBrain just shipped v0.40.0 gives your OpenClaw/Hermes Agent + GBrain a voice agent.
It’s based on Gemini Live. (Thanks @demishassabis it’s amazing) Large context, great tool use, full brain access.
Mars is a friend, Venus is your EA.
My open source gift to you. [159 ❤️ 8 🔄]
- Geoffrey Moore says startups die in the chasm because pragmatist buyers demand a “whole product.” These folks won’t tolerate gaps. They need references. They need the complete solution. The chasm is lethal because because the buyers won’t buy without perfection.
But Moore’s model assumes there’s an EXISTING solution the buyer is comparing you to. The whole framework assumes the buyer has a status quo they’re comfortable with.
When the bar is zero, when the alternative is literally “we die” or “we do this entirely by hand with 2,000 people” (Block’s compliance team) or “we just don’t have this capability at all”?
The chasm doesn’t exist for those.
Buyers start acting like visionaries instead of skeptics, because they have to buy. The alternative doesn’t exist. They’ll tolerate a 60% solution, missing features, no references, because 60% of something beats 100% of nothing.
The companies I get most excited about aren’t disrupting incumbents. They’re filling voids. 9 Mothers in the YC Spring 2026 batch is a counter-drone defense co for whom bar is zero, there is no viable close quarters defense otherwise! There’s no chasm to cross for that.
The practical implication for founders: if you’re in a market where the bar is zero, stop worrying about whole product, stop worrying about crossing the chasm, stop worrying about pragmatist references. Ship the 60% solution. They’re begging for it.
If you’re NOT in a bar-is-zero market (if there’s an incumbent, a status quo, a “good enough”) then Moore applies in full and you need the whole playbook (beachhead, bowling alley, whole product, the works).
The question every founder should ask: is my customer’s current alternative literally nothing? If yes, you’re in a different game than the textbooks describe. Ship it in whatever form you have. You’ll know. And it’s a great place to be. [77 ❤️ 8 🔄]
Matt Turck (@mattturck)
- AI progress has been pretty wild over the last few months but behind the scenes at @OpenAI, it’s continuous progress compounding - @yanndubs [16 ❤️ 2 🔄]
Nikunj Kothari (@nikunj)
- Docs signed, wire released, excited to lead the Series A of a special company..
Perfect way to start the weekend 🙏🕺
PS: It’s not AI [179 ❤️]
- This time is too important to NOT be doing your life’s best work.. [163 ❤️ 10 🔄]
Peter Steinberger (@steipete)
- We used bots so far to enforce a 10PR per “person” limit. Great to see GitHub shipping that natively! [232 ❤️ 2 🔄]
- Moved to [19 ❤️ 2 🔄]
Dan Shipper (@danshipper)
- I’ll be talking about my piece: After Automation! [24 ❤️ 1 🔄]
- so fun to speak at this! [27 ❤️ 1 🔄]
Claude (@claudeai)
- From The Problem Solvers, our series featuring founders taking on hard problems with Claude: [109 ❤️ 4 🔄]
- Kay Zhu is the co-founder and CTO of @genspark_ai, the all-in-one AI workspace built on Claude.
In a market moving this fast, where anyone can build, he thinks the team is what makes the difference: [1536 ❤️ 97 🔄]
📝 博客文章
由 Follow Builders 自动生成 · 2026-05-23