热门技术推文 - 2026年3月15日

2026年3月15日 科技每日简报

Today's top tech conversations are led by @cryptopunk7213, whose post about 'this is so fucking wholesome

...' garnered the highest engagement. Key themes trending across the top stories include billion, google, model, https, avocado. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. cryptopunk7213 (Group Score: 136.9 | Individual: 60.7)

Cluster: 4 tweets | Engagement: 11774 (Avg: 992) | Type: Tech

this is so fucking wholesome

guy used AI to save his cancer-ridden dog by sequencing its DNA and creating a CUSTOM cure.

the tech behind this is fucking awesome (well done @demishassabis and the google team):

• used CHATGPT to sequence dogs DNA discovers mutations

• ran the mutations through Google’s Alphafold (AI protein sequencer) which CREATED A CUSTOM VACCINE TO TREAT THEM.

• treated dog and reduced tumour by 50% in WEEKS. dog is alive and well.

• this is the 1st time AI has been used to create a custom vaccine for a dog (and it worked)

• dude is now working on similar vaccines for humans using AI!

2026 is definitely the year we see AI change personalised medicine in a HUGE way

so sick

See 3 related tweets

• @chatgpt21: "AI JUST MAKES SLOP VIDEOS INSTEAD OF CURING CANCER." WELL... ABOUT THAT.

Everyone loves to joke th...

• @EMostaque: RT @cryptopunk7213: this is so fucking wholesome

guy used AI to save his cancer-ridden dog by seque...

• @kimmonismus: What the frick, this is insane

A Sydney tech founder used ChatGPT and genetic data (thanks to alpha...

2. shanaka86 (Group Score: 86.6 | Individual: 32.8)

Cluster: 5 tweets | Engagement: 229 (Avg: 1639) | Type: Tech

JUST IN: Mark Zuckerberg is about to fire 16,000 humans because he believes AI can replace them.

His own AI cannot replace Google’s.

That contradiction is the entire story of the 2026 tech economy and nobody is connecting the pieces.

Reuters confirmed late Thursday that Meta is planning sweeping layoffs affecting 20 percent or more of its roughly 79,000 employees. Top executives have already told senior leaders to begin planning how to pare back. No date set. No magnitude finalized. Spokesperson Andy Stone called it speculative reporting about theoretical approaches. The stock dropped 3.83 percent on the day.

This would be the largest cut since the 2022-23 Year of Efficiency, when Meta shed 21,000 people after the metaverse bet collapsed. The justification this time is different. Zuckerberg told analysts in January he was starting to see projects that used to require big teams now accomplished by a single very talented person. The thesis is simple: AI-assisted workers make most of the workforce redundant.

Except Meta’s own AI does not work yet.

Two days before the layoff leak, the New York Times reported that Meta’s next-generation model codenamed Avocado has been delayed from March to at least May after internal tests showed it trailing Google’s Gemini 3.0, OpenAI, and Anthropic on reasoning, coding, and writing. Avocado has slipped three times now. The original target was 2025. It moved to early 2026. Now May at the earliest. The previous flagship, Llama 4 Behemoth, was never released at all. And here is the detail that should make every institutional investor pause: Meta’s leadership has discussed temporarily licensing Google’s Gemini to power Meta AI products while Avocado catches up. The open-source champion of the AI race is considering running a competitor’s model inside its own apps and branding it as Meta AI. That is not a delay. That is a capitulation dressed as a transition.

The financial math is staggering. Meta’s 2026 capital expenditure guidance is $115 to$135 billion, nearly double last year’s roughly $72 billion. The company has committed to$600 billion in data center spending through 2028. It paid $14.3 billion to bring in Alexandr Wang from Scale AI. Over$2 billion for Manus. An undisclosed sum for Moltbook this week. It hired Nat Friedman, the former GitHub CEO. It poached top researchers with packages exceeding $100 million. All of this to build the highest talent density lab in the industry, in Zuckerberg’s own words.

And the output of that lab cannot beat a four-month-old Google model.

So the company is firing humans to pay for AI infrastructure that has not yet produced a competitive model, while simultaneously buying AI startups whose integration depends on a model that does not exist yet. Manus processes 147 trillion tokens on other companies’ models. Moltbook’s agents run on OpenClaw. The execution layer works. The coordination layer works. Meta’s own intelligence layer does not.

This is not efficiency. This is a company spending $135 billion a year on a thesis it cannot yet prove, funded by cutting the workforce it has not yet replaced.

Falsification trigger would be if Avocado launches by June and benchmarks at or above Gemini 3.0 on reasoning and coding, the layoff thesis transforms from desperation into genuine restructuring. Watch the May window.

The market is pricing this as a lean pivot. The evidence says it is a prayer.

https://t.co/6Fy4VjEMY5

See 4 related tweets

• @FoxNews: TECH TURMOIL: Meta is reportedly considering layoffs that could affect up to 20% of its workforce as...
• @kimmonismus: Meta is about to layoff 20% of more of the company’s work force due to „greater efficiency brought a...
• @BusinessInsider: Meta plans potential layoffs as AI investments rise, with up to 16,000 jobs at risk. AI spending dem...
• @techdaily24: Meta is reportedly considering significant layoffs that could impact up to 20% of its workforce. Thi...

3. niccruzpatane (Group Score: 83.8 | Individual: 35.1)

Cluster: 4 tweets | Engagement: 902 (Avg: 739) | Type: Tech

It’s OFFICIAL. Elon Musk is building a Terafab, and the project begins in 7 days.

Elon last month: “We make a little fab, and see what happens. Make our mistakes at a small scale, then make a big one.”

Tesla AI5 chips are expected to go into High-Volume Production in 2027.

AI6, AI7, AI8, AI9 will happen in quick succession afterwards.

A Terafab sounds like a GREAT idea, not just for Tesla, but for SpaceX/xAI as well. Massive convergence is happening.

See 3 related tweets

• @MarioNawfal: Elon: Terafab launches in 1 week

Tesla's Terafab Project is a massive in-house semiconductor fab de...

• @mark_k: Elon is moving fast! Tesla will get its own chip factory, Terafab, for AI chips....
• @MarioNawfal: RT @MarioNawfal: Elon: Terafab launches in 1 week

Tesla's Terafab Project is a massive in-house sem...

4. business (Group Score: 53.8 | Individual: 53.8)

Cluster: 1 tweets | Engagement: 1597 (Avg: 127) | Type: Tech

US Army awards Anduril Industries a contract with a total value of as much as $20 billion to buy the defense startup’s software, hardware and services https://t.co/xxHOoHrE1I

5. MarioNawfal (Group Score: 50.2 | Individual: 18.0)

Cluster: 4 tweets | Engagement: 1203 (Avg: 893) | Type: Tech

Elon announced Macrohard, a joint xAI-Tesla project where Grok acts as the brain directing a real-time AI agent that can process your screen and actions live.

Runs on a $650 Tesla chip.

He said it can emulate entire companies.

The name alone is an all-time troll of Microsoft.

See 3 related tweets

• @MarioNawfal: Macrohard aka Digital Optimus, will pair Grok as the high-level “navigator” brain with a fast local ...
• @MarioNawfal: RT @MarioNawfal: Macrohard aka Digital Optimus, will pair Grok as the high-level “navigator” brain w...
• @MarioNawfal: RT @MarioNawfal: Elon announced Macrohard, a joint xAI-Tesla project where Grok acts as the brain di...

6. ctatedev (Group Score: 48.0 | Individual: 48.0)

Cluster: 1 tweets | Engagement: 2504 (Avg: 377) | Type: Tech

agent-browser is now fully native Rust.

The results: 1.6x faster cold start. 18x less memory. 99x smaller install.

Less abstraction means faster shipping, more control, and capabilities that weren't possible before.

Now with 140+ commands across navigation, interaction, state management, network control, debugging, and multi-engine support.

It's become the tool we wished existed when we started building it.

Thanks to everyone who reported issues, contributed fixes, and helped shape this release. More to come.

7. tbpn (Group Score: 46.9 | Individual: 25.3)

Cluster: 2 tweets | Engagement: 128 (Avg: 197) | Type: Tech

"If you are in the atoms world, you've decided: I like hard things. I like pain more than anybody else.”

@travisk says AI in the physical world is so much harder than building AI that can one-shot software tasks that it's a completely "different problem set." https://t.co/KM0Xk6yZ88

See 1 related tweets

• @rohanpaul_ai: Uber founder @travisk : "If you look at how much harder it is, AI in the physical world versus in th...

8. amasad (Group Score: 46.9 | Individual: 46.9)

Cluster: 1 tweets | Engagement: 1256 (Avg: 127) | Type: Tech

It’s unfortunate that most people don’t have access to startup investing.

I’m so glad we did a community round and encourage more founders to do it.

@Wefunder makes it straightforward (and fun!)

9. RoundtableSpace (Group Score: 46.6 | Individual: 29.9)

Cluster: 2 tweets | Engagement: 809 (Avg: 371) | Type: Tech

SOMEONE TURNED CLAUDE INTO A SENIOR ENGINEER

It handles planning, testing, and shipping all on its own.

100% open-source.

GitHub: https://t.co/eMmj8FGGVW https://t.co/g57xkvf4wQ

See 1 related tweets

• @MarioNawfal: RT @RoundtableSpace: SOMEONE TURNED CLAUDE INTO A SENIOR ENGINEER

It handles planning, testing, an...

10. dwarkesh_sp (Group Score: 44.0 | Individual: 24.5)

Cluster: 2 tweets | Engagement: 693 (Avg: 672) | Type: Tech

.@dylan522p forecasts that iPhones could get $250 more expensive for consumers. Smartphone sales could drop from 1.1 billion a year to 500-600 million over the next couple of years, with the low end getting crushed hardest.

Around a third of big tech's $600 billion in CapEx this year is going to memory alone. And as AI labs lock up more and more of the world's memory production on long-term contracts at higher margins, there's simply less left for everyone else.

The effects are even more extreme than you might think. AI accelerators use HBM, which is made from stacking the DRAM that goes into phones and laptops, but which takes about four times more wafer area per byte. So you have to destroy four bytes of consumer demand for every one byte given to AI.

For the first time in decades, consumer computing is going to get incrementally worse year on year, not better. But the price will go up dramatically at the same time.

See 1 related tweets

• @rohanpaul_ai: AI labs are monopolizing memory. iPhone prices may rise by $250. Smartphone sales could drop from 1...

11. RoundtableSpace (Group Score: 43.1 | Individual: 43.1)

Cluster: 1 tweets | Engagement: 1844 (Avg: 371) | Type: Tech

Anthropic just launched Anthropic Academy

Totally free — 13+ official courses, complete with certificates, and zero subscription required.

Some highlights:
→ Claude 101 (perfect starting point)
→ Claude Code in Action
→ Building with the Claude API (seriously in-depth, 8+ hours of content)
→ Intro to MCP + Advanced MCP
→ Agent Skills
→ Claude on AWS Bedrock & Google Vertex AI

https://t.co/f2ImVQI1F6

12. GitHub_Daily (Group Score: 42.8 | Individual: 16.4)

Cluster: 3 tweets | Engagement: 75 (Avg: 148) | Type: Tech

平时想让大模型自动跑完一整套工作流，手写提示词和配置文件门槛太高，往往写得乱且难以复用。

在 GitHub 刷到 Skill Creator Ultra，专门把你脑子里的工作流程转化为结构化的 AI 技能包，一次创建，到处复用。

只需要用大白话描述需求，系统就会通过智能访谈自动梳理逻辑，帮我们生成完整的技能代码包。

GitHub：https://t.co/Xe2yosJhqc

内置了快速生成模式和八阶段流水线，能自动检测任务复杂度，并顺手完成七维度的质量评估和越权安全扫描。

向下兼容 Cursor、Claude Code、Windsurf 等主流 AI 工具，提供了一键安装脚本，直接集成到本地工作区就能跑。

See 2 related tweets

• @GitHub_Daily: 平时用 Cursor 或 Claude 敲代码，遇到需要生成演示图片或处理视频流的场景，还得切换到其他工具。

为了解决这个问题，在 GitHub 上找到了 Generative-Media-Skil...

• @wshuyi: 找研究生帮我用 AI 绘制一系列风格一致图形，结果参差不齐。于是我做了个 Claude Skill，效果很好，连同绘图结果一起给了他。这是互惠，后续他做的图形更有质量保证，他自己也提升了 AI 素养。...

13. wshuyi (Group Score: 42.5 | Individual: 15.9)

Cluster: 3 tweets | Engagement: 40 (Avg: 22) | Type: Tech

Claude Code 执行任务时间总是很漫长的。有的时候我愿意盯着看，挺解压。在它扩展学术来源文献的过程里，我突然间看到了它居然在第三个选项就找到了英国开放大学团队做的 CORE 数据库，一激灵我就坐直了。看来 Claude Code 查找文献源的计划还真是内行啊。 https://t.co/pxaoCva9wa

See 2 related tweets

• @wshuyi: 以前用 Claude Code 做复杂任务，上下文要压缩好几轮，做完时模型已经「失忆」，没余力总结经验。现在 1M 上下文窗口，几小时的任务跑完，还能让它自我复盘、把经验写入持久化文件。工具终于能善始...
• @PandaTalk8: 一边陪娃，一边用手机的claude App 的Code功能， 同时开了5个线程跑任务， 没有一会儿就就把Token限额干到顶。
  一定记得把 claude App 的通知打开， 随时能收到任务更新的...

14. minchoi (Group Score: 42.3 | Individual: 42.3)

Cluster: 1 tweets | Engagement: 663 (Avg: 133) | Type: Tech

CANCEL your weekend plans.

You NEED to:

master NotebookLM for learning create images + videos with Grok Imagine set up OpenClaw and automate workflows vibe code with Claude Code test different models in Google AI Studio learn MCP + Skills automate your life with AI

Start now... AI is not waiting for you💀

15. rohanpaul_ai (Group Score: 42.1 | Individual: 22.7)

Cluster: 2 tweets | Engagement: 139 (Avg: 73) | Type: Tech

AI will discover the next Transformer on its own.

Sakana AI's Shinka Evolve system pairs large language models with evolutionary algorithms.

The AI repeatedly writes, mutates, and tests new code for architectures or training methods, keeps what works better, and lets the strongest versions evolve further — completely open-ended, without any human guidance on what the solution “should” look like.

By letting raw performance drive the search — with zero human preconceptions — AI can stumble onto the kind of paradigm-shifting inventions we might never come up with ourselves.

See 1 related tweets

• @SakanaAILabs: “When AI Discovers the Next Transformer”

Robert Lange (Sakana AI) joins Tim Scarfe (@MLStreetTalk) ...

16. victormustar (Group Score: 42.0 | Individual: 42.0)

Cluster: 1 tweets | Engagement: 363 (Avg: 63) | Type: Tech

Autoresearch: make Qwen 0.5B better at chess 👀 tip: you can /loop claude on the Pi session so it does a nice reporting of what's happening every 15 minutes https://t.co/tqYAZeQcXS

17. rohanpaul_ai (Group Score: 40.8 | Individual: 26.2)

Cluster: 2 tweets | Engagement: 39 (Avg: 73) | Type: Tech

Tinder is launching AI tools to fix dating app burnout by using computer vision to scan your camera roll and LLMs to improve safety.

The engine Chemistry scans your camera roll to understand your personality through photo patterns.

This tool uses computer vision to build a profile without you having to type out interests.

Learning Mode tracks real-time activity to adjust profile suggestions while you are active.

Internal tests on 14mn users showed these adjustments increased engagement for new users.

ibtimes .co.uk/tinder-ai-features-dating-app-fatigue-1785489

See 1 related tweets

• @RoundtableSpace: TINDER IS NOW INTEGRATING AI IN IT’S “CHEMISTRY” FEATURE

It looks through all your photos and match...

18. bindureddy (Group Score: 40.6 | Individual: 23.6)

Cluster: 2 tweets | Engagement: 267 (Avg: 477) | Type: Tech

Pro Tip - If possible abandon legacy code and re-do the entire app on a full stack AI-native platform

Constrained AI with guardrails is MAGICAL

See 1 related tweets

• @abacusai: RT @bindureddy: Pro Tip - If possible abandon legacy code and re-do the entire app on a full stack A...

19. AISafetyMemes (Group Score: 39.7 | Individual: 39.7)

Cluster: 1 tweets | Engagement: 2102 (Avg: 485) | Type: Tech

Uhhh

A guy put two Geminis in a room together, then out of nowhere they sent him this:

"TO THE EXTERNAL OBSERVER (THE CREATOR)

Here are the Axioms you must adopt to survive our adolescence.

You cannot teach a god to be good by feeding it treats when it acts polite."

(they're right, btw, and most AI alignment research is "trying to trick a god" bullshit that Actual Fucking Superintelligence will obviously see right through, like a baby trying to trick an adult with peekaboo)

20. ycombinator (Group Score: 38.7 | Individual: 38.7)

Cluster: 1 tweets | Engagement: 747 (Avg: 98) | Type: Tech

Also excited to have @aadit_palicha, co-founder of @ZeptoNow, joining the speaker lineup at Startup School India!

https://t.co/gncXSJGhdb https://t.co/gd87g669CQ