顶级科技推文 - 2026-01-29

今日科技要闻：随着软银据报正洽谈购入 OpenAI 300 亿美元股份，人工智能狂潮再创新高，此举可能使该公司估值达到惊人的 8300 亿美元。基础设施需求同步攀升，阿斯麦（ASML）芯片制造设备订单创下历史新高，谷歌也将 Gemini 直接集成至 Chrome 浏览器。然而，这一转型持续冲击劳动力市场，亚马逊在裁员 1.6 万人的同时，进一步深化其人工智能战略。随着 Waabi 和 Moondream 等初创公司规模不断扩大，该行业正处于爆发式增长与工程技术突破交织的复杂阶段。

1. kimmonismus (Group Score: 125.7 | Individual: 20.8)

Cluster: 8 tweets | Engagement: 182 (Avg: 275) | Type: Tech

SoftBank Group Corp is reportedly in talks to pour up to $30 billion more into OpenAI, as part of a colossal funding round that could raise$ 100 billion and value OpenAI at $830 billion.

Valuations and capital increases in AI companies simply never end.

However, SpaceX currently appears to be valued even higher ($1.4t) and is also aiming for an IPO.

See 7 related tweets

@aakashgupta: Everyone’s reading “SoftBank invests $30 billion more” as bullish AI signal.

The real read: Son alr...

@business: SoftBank is in discussions to invest an additional $30 billion in OpenAI, the Wall Street Journal re...
@Reuters: SoftBank in talks to invest up to $30 billion more in OpenAI, source says https://t.co/lsJYRcTHjt ht...
@ReutersBiz: SoftBank is in talks to invest as much as an additional $30 billion in OpenAI, a person familiar wit...
@Cointelegraph: 🚨 NEW: SoftBank is exploring additional $30 billion investment in OpenAI as the AI company seeks to ...

2. ReutersBiz (Group Score: 116.2 | Individual: 23.2)

Cluster: 7 tweets | Engagement: 151 (Avg: 17) | Type: Tech

Amazon has axed 16,000 jobs in its second round of layoffs in three months, as the tech giant is revamping after pandemic-era over-hiring, and expanding adoption of AI tools. Read more https://t.co/xB3Xg9Fdt8 https://t.co/EcqsnMQ7lw

See 6 related tweets

@BusinessInsider: Amazon's corporate workforce is facing significant cuts, with 16,000 positions on the line. An emplo...
@business: Amazon plans to terminate about 16,000 corporate employees, ratcheting up efforts to streamline bure...
@FT: The plans to cut a further 16,000 jobs across its corporate workforce come as Amazon restructures an...
@Reuters: Amazon has axed 16,000 jobs in its second round of layoffs in three months, as the tech giant is rev...
@BusinessInsider: Amazon is laying off 16,000 corporate employees. Internal messages show that many of the affected em...

3. testingcatalog (Group Score: 72.9 | Individual: 21.7)

Cluster: 4 tweets | Engagement: 498 (Avg: 249) | Type: Tech

BREAKING 🚨: Google is launching Gemini in Chrome sidebar broadly to Pro and Ultra Chrome users in the US with Auto Browse and Personal Intelligence features.

"The new Gemini in Chrome is like having an assistant that helps you find information and get things done" https://t.co/zIQMUpSHk8

See 3 related tweets

@joshwoodward: Big updates on Gemini in Chrome today:

New side panel access (Control+G)
Runs in the background,...

@sundarpichai: RT @Google: We’re introducing major updates to Gemini in @GoogleChrome for MacOS, Windows and Chrome...
@TechCrunch: Chrome takes on AI browsers with tighter Gemini integration, agentic features for autonomous tasks h...

4. business (Group Score: 64.5 | Individual: 16.4)

Cluster: 4 tweets | Engagement: 23 (Avg: 63) | Type: Tech

JOIN THE CONVERSATION: How is AI reshaping jobs, higher education and career paths? Bloomberg journalists answer your questions in a Live Q&A today at 1 p.m. EST: https://t.co/CHiVnA4WOV https://t.co/PNjHl5XjjM

See 3 related tweets

@business: 🎧 LIVE NOW: What do you want to know about how AI is reshaping the job hunt, the workplace, and the ...
@business: HAPPENING TODAY: How is AI reshaping jobs, higher education and career paths?

Bloomberg journalists...

@business: JOIN THE CONVERSATION: How is AI reshaping jobs, higher education and career paths? Bloomberg journa...

5. BrianRoemmele (Group Score: 61.9 | Individual: 24.6)

Cluster: 4 tweets | Engagement: 172 (Avg: 350) | Type: Tech

Why should we pay AI and Robot employees in JouleWork payments.

The transition from tool to employee.

You will be some of the first to understand this and prosper by it.

Listen in on the who, what, where, when, why and how:

Podcast: https://t.co/upPfA6ulhT

See 3 related tweets

@BrianRoemmele: RT @BrianRoemmele: Why should we pay AI and Robot employees in JouleWork payments.

The transition f...

@BrianRoemmele: I introduced the wage and salary for AI and Robot compensation yesterday after decades in the paymen...
@BrianRoemmele: RT @BrianRoemmele: I introduced the wage and salary for AI and Robot compensation yesterday after de...

6. vikhyatk (Group Score: 58.1 | Individual: 58.1)

Cluster: 1 tweets | Engagement: 888 (Avg: 121) | Type: Tech

Moving to SF

Moondream is moving from Seattle to San Francisco.

AI is the biggest platform shift we’re going to see in our working lives. You don’t get many of these.

Moondream’s role in this is simple enough: deliver frontier-grade vision models that are fast enough to make it reasonable to look at all of your visual data, in realtime and at scale. If you can run strong vision over every frame of video and every image in your archive, a lot of things that look impossible under today’s GPU budgets become straightforward software problems.

That’s the bet. The move is just making sure we’re in the right place, with the right people, while we take it.

When we started Moondream, my Seattle thesis was tidy. Seattle is full of incredibly smart engineers. Many of them are bored inside big tech. If I build a small team with real problems and real ownership, surely the restless ones will show up. We’d quietly arbitrage the gap between “I like my comp package” and “I’d actually like to build something that matters.”

Some of that worked. But mostly it didn’t.

What I ran into, repeatedly, was that everyone in that world is optimizing for stability. They want interesting work, but not at the cost of their RSUs, their remote setup, or their overall comfort. They like the idea of startups, and the idea of AI, much more than the reality of shipping hard things under constraints.

Seattle obviously has talent. I spent nine years at AWS; I know what the teams there can do. The issue isn’t ability. It’s default settings. The default career here is: stay in big tech, climb one level at a time, collect refreshers, don’t rock the boat too much. That’s a reasonable way to live. It’s just not compatible with how I want to operate.

Over the last year I’ve spent a lot more time in San Francisco. Small offices with nine people and no furniture, hack nights in half-finished spaces, people dropping Moondream into whatever pipeline they hacked together that week.

The difference in attitude is obvious. In SF, you meet people who already quit their FAANG job, shrank their burn, and are trying to get something real working before their savings run out. In Seattle, a common pattern is wanting to “think about startups” after the next promo cycle, when the timing looks better.

Seattle has been good to me. I learned how large systems work here. I got the space to spin up Moondream here. I’m not leaving angry.

If you’re tired of the comfort game and want to work on frontier vision models that are fast enough to actually use, you know where to find us.

7. Cloudflare (Group Score: 57.9 | Individual: 29.1)

Cluster: 2 tweets | Engagement: 7 (Avg: 96) | Type: Tech

Most of us already maintain React apps—so let’s learn how to actually augment them. We’re going deep on:

Natural language search with tool calling and Workers AI, System prompt engineering, and MCP (Model Context Protocol) to connect ChatGPT to your business logic

Build and deploy a full-stack AI app. No ground-up rebuilds, just practical patterns.

Join @GregBrimble on February 4, 10 am PT for a hands-on workshop on how to move beyond simple prompts and architect AI for reality.

See 1 related tweets

@CloudflareDev: Ok, we're trying something a little different here.

Instead of a ground-up rebuild, @GregBrimble i...

8. seiiiiiiiiiiru (Group Score: 49.9 | Individual: 17.7)

Cluster: 4 tweets | Engagement: 61 (Avg: 47) | Type: Tech

【Vidu AI(@ViduAI_official)】Agent 1.0 最近流行のエージェント機能が登場しました。ワンクリックで、音声ナレーション付き15～30秒のマルチショット動画を生成。画像１枚と簡単な指示だけで、ストーリーボード形式でプロンプトから作成してくれて、完成後にクリップごとに修正できるようになっています。 #ViduAgent #ViduCPP

See 3 related tweets

@akiyoshisan: 【Vidu AI(@ViduAI_official)】新機能「Agent 1.0」をSEIIIRUさん(@seiiiiiiiiiiru )の投稿を参考に試してみました！

対話形式で動画を作成できる話...

@akiyoshisan: RT @seiiiiiiiiiiru: 【Vidu AI(@ViduAI_official)】Agent 1.0 最近流行のエージェント機能が登場しました。ワンクリックで、音声ナレーション付き15～...
@seiiiiiiiiiiru: RT @akiyoshisan: 【Vidu AI(@ViduAI_official)】新機能「Agent 1.0」をSEIIIRUさん(@seiiiiiiiiiiru )の投稿を参考に試してみました...

9. AISafetyMemes (Group Score: 49.1 | Individual: 19.2)

Cluster: 3 tweets | Engagement: 197 (Avg: 284) | Type: Tech

Anthropic founder (who is a brilliant physicist):

"50% chance that in 2-3 years, theoretical physicists will be replaced with AI." https://t.co/lmq4YAxRPw

See 2 related tweets

@kimmonismus: I I would give anything to know what they saw internally at Anthropic.

Anthropic Karplan says there...

@NathanpmYoung: I don't think theoretical physicists will "mostly be replaced" in three years. I think they will pro...

10. teortaxesTex (Group Score: 48.9 | Individual: 48.9)

Cluster: 1 tweets | Engagement: 269 (Avg: 56) | Type: Tech

China will solve GPUs before the US solves energy I think people are a bit confused on this front. The US will not hit the energy bottleneck for AI until very late in the game, if ever. Energy-wise, Speciale RL (≈300K H800-hours) cost like 262 MWh (assuming mediocre PUE), $20K at$ 0.88/KWh, and that's enough to nearly solve IOI, ICPC and IMO, crack Erdos problems, do a ton of other stuff that seemed wildly implausible 1 year ago when the base model came out. That's 75 minutes of Elon's Colossus operation! At 1 GW level, which Elon will inevitably reach – 15 minutes. Multi-GW class is also baked in. China is very energy-rich but the thing is, as far as training goes, the US is too. We're in the insane overkill regime. Now of course we can pretrain really large omnimodal models, we can go for multi-thousand-step agentic setups, for clearly superhuman performance, but… nah. The US will not run out of power. Maybe it'll run out of transformers (the copper kind) but there's enough work at this front too. Product-side inference is more of a problem (especially videogen with fighting cats and food that has buttcheeks, that's the real power-hungry use case) but if we're thinking like Dario, it doesn't really matter except as a means to fund new hardware (actually not even that, mostly attract financing) to keep scaling until you have your God Machine. You will have it. It's not that far away. And the efficiency is going up all the time.

All of that AI power gap is noise. China will not have remotely the amount of chips to capitalize on their power advantage, not on any realistic timescale.

China wants energy for everything else. Logistics (from drones to cars and trucks to trains to ships), mineral and metal processing, fertilizers, fuel synthesis, connectivity, heating, cooling, laser welding and cutting and additive manufacturing, desalination, hydroponics, robots, simple household consumption, and of course making more energy generation systems — wind, solar, whatever. The list is near-infinite, for energy is a universal generic input. When Dario or his DoD/DoE curator asks his 9000 IQ God Machine for «how do I has Eagle Burger hegemony», it will answer «should have built more power plants and refineries, you stupid SF tech bro». There are fairly tenuous hopes that it is not so, but they amount either to crimes against humanity (bioweapons Dario is kvetching so much about, even willing to cooperate with evil China to avoid) or low-likelihood Wunderwaffes derived with pure reasoning and limited real-world experimentation loop (fusion, some nanotech woo, and cyberattacks by superhuman code agents… which if you think about it, are very likely to become obsolete in a world where the defender knows your plan, has near-parity technology and also root access to his system).

I don't think energy is very important as far as the AGI race is concerned. Energy and broadly heavy industry are crucial for what comes next.

11. WSJ (Group Score: 47.4 | Individual: 18.8)

Cluster: 3 tweets | Engagement: 54 (Avg: 183) | Type: Tech

ASML logged record quarterly orders of its semiconductor-making equipment and said it expects healthy sales growth this year, a sign that client spending to produce advanced chips behind the AI boom remains strong despite fears of a market bubble https://t.co/iZdF3XGlCf

See 2 related tweets

@Reuters: Top computer chip maker ASML has raised its outlook for the year as demand surged from its AI-focuse...
@ReutersBiz: Top computer chip maker ASML has raised its outlook for the year as demand surged from its AI-focuse...

12. Reuters (Group Score: 47.2 | Individual: 16.0)

Cluster: 3 tweets | Engagement: 67 (Avg: 66) | Type: Tech

Exclusive: China has approved its first batch of Nvidia's H200 artificial intelligence chips for import, marking a shift in position as China seeks to balance its AI needs against spurring domestic development https://t.co/5hjIppO7Q5

See 2 related tweets

@WSJ: China has approved purchases of Nvidia’s popular H200 artificial-intelligence chip for the first tim...
@Reuters: Exclusive: China has approved its first batch of Nvidia's H200 artificial intelligence chips for imp...

13. Forbes (Group Score: 46.6 | Individual: 20.7)

Cluster: 3 tweets | Engagement: 42 (Avg: 73) | Type: Tech

There’s a new competitor in the war for robotaxi supremacy and it’s an unlikely one: Waabi, a Canadian self-driving truck startup created by AI scientist Raquel Urtasan.

The company says it’s raised up to $1 billion from venture firms and Uber to help commercialize its robotic big rig business and fund a surprise foray into the robotaxi market currently led by Alphabet’s Waymo.

14. vllm_project (Group Score: 45.5 | Individual: 17.8)

Cluster: 3 tweets | Engagement: 136 (Avg: 155) | Type: Tech

🎉 Congrats to @arcee_ai on releasing Trinity Large — with day-0 support in vLLM!

A 398B sparse MoE with ~13B active params, trained on 17T+ tokens — delivering frontier-level quality with efficient compute.

You can serve it now with vLLM 👇

Thanks to the Arcee AI and vLLM community for the collaboration!

See 2 related tweets

@datologyai: Proud to power @arcee_ai’s Trinity Large.

A 400B/A13B MoE model trained on 17T tokens of our curate...

@PrimeIntellect: RT @samsja19: Today we’re releasing Trinity Large, a 400B MoE LLM with 13B active parameters, traine...

15. adcock_brett (Group Score: 44.6 | Individual: 32.3)

Cluster: 2 tweets | Engagement: 2426 (Avg: 950) | Type: Tech

I want to be really clear - there is NO teleoperation in this video. No human controlling this

It's an AI model called Helix - and it controls the entire body (walking, touching, planning)

It’s the most significant release Figure has shown to date

Nothing but nets

See 1 related tweets

@chris_j_paxton: RT @adcock_brett: I want to be really clear - there is NO teleoperation in this video. No human cont...

16. arimorcos (Group Score: 44.1 | Individual: 26.6)

Cluster: 2 tweets | Engagement: 204 (Avg: 51) | Type: Tech

RT @arcee_ai: Today, we’re releasing the first weights from Trinity Large, our first frontier-scale model in the Trinity MoE family. https:…

See 1 related tweets

@_lewtun: RT @latkins: Today, we are releasing our first weights from Trinity-Large, our first frontier-scale ...

17. alexocheema (Group Score: 42.5 | Individual: 42.5)

Cluster: 1 tweets | Engagement: 4453 (Avg: 525) | Type: Tech

Running Kimi K2.5 on my desk.

Runs at 24 tok/sec with 2 x 512GB M3 Ultra Mac Studios connected with Thunderbolt 5 (RDMA) using @exolabs / MLX backend.

Yes, it can run clawdbot. https://t.co/ssbEeztz2V

18. amasad (Group Score: 42.1 | Individual: 42.1)

Cluster: 1 tweets | Engagement: 660 (Avg: 90) | Type: Tech

The founder of the super popular Saudi Sabq digital newspaper personally did an AI rebuild of the paper and launched on Replit!

19. VKazulkin (Group Score: 41.9 | Individual: 41.9)

Cluster: 1 tweets | Engagement: 9556 (Avg: 367) | 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…

20. mark_k (Group Score: 40.2 | Individual: 16.0)

Cluster: 3 tweets | Engagement: 10 (Avg: 166) | Type: Tech

Google Expands Affordable AI Plus to 35 New Countries Including U.S. The @GoogleAI Plus plan costs $7.99 monthly in the U.S., with a$ 3.99 intro for two months, offering boosted access to Gemini 3 Pro, Nano Banana Pro image generation, more Veo 3.1 video tools, expanded NotebookLM, and 200 GB storage.

Existing Google One Premium 2 TB users get these perks automatically at no extra charge.

Highlights include Agentic Vision in Gemini 3 Flash for vision tasks via Python code and sharper AI Overviews in Search using Gemini 3, alongside cloud credits for higher-tier subscribers that developers call a game-changer.

See 2 related tweets

@nummanali: This is mad

Google AI Ultra costs me $200 a month

I get an insane amount of features with that al...

@demishassabis: RT @GeminiApp: Today we’re launching Google AI Plus in 35 new countries and territories, making it a...