- Published on
科技推特精选 - 2026年1月30日
- Authors
- Name
- geeknotes
今日科技动态：拥抱 AI 转型已成必然，否则将面临被淘汰的风险。知名投资人比尔·葛利（Bill Gurley）与 a16z 强调，当前正处于结构性变革之中，工作职能正被重新定义为一系列适合自动化的“任务集合”。开发者工具正加速这一进程：Cloudflare 通过 Moltbot 推出了安全 AI 托管服务，Replit 则引入了语音 AI 以简化创作流程。尽管“氛围编程”（vibe coding）正受到审视，且投资者在 AI 支出与增长回报之间反复权衡，但行业整体正坚定地迈向复杂的智能体工作流及下一代内容管理系统。
1. bgurley (Group Score: 60.4 | Individual: 44.2)
Cluster: 2 tweets | Engagement: 2682 (Avg: 309) | Type: Tech
100%! The number one way to protect your career from AI is to be the most aggressive user of AI in your field.
RUN AT IT.
Humans evolve with their tools. Keep evolving.
See 1 related tweets
- @eastdakota: RT @bgurley: 100%! The number one way to protect your career from AI is to be the most aggressive us...
2. CloudflareDev (Group Score: 47.9 | Individual: 47.9)
Cluster: 1 tweets | Engagement: 2238 (Avg: 173) | Type: Tech
We took @moltbot and brought it to @Cloudflare Workers! With the Sandbox SDK, it's secure by default. Check out the blog post to learn how it all fits together...
Or skip straight to deploying 😎⬇️
3. Cloudflare (Group Score: 46.0 | Individual: 46.0)
Cluster: 1 tweets | Engagement: 6198 (Avg: 665) | Type: Tech
Moltworker is a middleware Worker and adapted scripts that allows running Moltbot (formerly Clawdbot) on Cloudflare's Sandbox SDK and our Developer Platform APIs. So you can self-host an AI personal assistant — without any new hardware. https://t.co/BUlxsyu1fa
4. javinpaul (Group Score: 45.6 | Individual: 18.9)
Cluster: 3 tweets | Engagement: 47 (Avg: 9) | Type: Tech
I Joined 30+ Agentic AI Online Courses: Here Are My Top 6 Recommendations for 2026 https://t.co/c1HJQ3JWvx https://t.co/cis1vYsU5y
See 2 related tweets
- @javinpaul: I Joined 30+ Agentic AI Online Courses: Here Are My Top 6 Recommendations for 2026 https://t.co/Mu8q...
- @javarevisited: RT @javinpaul: I Joined 30+ Agentic AI Online Courses: Here Are My Top 6 Recommendations for 2026 ht...
5. Cohere_Labs (Group Score: 43.8 | Individual: 43.8)
Cluster: 1 tweets | Engagement: 423 (Avg: 81) | Type: Tech
Our community-led beginners in research driven studies (BIRDS) group is excited to announce their next sprint which offers 3⃣ specialized tracks designed to build deep expertise in NLP, CV and RL.
Participants will develop strong research literacy, reproducibility skills, and hands-on experience with ML and PyTorch.
Key topics include: 🏗️ RL concepts like MDPs, policy iteration, and value iteration 👀vision transformers (ViT) and CLIP 💬multilingual model challenges such as tokenization biases and cross-lingual retrieval 🌈advanced RL techniques like double Q-learning and Rainbow
Led by @VivekS100, @nuna_ade and Rafi Sudrajat - this sprint combines theoretical learning with practical implementation, equipping attendees with the skills to tackle cutting-edge AI challenges.
Join our open science community today: https://t.co/UZeGpcOkzC
6. bibryam (Group Score: 41.9 | Individual: 41.9)
Cluster: 1 tweets | Engagement: 10052 (Avg: 404) | Type: Tech
RT @karpathy: A few random notes from claude coding quite a bit last few weeks.
Coding workflow. Given the latest lift in LLM coding capab…
7. Replit (Group Score: 41.0 | Individual: 41.0)
Cluster: 1 tweets | Engagement: 243 (Avg: 40) | Type: Tech
Voice AI just landed on Replit.
No API keys. No setup.
Just vibe code voice apps. https://t.co/niadjir6iK
8. TDataScience (Group Score: 40.4 | Individual: 25.4)
Cluster: 2 tweets | Engagement: 6 (Avg: 4) | Type: Tech
The true potential of coding agents is unlocked not by running one, but by running many. However, this requires a structured approach. @EivindKjos shares a guide to running agents in parallel.
See 1 related tweets
- @TDataScience: If you’re still running one coding agent at a time, you’re leaving performance on the table. @Eivind...
9. a16z (Group Score: 39.6 | Individual: 39.6)
Cluster: 1 tweets | Engagement: 1418 (Avg: 638) | Type: Tech
Marc Andreessen:
"The job is not actually the atomic unit of what happens in the workplace. The atomic unit of what happens in the workplace is the task."
"A job is a bundle of tasks."
“Everybody wants to talk about job loss, but really what you want to look at is task loss.”
"As the tasks change enough, then that’s when the jobs change."
"Ten years from now, is your job title coder, or coder-designer-product manager, or is it just, ‘I build products,’ or is it just, ‘I tell the AI how to build products.’ Whatever that job is called, it’s going to be incredibly important, because the people doing that job are going to be orchestrating the AI."
@pmarca on Lenny's Podcast with @lennysan
10. OpenRouterAI (Group Score: 39.5 | Individual: 39.5)
Cluster: 1 tweets | Engagement: 519 (Avg: 72) | Type: Tech
GPT-5.2 Pro does about 50% of the $ volume on OpenRouter as GPT-5.2.
What are people using GPT-5.2 Pro for? Here's the category breakdown:
GPT-5.2 Pro is more heavily used for:
- Science (6.7% vs 2.8% for Standard)
- Finance (2.6% vs 1.3% for Standard)
- Legal (1.2% vs 0.5% for Standard)
GPT-5.2 Standard is more heavily used for:
- Academia (3.4% vs 2.0% for Pro)
- Programming (10.7% vs 9.8% for Pro)
- Technology (5.6% vs 4.7% for Pro)
See https://t.co/i04WYOw5u2 for more insights
11. AlexFinn (Group Score: 35.1 | Individual: 35.1)
Cluster: 1 tweets | Engagement: 995 (Avg: 783) | Type: Tech
"ClawdBot is nothing special. ChatGPT can do all of this"
OK, if you can show me how ChatGPT does the following use cases I've accomplished, I'll write you a check for a million dollars:
• Every night vibe codes me new tooling without me asking
• Self improves itself by building new memory systems. Now remembers every detail of every conversation (again, without me asking)
• Texts me proactively when a competitor posts a Youtube video that performs better than their average
• Continuously analyzing my own youtube videos, writing new scripts based on structures and hooks that work. Delivers a new script to my inbox every morning (one video it scripted already has $2,600 of ad revenue)
If you can show me how to do any of these use cases with ChatGPT without any additional tooling, the million dollars is yours.
12. ReutersBiz (Group Score: 34.3 | Individual: 19.0)
Cluster: 2 tweets | Engagement: 23 (Avg: 17) | Type: Tech
Big Tech earnings so far this week sent a clear warning: investors are willing to overlook soaring spending on AI if it fuels strong growth, but are quick to punish companies that fall short. Read: https://t.co/l51xZvfgoJ https://t.co/s5gX2nuM9y
See 1 related tweets
- @BusinessInsider: Big AI spending is being forgiven, as long as a company's core business is thriving. https://t.co/KK...
13. ibuildthecloud (Group Score: 33.9 | Individual: 22.5)
Cluster: 2 tweets | Engagement: 17 (Avg: 26) | Type: Tech
Trying to vibe code react has been really enlightening. It's kind of a worst case scenario. It's extremely hard and AI doesn't magically make it easy.
See 1 related tweets
- @BusinessInsider: OpenAI's chair says vibe coding will stick around, but the real disruption lies in how AI agents are...
14. webflow (Group Score: 33.3 | Individual: 33.3)
Cluster: 1 tweets | Engagement: 44 (Avg: 30) | Type: Tech
Today, we’re introducing the Webflow Collective Fund: a commitment to supporting community-led innovation and education.
The fund unlocks: – Application-based support for community-led projects – Funding for open-source, Marketplace, and developer innovation – Sponsorship for educational resources and masterclasses
Whether you’re hosting an event, teaching AI workflows, building Marketplace assets, organizing a community challenge, or launching a new community initiative, this fund exists to help you create impact.
To kick things off, we’re proud to sponsor our first project: a free, advanced Webflow Interactions course by @victorwork_, a 2× Community Challenge winner. The course launches in mid-march, sign up here to stay in the loop: https://t.co/9rsiHoLxib
👉 Learn more about the Collective Fund: https://t.co/GRs935k99d
15. webflow (Group Score: 33.1 | Individual: 33.1)
Cluster: 1 tweets | Engagement: 54 (Avg: 30) | Type: Tech
Next-gen CMS is now available for Enterprise.
As discovery shifts toward answer engines, structured content at scale matters more than ever.
Today, we’re excited to share that all Enterprise sites have been migrated to our next-gen CMS — a complete backend re-architecture unlocking significantly greater storage capacity, richer data models, and increased design flexibility.
The result? A future ready content foundation that scales with your business.
Broader availability is on the way.
👉 Read the full launch post: https://t.co/21lDEIb8Fn
16. teortaxesTex (Group Score: 32.8 | Individual: 32.8)
Cluster: 1 tweets | Engagement: 314 (Avg: 55) | Type: Tech
This is the vision that basically everyone serious has in mind ofc. If you don't think we're going to self-replicating industry that scales down to molecular and up to astronomical level, you don't take AI seriously the question is of time and cost
17. garrytan (Group Score: 32.8 | Individual: 32.8)
Cluster: 1 tweets | Engagement: 1740 (Avg: 303) | Type: Tech
Design is the last frontier that Claude Code can’t really help you with. You’re going to need Variant.
18. brk0v (Group Score: 32.6 | Individual: 32.6)
Cluster: 1 tweets | Engagement: 849 (Avg: 323) | Type: Tech
🦀 Building an LLM from scratch in Rust - Part 4 is where the model actually learns.
No autograd magic: you hand-roll backprop through linear layers, attention, layer norm, MLP.
Watching it go from pure noise to Shakespeare.
https://t.co/ggHZl87xRj #rust #rustlang #LLM #gpt https://t.co/7G1LuVGYec
19. garrytan (Group Score: 32.5 | Individual: 32.5)
Cluster: 1 tweets | Engagement: 756 (Avg: 303) | Type: Tech
Is there any way I can connect Claude Code to use my browser and iOS simulator so it can find mobile css bugs and make things responsive on tablet, web, and mobile automatically?
20. rasbt (Group Score: 32.1 | Individual: 32.1)
Cluster: 1 tweets | Engagement: 857 (Avg: 461) | Type: Tech
It's been a while since I did an LLM architecture post. Just stumbled upon the Arcee AI Trinity Large release + technical report released yesterday and couldn't resist:
- 400B param MoE (13B active params)
- Base model performance similar to GLM 4.5 base
- Alternating local:global (sliding window) attention layers in 3:1 ratio like Olmo 3
- QK-Norm (e.g. popular since Olmo 2) and NoPE (e.g., SMolLM3)
- Gated attention like Qwen3-Next
- Sandwich RMSNorm (kind of like Gemma 3 but depth-scaled)
- DeepSeek-like MoE with lots of small experts, but made it coarser as that helps with inference throughput (something we have also seen in Mistral 3 Large when they adopted the DeepSeek V3 architecture)
Added a slightly longer write-up to my The Big Architecture Comparison article: https://t.co/uCt4rFAK3a