科技推文日报 - 2026-03-30

2026年3月30日科技每日简报

Today's top tech conversations are led by @PandaTalk8, whose post about 'RT @_chenglou: My dear front-e...' garnered the highest engagement. Key themes trending across the top stories include https, first, agent, apple, models. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. PandaTalk8 (Group Score: 104.1 | Individual: 44.8)

Cluster: 5 tweets | Engagement: 12728 (Avg: 165) | Type: Tech

RT @_chenglou: My dear front-end developers (and anyone who’s interested in the future of interfaces):

I have crawled through depths of hell to bring you, for the foreseeable years, one of the more important foundational pieces of UI engineering (if not in implementation then certainly at least in concept): Fast, accurate and comprehensive userland text measurement algorithm in pure TypeScript, usable for laying out entire web pages without CSS, bypassing DOM measurements and reflow

See 4 related tweets

@RoundtableSpace: RIP CSS

https://t.co/BCCayMxQQn\n\nQT @_chenglou: My dear front-end developers (and anyone who’s in...

@arvidkahl: why is this\n\nQT @_chenglou: My dear front-end developers (and anyone who’s interested in the futur...
@xicilion: 这个好，今天来看看能不能用在 drawio 和 planuml 渲染上\n\nQT @_chenglou: My dear front-end developers (and anyone who’s...
@seconds_0: actual wizardry\n\nQT @_chenglou: My dear front-end developers (and anyone who’s interested in the f...

2. chhddavid (Group Score: 98.5 | Individual: 34.1)

Cluster: 3 tweets | Engagement: 27 (Avg: 22) | Type: Tech

Introducing Shipper

Claude Opus 4.6 can now build a business autonomously.

1️⃣ send a prompt in @shipper_now 2️⃣ claude designs, codes, launches, monetizes, translates, sends emails 3️⃣ you touch grass and make $$$

Done. Your Mac is now your co-founder. https://t.co/Rv2VZKC1vs

See 2 related tweets

@chddaniel: THIS IS ACTUALLY SCARY.\n\nQT @chhddavid: Introducing Shipper

Claude Opus 4.6 can now build a busin...

@chddaniel: this is actually frightening...\n\nQT @chhddavid: Introducing Shipper

Claude Opus 4.6 can now build...

3. business (Group Score: 91.5 | Individual: 20.8)

Cluster: 6 tweets | Engagement: 192 (Avg: 142) | Type: Tech

Eli Lilly signed an AI-powered drug development deal with Insilico Medicine that could be worth up to $2.75 billion. https://t.co/Ka90EfKnsf

See 5 related tweets

@CNBC: Eli Lilly reaches $2.75 billion deal with Insilico to bring AI-developed drugs to the global market ...
@Reuters: Eli Lilly to sign $2 billion deal for AI drug development with Hong Kong's Insilico Medicine, FT say...
@FT: US drugmaker Eli Lilly will sign a $2bn deal with a Hong Kong-listed company that uses AI for drug d...
@FT: Eli Lilly to sign $2bn deal for AI drug development with Hong Kong biotech https://t.co/G4038sb8eU...
@FirstSquawk: ELI LILLY TO ENTER $2B AGREEMENT FOR AI-DRIVEN DRUG DISCOVERY WITH HONG KONG BIOTECH FIRM – FT...

4. chhddavid (Group Score: 88.8 | Individual: 31.0)

Cluster: 3 tweets | Engagement: 8 (Avg: 22) | Type: Tech

We killed Claude Cowork.

Shipper builds an entire business from a single prompt... design, code, email marketing, launch, translation. Everything. Instantly.

All you have to do is talk to AI. It's live today. 🚢

https://t.co/RA080758o8 https://t.co/HRSzQvaIZD

See 2 related tweets

@chhddavid: 🚨 We just killed @cursor_ai.

Shipper builds an entire business from a single prompt... design, code...

@chddaniel: We might've killed Lovable.

Shipper builds an entire business from a single prompt... design, code,...

5. cryptopunk7213 (Group Score: 85.0 | Individual: 26.7)

Cluster: 4 tweets | Engagement: 307 (Avg: 1308) | Type: Tech

Apple really nailed AI by doing fucking nothing lol.

$135 billion in the bank. stole google’s model for a measly$ 1B, now forcing competitors to plug their models into siri if they want access to 2.5B apple users

patience (or laziness) paid off massively https://t.co/SSuQzD3Zeg\n\nQT @cryptopunk7213: i find it fucking hilarious how Apple "failing" at AI is now the exact reason they're about to win it:

watched everyone else burn $1.4T+ building models... then picked the winner (gemini) to use for...$ 1B
while everyone fights to grow users, apple flips a switch and 2.5 billion devices get AI siri tmrw.
$150B to splurge on the device / app layer. zero competition (because everyones spent their cash).
while openAI charges $200/mo subscriptions, Apple lets you run models on-device (cheaper, faster, private, personal)
while openAI struggles to build an AI device, Apple just dropped 5 powered by the best AI chips for hand-held devices.

they "lost" the model race because they didn't need to win it in the first place

greatest to (accidentally) ever do it.

See 3 related tweets

@business: Apple’s revamped AI and Siri strategy shows the company is recommitting to its core business model: ...
@business: Apple’s new AI strategy shows that it’s doubling down on being a hardware and services company, @mar...
@seraleev: Interesting strategy:

Apple lets you plug in third-party AI (Gemini, Claude, etc.) Users choose...

6. amasad (Group Score: 77.8 | Individual: 41.9)

Cluster: 3 tweets | Engagement: 411 (Avg: 121) | Type: Tech

Wow — $8m ARR.

Jon (@cheneypiano), one of the earliest Replit vibecoding success stories, is on his way to making millions in recurring revenue.

Started the business in a week and $400 barely a year ago.\n\nQT @mhp_guy: This guy vibecoded an app that landed him a$ 15,000 customer almost Instantly.

A year later, he's made $2.5M This year? He's projected to make$ 7M

Guys, please listen carefully to me with this:

He is not a software developer
He started this with $400

If you are watching this, you are the minority.

Stop assuming other people know what you know about AI. You are a first mover.

This is 100% possible for anyone else to do.

You've gotta check this one out!

See 2 related tweets

@amasad: RT @mhp_guy: This guy vibecoded an app that landed him a $15,000 customer almost Instantly.

A year...

@Replit: RT @amasad: Wow — $8m ARR.

Jon (@cheneypiano), one of the earliest Replit vibecoding success stori...

7. danshipper (Group Score: 77.3 | Individual: 30.1)

Cluster: 4 tweets | Engagement: 218 (Avg: 38) | Type: Tech

pirates, architects, hot people, grown ups\n\nQT @chintanzalani: The only 4 jobs that will remain at tech companies.

Credits: @yrechtman https://t.co/GsB7kIHqAt

See 3 related tweets

@clairevo: go heads down and grind and you can be all 4\n\nQT @chintanzalani: The only 4 jobs that will remain ...
@hammer_mt: RT @chintanzalani: The only 4 jobs that will remain at tech companies.

Credits: @yrechtman https://...

@kylebrussell: My niche is the intersection between slop cannon and hot people\n\nQT @chintanzalani: The only 4 job...

8. badlogicgames (Group Score: 69.2 | Individual: 52.7)

Cluster: 2 tweets | Engagement: 2122 (Avg: 130) | Type: Tech

RT @aakashgupta: OpenAI may have caused the worst consumer hardware crisis in a decade with purchase orders that were never real.

In October 2025, Sam Altman flew to Seoul and signed simultaneous deals with Samsung and SK Hynix for 900,000 DRAM wafers per month. That's 40% of global supply. Neither company knew the other was signing a similar commitment at the same time. The pricing and terms would have looked very different if they had.

Those "deals" were letters of intent, not binding purchase orders. No RAM actually changed hands. But the market treated them as real. Contract DRAM prices jumped 171%. A 64GB DDR5 kit went from $190 to$ 700 in three months. DDR4 kits that should have been in oversupply doubled. Retailers stopped posting prices entirely.

The Abilene Stargate expansion just got cancelled because OpenAI couldn't forecast its own demand. Oracle couldn't agree on financing. The partners are squabbling. Bloomberg reported the $500B project hadn't started and no funds were raised to meet the initial budget. Multiple data center buildouts are delayed or shelved.

Now DDR5 prices are finally dropping for the first time in months, and it has nothing to do with OpenAI walking away from anything. Google released TurboQuant on March 24, a compression algorithm that cuts AI memory requirements by 6x. SK Hynix and Samsung stocks dropped 6% and 5% overnight. Corsair kits fell $60-100 from their highs within days.

One company locked up 40% of global memory with commitments it may never fulfill. A different company published a research paper. The research paper is doing more for RAM prices than the entire supply chain has done in six months.

See 1 related tweets

@edzitron: OpenAI never had a commitment of any kind, they did not agree to buy $71bn of anything, the LOI was ...

9. SakanaAILabs (Group Score: 67.2 | Individual: 44.5)

Cluster: 2 tweets | Engagement: 776 (Avg: 160) | Type: Tech

RT @SakanaAILabs: The AI Scientist: Towards Fully Automated AI Research, Now Published in Nature

Nature: https://t.co/nNfpSV5e5I Blog: https://t.co/i6h8LVQOdl

When we first introduced The AI Scientist, we shared an ambitious vision of an agent powered by foundation models capable of executing the entire machine learning research lifecycle.

From inventing ideas and writing code to executing experiments and drafting the manuscript, the system demonstrated that end-to-end automation of the scientific process is possible.

Soon after, we shared a historic update: the improved AI Scientist-v2 produced the first fully AI-generated paper to pass a rigorous human peer-review process.

Today, we are happy to announce that “The AI Scientist: Towards Fully Automated AI Research,” our paper describing all of this work, along with fresh new insights, has been published in @Nature!

This Nature publication consolidates these milestones and details the underlying foundation model orchestration. It also introduces our Automated Reviewer, which matches human review judgments and actually exceeds standard inter-human agreement.

Crucially, by using this reviewer to grade papers generated by different foundation models, we discovered a clear scaling law of science. As the underlying foundation models improve, the quality of the generated scientific papers increases correspondingly. This implies that as compute costs decrease and model capabilities continue to exponentially increase, future versions of The AI Scientist will be substantially more capable.

Building upon our previous open-source releases (https://t.co/H1tBT14Yx8), this open-access Nature publication comprehensively details our system's architecture, outlines several new scaling results, and discusses the promise and challenges of AI-generated science.

This substantial milestone is the result of a close and fruitful collaboration between researchers at Sakana AI, the University of British Columbia (UBC) and the Vector Institute, and the University of Oxford. Congrats to the team!

@chris_lu @cong_ml @RobertTLange @_yutaroyamada @shengranhu @j_foerst @hardmaru @jeffclune

See 1 related tweets

@heynavtoor: RT @heynavtoor: 🚨 An AI just wrote a scientific paper.

Came up with the hypothesis. Designed the ex...

10. clairevo (Group Score: 63.5 | Individual: 39.5)

Cluster: 2 tweets | Engagement: 180 (Avg: 149) | Type: Tech

You all know I'll give you my honest take on these AI tools, and my first experience with @openclaw (RIP "Clawdbot") was not a smooth one.

But there was enough there I had to keep going.

Now, I cannot shut up about my army of OC agents, so it was a treat to hang out with @lennysan and yap about my love for Polly 🦞 and her AI agent crew.

In this very special How I AI x Lenny's Podcast crossover ep, we walk through what openclaw is, how to set it up, and my journey to 10 agents running @chatprd, my @MavenHQ course, and my life.

It might feel like a hacker's tool right now, but I believe this is the future of agent experience.

Don't sleep on our little lobster friend.

Snap to it ➡️ https://t.co/xHyloo5APP\n\nQT @lennysan: Claire Vo's first day with @OpenClaw it deleted her family calendar.

Now she runs 9 agents across 3 Mac Minis, and said "I haven't felt like this since I was a teenager learning to code."

Her sales agent Sam does a daily CRM sweep, identifies decision-makers from new signups, and sends personalized outreach—replacing a part-time salesperson she was paying 10 hours a week.

Her home agent Finn pings her and her husband every day at 3pm: "Which of you is picking up which kids?" Then flags when the oldest's basketball conflicts with the middle kid's soccer and asks how they want to split duties.

She also has agents for podcast prep, kids' homework help, and course project management.

Claire (host of How I AI, founder of @ChatPRD) started as one of OpenClaw's most vocal skeptics. She now calls it "a ChatGPT moment."

In our in-depth conversation, she breaks down: 🔸 Her exact setup: Mac Mini, separate Gmail, dedicated local account 🔸 The progressive trust model: first calendar access, then read email, then draft, then send — just like onboarding an EA 🔸 Why one agent is a mistake—and why she thinks about it like Slack channels, not a single assistant 🔸 How to use Claude Code as a "brain surgeon" to fix and manage your OpenClaw when things break 🔸 "The yappers API" — why rambling into a voice note is the highest-bandwidth way to set up your agent 🔸 Why management skills matter more than technical skills for making this actually work

Listen now 👇 https://t.co/PXMDm2LEGz

See 1 related tweets

@danshipper: RT @clairevo: You all know I'll give you my honest take on these AI tools, and my first experience w...

11. blackboxai (Group Score: 59.5 | Individual: 35.1)

Cluster: 2 tweets | Engagement: 301 (Avg: 100) | Type: Tech

New release on VSCode in the next few hours:

Run Claude Code with OpenAI models
Run Codex with Anthropic Models https://t.co/xKuWrnCfjN

See 1 related tweets

@RobRizk1: CLAUDEX...\n\nQT @blackboxai: New release on VSCode in the next few hours:
Run Claude Code with O...

12. burkov (Group Score: 58.8 | Individual: 49.0)

Cluster: 2 tweets | Engagement: 936 (Avg: 80) | Type: Tech

People crying they miss coding are annoying. So many of them here, half of my feed.

Miss coding? Start the fuck coding! Somebody banned you from the keyboard?\n\nQT @vasuman: I miss coding

Like real coding. Where you name a variable something stupid and refactor methods one at a time. And you don’t know if it compiles until you run it.

The dopamine hit when it does.

Sad

See 1 related tweets

@burkov: RT @burkov: People crying they miss coding are annoying. So many of them here, half of my feed.

Mis...

13. ivanfioravanti (Group Score: 58.3 | Individual: 34.4)

Cluster: 2 tweets | Engagement: 330 (Avg: 76) | Type: Tech

RT @0xSero: Best models to run on your hardware:

—— 64 GB ——

Qwen3-coder-next-80B-4bit (coding, Claude code, general agent)
Qwen3.5-122B-reap: (browser use, multimodal, tool calling, general agent)

—— 96 GB ——

GLM-4.6V (multimodal and tool calls)
Hermes-70B (Jailbroken)
Nemotron-120B-Super: (openclaw)
Mistral-4-Small (general agent)

—— 192 GB ——

All these are excellent top tier LLMs and approach sonnet in capabilities

Step-3.5-Flash
Qwen3.5-397B-REAP
MiniMax-M2.5 (soon M2.7)
GLM-4.7-Reap

See 1 related tweets

@ferologics: god damn i need another 5090 to run Qwen3-coder-next-80B-4bit\n\nQT @0xSero: Best models to run on y...

14. ivanfioravanti (Group Score: 53.9 | Individual: 27.2)

Cluster: 2 tweets | Engagement: 15 (Avg: 76) | Type: Tech

This book rocks! Not only for the Reasoning Deep Dive, but for all coding, appendixes and living repo around it.\n\nQT @rasbt: It’s done.

All chapters of Build A Reasoning Model (From Scratch) are now available in early access.

The book is currently in production and should be out in the next months, including full-color print and syntax highlighting.

There’s also a preorder up on Amazon. https://t.co/ANaJHjpC2s

See 1 related tweets

@rasbt: It’s done.

All chapters of Build A Reasoning Model (From Scratch) are now available in early access...

15. penberg (Group Score: 52.4 | Individual: 30.7)

Cluster: 2 tweets | Engagement: 51 (Avg: 35) | Type: Tech

This matches my empirical experience. If you just let the coding agent work without manually or automatically cleaning up the code, the agent starts to have difficulties working on the code.\n\nQT @GOrlanski: We found that agents generate progressively worse code with each iteration. Real developers do not.

SlopCodeBench is the only eval that faithfully measures quality degradation on iterative, long-horizon coding tasks.

https://t.co/JXGHC4w0bv https://t.co/RQkB8wdzAu 🧵 https://t.co/dOvNkrFv2c

See 1 related tweets

@MLStreetTalk: RT @GOrlanski: We found that agents generate progressively worse code with each iteration. Real deve...

16. manthanguptaa (Group Score: 50.2 | Individual: 29.3)

Cluster: 2 tweets | Engagement: 109 (Avg: 106) | Type: Tech

MCPs had a good run, but they haven’t really lived up to the expectations.

A lot of people advocating for MCP clearly haven’t dealt with real production systems. In practice, MCPs tend to bloat the context heavily. You end up pushing tool definitions, schemas, and intermediate state into the model again and again, which just increases token usage and slows everything down.

A simple CLI or tightly scoped tool interface is usually much more efficient. Smaller surface area, clearer contracts, less noise for the model to deal with.

The deeper issue is that MCPs optimize for flexibility, not reliability. That tradeoff sounds good in theory, but in production, you care about predictability, latency, cost, and debuggability. MCPs make all of that harder. More context means worse signal-to-noise, more tokens means higher cost, and all the dynamic behavior makes failures harder to reason about.

They look great in demos because they can do a bit of everything, but real systems don’t need that kind of open-ended orchestration. They need tight loops, clear boundaries, and systems you can actually debug when things go wrong.

MCPs are great for exploration, but not something I would trust in a production-grade setup.\n\nQT @KaranVaidya6: Okay, @gdb is team CLI all the way. @garrytan thinks MCPs suck.

So we hit the streets of SF to see if the city agreed.

We posed a simple question: MCP or CLI?

Basically everyone under the age of 35 said CLI
One person said MCP was as bloated as Java
& unsurprisingly, numerous people told us to touch grass

Final score- MCP: 3 vs CLI: 17

SF has spoken, and @composio listened.

Our universal CLI is now live!

Drop your best CLI vs MCP hot take in the comments and we'll send the best ones some very sick gear 👀

Link to try our CLI in the next thread ⬇️

See 1 related tweets

@manthanguptaa: RT @manthanguptaa: MCPs had a good run, but they haven’t really lived up to the expectations.

A lot...

17. teortaxesTex (Group Score: 48.2 | Individual: 18.9)

Cluster: 3 tweets | Engagement: 34 (Avg: 63) | Type: Tech

> At some point it's already pretty damn spherical yes, but we've never been remotely close to that point The whole question is whether our intelligence is more like an approximate sphere, or more like a damn dodecahedron peeled potato\n\nQT @fchollet: One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". At some point it's already pretty damn spherical and any improvement is marginal.

Now of course smart humans aren't quite at the optimal bound yet on an individual level, and machines will have many advantages besides intelligence -- mostly the removal of biological bottlenecks: greater processing speed, unlimited working memory, unlimited memory with perfect recall... but these are mostly things humans can also access through externalized cognitive tools.

See 2 related tweets

@JaredSleeper: Spot on.\n\nQT @fchollet: One of the biggest misconceptions people have about intelligence is seeing...
@MLStreetTalk: RT @fchollet: One of the biggest misconceptions people have about intelligence is seeing it as some ...

18. burkov (Group Score: 47.2 | Individual: 47.2)

Cluster: 1 tweets | Engagement: 541 (Avg: 80) | Type: Tech

With all due respect to Andrew, in his motivational post, he didn't explain why anyone would write code by hand. I can code, but I consider coding by hand a waste of time.

So, if I, the one who already knows how to code, consider this a waste of time, why would anyone learn something which is very hard to learn only to then consider it a waste of time, like I do?

19. chris_j_paxton (Group Score: 46.8 | Individual: 27.3)

Cluster: 2 tweets | Engagement: 53 (Avg: 109) | Type: Tech

My opinion of Sunday's marketing and product design is incredibly high. You can see the influence from other favorite home robots company Matic as well, and I believe in fact one of their industrial design folks is an alumnus.\n\nQT @carlosdponx: friends I talk to in humanoid robotics are often surprised when I tell them I have family members, who I hadn't talked to about robots before, that told me over Christmas that they were dying to buy a @sundayrobotics

My cousin's wife signed up and got into their data collection program, in hopes of getting on the early list, and she's super excited to use the gloves and help Sunday learn!

The Sunday team found a design and a marketing scheme that penetrated very quickly into the mainstream, in a way that has normies wanting the product in their home today, rather than cautious curiosity. They can easily imagine it alongside them in their homes.

More teams would probably do well to follow their example if targeting consumers!

See 1 related tweets

@chris_j_paxton: RT @carlosdponx: friends I talk to in humanoid robotics are often surprised when I tell them I have ...

20. garrytan (Group Score: 46.3 | Individual: 29.2)

Cluster: 2 tweets | Engagement: 1718 (Avg: 333) | Type: Tech

The unit of software production has changed from team-years to founder-days.

Act accordingly.

See 1 related tweets

@tunguz: That was so last week. Now it’s agent-seconds.\n\nQT @garrytan: The unit of software production has ...