科技推特精选 - 2026年2月10日

今日科技要闻：随着 FDA 报告多起与人工智能医疗设备相关的重大召回，科技行业正处于创新与安全的平衡博弈之中。尽管代理式 AI (Agentic AI) 的突破及 Stilta 等专业工具推进了自动化进程，但 Snowflake 警告称，多数企业尚未为正式投产做好准备。虽然 AI 对法律服务的影响引发了持续的市场波动，但 Palantir 市值的 3240 亿美元飙升以及 Workday 的领导层变动，凸显了持续的工程研究和初创公司的果断转型依然是增长的主要驱动力。

1. javarevisited (Group Score: 87.8 | Individual: 46.3)

Cluster: 2 tweets | Engagement: 2003 (Avg: 73) | Type: Tech

Software Engineering is the only industry where 5 days of study is more important than 5 years of experience

See 1 related tweets

@javarevisited: RT @javarevisited: Software engineering is also the only career where you couldn't get the job after...

2. aakashgupta (Group Score: 67.9 | Individual: 67.9)

Cluster: 1 tweets | Engagement: 12829 (Avg: 416) | Type: Tech

If you pitched this as a screenplay, every studio would reject it for being too unrealistic.

A 28-year-old Vietnamese developer living with his parents programs a game over a single holiday weekend in his bedroom. He uses free ad monetization because he hopes to make a few hundred dollars a month. The game sits untouched in the App Store for eight months.

Then something inexplicable happens. With zero marketing spend, the game goes viral. Within weeks it tops the charts in 102 countries, hits 50 million downloads, and starts generating $50,000 per day in pure ad revenue. He’s making$ 18 million annualized from a game he built in three days.

So what does he do? He kills it. Not because Apple forced him. Not because Nintendo sued him. Not because a competitor acquired him. He kills it because he felt guilty that people were too addicted. The guilt was ruining his sleep. He tweeted “I cannot take this anymore” and pulled it the next day.

Within hours, people listed phones with the game on eBay for $99,900. Fans sent him death threats demanding he put it back. His only response was “And I still make games.”

The part nobody talks about: he turned down every acquisition offer afterward. A game generating $1.5 million per month, and a solo developer in Hanoi said no to everyone because he refused to compromise his independence.

Every founder in Silicon Valley talks about “mission over money.” Dong Nguyen actually did it, and the internet tried to destroy him for it.

3. Reuters (Group Score: 66.3 | Individual: 21.7)

Cluster: 4 tweets | Engagement: 106 (Avg: 98) | Type: Tech

Researchers found 60 FDA-authorized medical devices using AI were linked to 182 product recalls, a letter in the JAMA Health Forum shows. Reuters documents the hazards as device makers, tech giants and software developers race to roll out AI in medicine https://t.co/FVKgs73gi5

See 3 related tweets

@Reuters: The AI boom poses a problem for the FDA, current and former agency scientists told Reuters: The agen...
@Reuters: AI is beginning to transform health care. A Reuters investigation shows some of the hazards as devic...
@Reuters: AI made its way into medicine long before chatbots like ChatGPT appeared — the FDA authorized its fi...

4. Reuters (Group Score: 62.4 | Individual: 43.6)

Cluster: 2 tweets | Engagement: 848 (Avg: 98) | Type: Tech

The FDA received unconfirmed reports of at least 100 malfunctions and adverse events after AI was added to a medical device used to treat chronic sinusitis. In one case, a surgeon mistakenly punctured the base of a patient’s skull https://t.co/Dvb8BSrLgf @specialreports https://t.co/m5V9GfLw4t

See 1 related tweets

@Reuters: After artificial intelligence was added to a medical device used to treat chronic sinusitis, the FDA...

5. gdgtify (Group Score: 56.8 | Individual: 28.8)

Cluster: 2 tweets | Engagement: 28 (Avg: 56) | Type: Tech

I asked Nano Banana to visualize the history of computing.

 Input Variable:   The history of computing

  System Instruction:  
Generate a hyper-realistic, 3D "Kinetic Museum Sculpture."
  Structure:   A continuous, twisting   Figure-8 (Infinity Symbol) Track   suspended in mid-air by a central mechanical pedestal.

  1.   Era Deconstruction:  
      Analyze the Input:   Break the history into   4 Distinct Stages  .
         Stage 1 (The Origin):  Ancient/Natural.
         Stage 2 (The Expansion):  Medieval/Trade.
         Stage 3 (The Revolution):  Industrial/Mechanical.
         Stage 4 (The Horizon):  Future/Digital.

  2.   Container  
      The Rails:   The timeline is a   Translucent Glowing Tube   or   Conveyor Belt  .
         Inside the Tube:  A flowing substance representing the topic (e.g., Coins flowing for Currency; Data stream for AI).
      The Base:   A heavy   Mahogany and Brass Gearbox   sits at the bottom, physically driving the loop.

  3. The Platforms :  
      Placement:   4 distinct "Islands" or platforms sit on top of the loop at the curves.
      Procedural Population:  
         Island 1 (Origin):  Primitive figures using simple tools. Earthy textures.
         Island 2 (Expansion):  Stone architecture, merchants, horses.
         Island 3 (Revolution):  Smokestacks, gears, steam trains. Grey/Rust palette.
         Island 4 (Horizon):  Holograms, chrome robots, floating cars. Neon palette.

  4. Visual Syntax:  
      Connectivity:   The figures are traversing the loop. (e.g., A horse cart on the left turns into a Model T Ford in the middle, turning into a hover-car on the right).
      Lighting:   The loop itself emits light, changing color as it progresses through time (Amber -> White -> Blue).

  Output:   ONE image, 1:1 Aspect Ratio, Studio Lighting, "Installation Art" Aesthetic, 8k.

See 1 related tweets

@gdgtify: I love visualizing inventions with Nano Banana. Here is the prompt:

Do this for Wright Brothers: ...

6. QodoAI (Group Score: 56.3 | Individual: 56.3)

Cluster: 1 tweets | Engagement: 765 (Avg: 4) | Type: Tech

Analogue 3D - Prototype Limited Editions

Available now. Shipping in 24-48hrs.

A reimagining of the N64. In 4K resolution. https://t.co/GZii0UD3iX

7. Reuters (Group Score: 53.7 | Individual: 27.2)

Cluster: 2 tweets | Engagement: 86 (Avg: 98) | Type: Tech

Novo Nordisk shares jumped more than 8% after Hims & Hers abandoned the launch of their $49 weight-loss pill. The company withdrew the product following legal threats from Novo Nordisk and the FDA https://t.co/8SZIRX8cQd https://t.co/zfoz1YVvOP

See 1 related tweets

@ReutersBiz: Novo Nordisk shares jumped more than 8% after Hims & Hers abandoned the launch of their $49 weig...

8. Snowflake (Group Score: 45.1 | Individual: 28.8)

Cluster: 2 tweets | Engagement: 14 (Avg: 22) | Type: Tech

"95% of enterprises aren't actually ready for agentic AI." Ouch. 😬

We sat down with 8 VCs to find out why most "agents" are just LLMs on a loop and how the real winners are building for production in 2026.

Read the latest in our new report - “Startup 2026: AI Agents Mean Business” https://t.co/H3mRByYGn8

See 1 related tweets

@ethereum: RT @Optimism: AI agents are about to manage trillions in assets.

But an agent built by one company ...

9. aakashgupta (Group Score: 42.6 | Individual: 42.6)

Cluster: 1 tweets | Engagement: 1188 (Avg: 416) | Type: Tech

Alex Karp received a $12,000 inheritance from his grandfather. His goal was to save$ 250,000 so he could move to Berlin and read books for a living.

Today Palantir has a market cap of $324 billion. The stock is up over 1,300% from its first-day close of$ 9.50 in September 2020. Karp was the highest-paid CEO of a publicly traded company in the United States in 2024, with compensation calculated at $6.8 billion.

His credentials for running a defense tech company? A philosophy degree from Haverford. A JD from Stanford he called “the worst three years of my adult life.” A PhD from Goethe University Frankfurt where he wrote his dissertation in German on aggression, jargon, and culture through Parsonian social theory. His first job after graduating was research associate at the Sigmund Freud Institute.

Every other CEO who built a company this large in defense came through engineering, military service, or finance. Karp came through continental philosophy and psychoanalysis. And he built the platform the CIA, NSA, FBI, U.S. Army, and Special Operations Command all depend on. The company recently landed a $10 billion Army deal.

Q4 2025: $1.407 billion in revenue, up 70% year over year. U.S. commercial revenue surged 137%. Full year 2025 hit$ 4.475 billion. They’re guiding $7.2 billion for 2026. Rule of 40 score: 127%. They do all of this with 4,414 employees, roughly$ 1 million in revenue per head.

He delivers earnings calls from New Hampshire in ski gear. Keeps Tai Chi swords in his offices. Practices qigong meditation. In December 2025, he paid $120 million for a 3,700-acre former Trappist monastery outside Aspen. He calls himself a socialist. He voted for Hillary Clinton. And he tells the New York Times the West needs autonomous weapons to survive a three-front war against China, Russia, and Iran.

He hires philosophers, physicists, and historians alongside engineers. He moved the company out of Silicon Valley because he grew disgruntled with its monoculture. He wrote in Palantir’s IPO prospectus that the engineering elite “do not know more about how society should be organized or what justice requires.”

His PhD was about how language, aggression, and culture interact in systems. Palantir does the same thing with data. The philosophy is the entire foundation.

$12,000 inheritance to the 42nd most valuable company on earth. The most unconventional path to building a$ 324 billion company in American history.

10. business (Group Score: 42.2 | Individual: 18.4)

Cluster: 3 tweets | Engagement: 24 (Avg: 66) | Type: Tech

Workday announced co-founder Aneel Bhusri is returning to head the software company, replacing chief executive officer Carl Eschenbach after the company’s shares have plummeted over the past year https://t.co/j8jQXcB0lD

See 2 related tweets

@TechCrunch: Workday CEO Eschenbach departs, with co-founder Aneel Bhusri returning as CEO https://t.co/YZVsneBh...
@BusinessInsider: Cofounder Aneel Bhusri returns as CEO of the software company, which recently laid off about 400 emp...

11. AdinaYakup (Group Score: 38.8 | Individual: 22.4)

Cluster: 2 tweets | Engagement: 35 (Avg: 216) | Type: Tech

New release from GAIR: daVinci-Agency 🔥 A long-horizon agent finetuned on GLM-4.6, trained on real GitHub chain of PRs. Model: https://t.co/MV2Xj7bxQa Paper: https://t.co/U6l6ECP9OU

See 1 related tweets

@HuggingPapers: RT @AdinaYakup: New release from GAIR: daVinci-Agency 🔥 A long-horizon agent finetuned on GLM-4.6, t...

12. ReutersBiz (Group Score: 38.3 | Individual: 19.4)

Cluster: 2 tweets | Engagement: 33 (Avg: 13) | Type: Tech

Global markets saw a wild swing in the first week of February as nerves about AI replacing legal software sent tech firms and chipmakers sliding. Listen to experts on Market Talk https://t.co/JwaoVoUWyn

See 1 related tweets

@Reuters: Global markets saw a wild swing in the first week of February as nerves about AI replacing legal sof...

13. dair_ai (Group Score: 37.7 | Individual: 37.7)

Cluster: 1 tweets | Engagement: 110 (Avg: 96) | Type: Tech

What if you could get multi-agent performance from a single model?

Multi-agent debate systems are powerful. Multiple LLMs can critique each other's reasoning, catch errors, and converge on better answers.

However, the cost scales linearly with the number of agents. Five agents means 5x the compute. Twenty agents means 20x and so on.

But the intelligence gained from debate doesn't have to stay locked behind a compute wall.

This new research introduces AgentArk, a framework that distills the reasoning capabilities of multi-agent debate into a single LLM through trajectory extraction and targeted fine-tuning.

This work addresses an important problem: multi-agent systems are effective but expensive at inference time. AgentArk moves that cost to training time, letting a single model carry the reasoning depth of an entire agent team.

The key idea: run multi-agent debate offline to generate high-quality reasoning traces, then train a smaller model to internalize those patterns.

Five agents debate, one student learns.

AgentArk tests three distillation methods. RSFT uses supervised fine-tuning on correct trajectories. DA filters for diverse reasoning paths. PAD, their strongest method, preserves the full structure of multi-agent deliberation, capturing how agents verify intermediate steps and localize errors.

The results across 120 experiments:

PAD achieves a 4.8% average gain over single-agent baselines, with in-domain improvements reaching up to 30%. On reasoning quality metrics,

PAD scores highest in intermediate verification (4.07 vs 2.41 baseline) and reasoning coherence (3.96 vs 1.88 baseline).

The distilled models also transfer: trained on math, they improve on TruthfulQA with ROUGE-L jumping from 0.613 to 0.657.

Scaling from Qwen3-32B teachers down to Qwen3-0.6B students, the framework holds up. Even sub-billion parameter models absorb meaningful reasoning improvements from multi-agent debate.

Paper: https://t.co/cyPTig221s

Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c

14. ednewtonrex (Group Score: 37.1 | Individual: 37.1)

Cluster: 1 tweets | Engagement: 1431 (Avg: 1101) | Type: Tech

It is crazy that massive book piracy was so critical to the early days of both OpenAI and Anthropic.

In 2025, Anthropic co-founder Ben Mann testified that he himself had downloaded the massive pirate library LibGen at Anthropic. He did so only a few months after the company was founded in 2021.

He also downloaded another pirate library, Books3, the very month Anthropic was founded. It was used to train Claude 1 & Claude 2.

He also said he had previously downloaded LibGen when he was at OpenAI. OpenAI used books from LibGen to train GPT-3 and GPT-3.5.

And he went further - he said he understood that LibGen had also been downloaded by Alec Radford at OpenAI. Radford led the development of GPT-1 and GPT-2.

Whatever today’s models were trained on, billions of dollars were raised off the back of models trained on pirated books, and those books were downloaded by some of the most important researchers at the companies in question.

The AI industry is built on piracy.

15. ycombinator (Group Score: 37.0 | Individual: 37.0)

Cluster: 1 tweets | Engagement: 719 (Avg: 184) | Type: Tech

Stilta (@StiltaIP) is building the Cursor for patent practitioners.

Unlike rigid or generic tools, Stilta uses agentic AI for deep patent research that's interactive and traceable.

Congrats @oskarblock, @petruswrn, @testreen & @oscaradamssons!

https://t.co/wlSEIbHtOp https://t.co/62YFmm4otI

16. llama_index (Group Score: 36.8 | Individual: 30.3)

Cluster: 2 tweets | Engagement: 74 (Avg: 47) | Type: Tech

Everybody’s talking about @openclaw, so @itsclelia decided to build her own crustacean AI assistant for document workflows: LobsterX 🦞

LobsterX is built on top of our Agent Workflows and leverages LlamaCloud for document parsing, structured data extraction, and classification via powerful, modular tools🛠️

Designed with a safety-first mindset, the agent runs on AgentFS (by @tursodatabase) to protect your real filesystem and intentionally avoids full shell access to prevent security-critical or destructive operations 🔒 It’s fully self-hostable, can be run as a uv tool or @Docker container, and works out of the box as a @telegram bot 💬

📦 Install: 𝘶𝘷 𝘵𝘰𝘰𝘭 𝘪𝘯𝘴𝘵𝘢𝘭𝘭 𝘭𝘰𝘣𝘴𝘵𝘦𝘳𝘹 --𝘱𝘳𝘦𝘳𝘦𝘭𝘦𝘢𝘴𝘦=𝘢𝘭𝘭𝘰𝘸 📚 Read more: https://t.co/DrAkucc0B2 🦙 Sign up to LlamaCloud: https://t.co/wZjhFV29gN

See 1 related tweets

@jerryjliu0: RT @llama_index: Everybody’s talking about @openclaw, so @itsclelia decided to build her own crustac...

17. rohanpaul_ai (Group Score: 36.3 | Individual: 36.3)

Cluster: 1 tweets | Engagement: 764 (Avg: 94) | Type: Tech

RT @rohanpaul_ai: Marc Andreessen explains future belongs to generalist in the AI era.🎯

Founders will need skills across 6–8 fields. Deep…

18. webflow (Group Score: 35.2 | Individual: 35.2)

Cluster: 1 tweets | Engagement: 300 (Avg: 47) | Type: Tech

We partnered with Anthropic so you can use the Webflow connector directly in @claudeai 🙌🏽

Instead of stopping at ideas or recommendations, Claude can now take real action in Webflow.

Here are some of the things you can do with Claude + Webflow:

Bulk update CMS content across collections
Run SEO, content, and usability audits (and apply the fixes)
Clean up design systems with consistent variables, classes, and naming

👉 Learn more: https://t.co/IVDiG9nP21

19. jasonlk (Group Score: 34.9 | Individual: 34.9)

Cluster: 1 tweets | Engagement: 176 (Avg: 148) | Type: Tech

SaaS isn’t dead, even with the massive sell off this year (and really since July 2025)

But … what is dead is the classic pattern and way to scale:

1, maybe 2 players, take over a space by $20m ARR
Massive sales & marketing spend gets you to $100m+ ARR, often takes years
Product then mostly stagnant, quarterly releases, big platform advance only every 5 years or so
High NRR of 130%+ in enterprise, 110%+ in SMBs basically on autopilot
Customers happily renew and keep spending more with you

That is dead

It’s been replaced with:

Prototype built in days
AI creating 100s of new competitors, many dramatically better at agentic functionality
Outliers get to $100m in 12 months, become budget magnets
AI radically expanding functionality, obliterating prior pace software development
Massive reluctance to buy more products or seats from slow moving vendors they already pay for
High reluctance to contribute any budget that doesn’t show massive ROI almost immediately (from AI)
Endless price increases since 2022 leading to frustration and even anger at existing leaders and brands

That is "SaaS" today. If you want to win.

20. Forbes (Group Score: 34.8 | Individual: 34.8)

Cluster: 1 tweets | Engagement: 65 (Avg: 138) | Type: Tech

ChatGPU | A GPU chip that trained an early version of ChatGPT. “It looks so old fashioned and kind of crappy and tiny,” Sam Altman says now. After the hardware became too old to use, he gave the chips to all the employees who worked on the project.

FULL STORY: Sam Altman Reveals Why OpenAI Is Poised To Make The Biggest Business Bets Ever: https://t.co/4CNTnfMosn (Photo: Cody Pickens for Forbes) #Forbes250