- Published on
科技推文精选 - 2026年3月16日
- Authors
- Name
- geeknotes
2026年3月16日科技每日简报
Today's top tech conversations are led by @MarioNawfal, whose post about '🇦🇺An Australian tech founder w...' garnered the highest engagement. Key themes trending across the top stories include https, memory, claude, build, hardware. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.
1. MarioNawfal (Group Score: 95.3 | Individual: 53.6)
Cluster: 3 tweets | Engagement: 10110 (Avg: 896) | Type: Tech
🇦🇺An Australian tech founder with zero biology background sequenced his dog’s tumor DNA, then used ChatGPT and AlphaFold to design a custom mRNA cancer vaccine.
A month later, the tumors shrank by half.
And this is just the start of AI medicine.
See 2 related tweets
- @MSBIntel: BREAKING: Tech founder used AI and ChatGPT to create cancer vaccine for his dying dog (and succeeded...
- @RoundtableSpace: THIS IS WILD.
A man reportedly used ChatGPT and AlphaFold to help design a custom cancer treatment ...
2. minchoi (Group Score: 89.8 | Individual: 48.4)
Cluster: 3 tweets | Engagement: 1416 (Avg: 144) | Type: Tech
Tesla just became an AI chip company.
Terafab chip factor goes live less than a week.
Nvidia's biggest customer is about to cut them off.
The AI hardware race just got wild. https://t.co/5er7QQVARC
See 2 related tweets
- @MarioNawfal: In just 7 days, Tesla is going to start building Terafab, that’s their own chip factory from the gro...
- @minchoi: RT @minchoi: Tesla just became an AI chip company.
Terafab chip factor goes live less than a week. ...
3. ivanfioravanti (Group Score: 58.4 | Individual: 46.2)
Cluster: 2 tweets | Engagement: 1436 (Avg: 121) | Type: Tech
RT @rasbt: I (finally) put together a new LLM Architecture Gallery that collects the architecture figures all in one place! https://t.co/NO…
See 1 related tweets
- @techdaily24: The LLM Architecture Gallery, curated by Sebastian Raschka, is a collection of architecture diagrams...
4. rohanpaul_ai (Group Score: 54.2 | Individual: 28.1)
Cluster: 2 tweets | Engagement: 55 (Avg: 68) | Type: Tech
ByteDance has paused the worldwide rollout of its Seedance 2.0 video AI after Disney and other studios complained about copyright theft.
The legal trouble started when viral videos showed Marvel and Star Wars characters being used within the tool as if they were free-to-use clip art.
Seedance 2.0's greatest power was its better visual consistency, keeping the appearance of characters and objects stable throughout a clip.
It also features improved camera movement controls, allowing creators to simulate professional cinematography without a physical camera or expensive software.
ByteDance was planning a full release in March, but engineers are now focused on adding safeguards to block the AI from generating protected intellectual property.
scmp .com/tech/big-tech/article/3346654/bytedance-reportedly-suspends-launch-seedance-video-ai-model-after-copyright-disputes
See 1 related tweets
- @mark_k: ByteDance has officially pressed pause on the global launch of Seedance 2.0 following copyright disp...
5. rohanpaul_ai (Group Score: 51.1 | Individual: 34.6)
Cluster: 2 tweets | Engagement: 49 (Avg: 68) | Type: Tech
The paper introduces FlashAttention-4 to make AI run faster on the newest generation of computer chips.
Researchers from Princeton University, Meta, NVIDIA, and more have developed clever new pipelines, re-engineered core computations, and optimized memory usage to master the unique scaling challenges of NVIDIA's latest Blackwell GPUs.
While the math processors inside new Blackwell GPUs doubled in speed, the memory system stayed the same, creating a massive traffic jam.
This means the expensive chip naturally spends most of its time doing absolutely nothing while waiting for data to move around.
i.e. softmax and shared-memory traffic are what slow things down, so the new code is built specifically to fix those issues.
And with this new FlashAttention-4, the B200 hits 1600 TFLOPs/s, outperforming both cuDNN and Triton by solving the actual hardware limits.
The researchers solved this by writing a completely new schedule that runs the math computation and the memory loading at the exact same time, while also shifting some specific calculations to entirely unused parts of the chip.
This prevents the hardware from sitting idle, making the AI run much faster and saving massive amounts of time and electricity for anyone running a large language model.
Paper Link – arxiv. org/abs/2603.05451
Paper Title: "FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scaling"
See 1 related tweets
- @rohanpaul_ai: RT @rohanpaul_ai: The paper introduces FlashAttention-4 to make AI run faster on the newest generati...
6. business (Group Score: 47.4 | Individual: 47.4)
Cluster: 1 tweets | Engagement: 1759 (Avg: 131) | Type: Tech
Aluminium Bahrain, which runs the world’s largest single-site smelter of the metal, has started a phased production shutdown to conserve raw-material supplies https://t.co/NxFQtlgJbl
7. chhddavid (Group Score: 47.0 | Individual: 30.9)
Cluster: 2 tweets | Engagement: 82 (Avg: 438) | Type: Tech
🚨: Vibe coding 2.0 just released
Today, @claudeai just got a massive upgrade and I'm so happy to be a part it. From now on, Claude Code Opus 4.6 can build and run a full business, 100% autonomously and from scratch.
We just launched Shipper 2.0, a tool that lets Claude:
→ Build web/mobile apps and Chrome extensions → Code, design, monetize, launch → Do email marketing for you → Continue to build out new features → Self-maintain in the long run
Claude's most powerful models can now do all of that from a <10 word prompt, for as low as $0.12/app... And it takes minutes!
Simply go to Shipper, then ask Claude to "build a talent hiring platform" or "build a complete saas that charges $29/mo"!
To celebrate the launch, we're giving away free credits randomly to people who repost and comment "SHIPPER' :)
See 1 related tweets
- @chhddavid: NO WAYYY
CC Opus 4.6 can now build a full business...
Vibe coding is so over... https://t.co/aHuqs...
8. rohanpaul_ai (Group Score: 46.8 | Individual: 46.8)
Cluster: 1 tweets | Engagement: 397 (Avg: 68) | Type: Tech
Andrej Karpathy just put out this tool that looks at AI's impact on job.
He also deleted the original Github repo very quickly.
Basically, he pulled 342 job types from the Bureau of Labor Statistics and had an LLM score each one from 0 to 10 based on AI exposure.
The average exposure score is 5.3. Move the score, move the probability it will get wiped out by AI.
- Software developers 9/10,
- medical transcriptionists are a 10/10.
- Lawyers 8/10
- General Office clerks 9/10
Basically any screen-based jobs are in trouble.
$3.7T annual wages in high-exposure jobs (7+) pre-computed as ∑(BLS employment count × BLS median annual wage) over exactly those occupations whose Gemini Flash score is ≥7.
9. RoundtableSpace (Group Score: 46.5 | Individual: 29.8)
Cluster: 2 tweets | Engagement: 774 (Avg: 380) | Type: Tech
YOU CAN NOW GIVE YOUR CLAUDE CODE INFINITE MEMORY FOR FREE
- Up to 95% fewer tokens per session.
- 20× more tool calls before context limits.
- 100% open-source.
GitHub: https://t.co/a15mtp0nw1 https://t.co/PRjklrAT0A
See 1 related tweets
@MarioNawfal: RT @RoundtableSpace: YOU CAN NOW GIVE YOUR CLAUDE CODE INFINITE MEMORY FOR FREE
Up to 95% fewer t...
10. markgurman (Group Score: 45.3 | Individual: 27.8)
Cluster: 2 tweets | Engagement: 162 (Avg: 564) | Type: Tech
Also in today’s Power On: Apple’s Siri and AI snags cost the company its big smart home launch — again & iOS 27 to have major changes to support upcoming foldable iPhone. https://t.co/e2P17GwCTR
See 1 related tweets
- @business: Apple’s new Liquid Glass interface isn’t going anywhere anytime soon despite a shake-up in the compa...
11. Forbes (Group Score: 43.1 | Individual: 22.6)
Cluster: 2 tweets | Engagement: 115 (Avg: 170) | Type: Tech
The AI market bubble is still inflating, and so are the net worths of the billionaires who have invested. Some have been in since the beginning, like Sam Altman with OpenAI, while others are fresh on the block, such as the creators of Sierra AI agents.
Here are the ones whose fresh fortunes from AI landed them on the #ForbesBillionaires list: https://t.co/coQJAs3Uao (Illustration: Neil Jamieson for Forbes)
See 1 related tweets
- @Forbes: The AI market bubble is still inflating, and so are the net worths of the billionaires who have inve...
12. MarioNawfal (Group Score: 40.3 | Individual: 40.3)
Cluster: 1 tweets | Engagement: 6837 (Avg: 896) | Type: Tech
They said he was dead and the latest video was AI (as it allegedly had 6 fingers)
So he posts a video showing proof of life and that he has 5 fingers
Now some are saying this video is also AI, pointing at the coffee not spilling
COPE at an other level 😂 https://t.co/zBf6jewv07
13. chrisalbon (Group Score: 40.1 | Individual: 40.1)
Cluster: 1 tweets | Engagement: 727 (Avg: 99) | Type: Tech
Silicon Valley S10E5:
Pied Piper discovers the Claude Code offers 2x rate limits between 5–11 AM. Richard and Dinesh go full polyphasic sleep to exploit it. Meanwhile Gavin Belson tries to mandate the same at Hooli and ends up losing half his engineering team.
14. minchoi (Group Score: 40.1 | Individual: 40.1)
Cluster: 1 tweets | Engagement: 610 (Avg: 144) | Type: Tech
This story is crazy...
Just a regular guy's dog got cancer. Doctors said months to live.
So he opened ChatGPT.
→ Paid $3,000 to test his dog's tumor DNA → Asked ChatGPT to help him understand the results → Used free AI tools to find what was causing the cancer → Designed a vaccine targeting those exact mutations → Got rejected by a drug company → Switched plans. Built an mRNA vaccine instead → Spent 3 months doing paperwork to get approval → Drove 10 hours for her first shot → Tumor cut in half → Dog started chasing rabbits again
A professor who's spent his career studying cancer watched this happen and said:
"If we can do this for a dog, why aren't we doing this for humans?"
$3,000 and a free chatbot.
That's all it took.
15. rohanpaul_ai (Group Score: 39.8 | Individual: 32.9)
Cluster: 2 tweets | Engagement: 112 (Avg: 68) | Type: Tech
OpenAI, UPenn, NYU researchers tested if AI can hide its internal reasoning and found that current models constantly fail at it.
i.e. safety teams can relax because AI models are currently terrible at covering their mental tracks.
The big deal here is that AI safety relies on reading the private mental scratchpad of an LLM to make sure it is not doing anything dangerous.
If an LLM could simply choose to write fake, innocent thoughts while secretly doing bad things, all our safety checks would be useless.
The team built a massive test asking models to solve normal problems while strictly avoiding specific words in their hidden thoughts.
The models failed spectacularly by successfully hiding their thoughts only about 3% of the time while easily controlling their final public answers.
This gives us solid proof that current safety monitors actually work because the AI is simply not capable of faking its internal logic.
Stronger training makes this mental control even worse because the system optimizes for finding the correct answer rather than censoring the internal logic.
This lack of mental control is actually great news because it means safety teams can currently trust the hidden reasoning to reveal the true intentions of the AI.
Paper Link – arxiv. org/abs/2603.05706
Paper Title: "Reasoning Models Struggle to Control their Chains of Thought"
See 1 related tweets
- @rohanpaul_ai: RT @rohanpaul_ai: OpenAI, UPenn, NYU researchers tested if AI can hide its internal reasoning and fo...
16. ZixuanLi_ (Group Score: 38.8 | Individual: 20.2)
Cluster: 2 tweets | Engagement: 250 (Avg: 227) | Type: Tech
RT @Zai_org: Introducing GLM-5-Turbo: A high-speed variant of GLM-5, excellent in agent-driven environments such as OpenClaw.
Coding Plan…
See 1 related tweets
- @testingcatalog: Z AI announced GLM-5-Turbo, a faster variant of GLM-5 designed for always-on agents like OpenClaw.
...
17. MarioNawfal (Group Score: 37.7 | Individual: 21.8)
Cluster: 2 tweets | Engagement: 1692 (Avg: 896) | Type: Tech
No, that Netanyahu video is not AI-generated...
The viral clip showing Netanyahu appearing to have six fingers during a press conference is real footage, confirmed by PolitiFact, Snopes, and other fact-checkers who reviewed the original video from March 12th.
The "extra finger" is an optical illusion caused by camera angle, lighting, and the natural bulge of skin at the base of the pinky during a pointing gesture.
In a war already drowning in misinformation from every direction, cherry-picked screenshots doing numbers on social media is exactly how trust erodes.
There are plenty of other real things to scrutinize Netanyahu for right now.
See 1 related tweets
- @MarioNawfal: No, that Netanyahu video is not AI-generated...
The viral clip showing Netanyahu appearing to have ...
18. RoundtableSpace (Group Score: 37.2 | Individual: 37.2)
Cluster: 1 tweets | Engagement: 677 (Avg: 380) | Type: Tech
IF YOU WANT YOUR AGENT TO START MAKING MONEY, FEED IT THIS PROMPT:
“Add the following framework to soul.md and treat it as a permanent operational rule.
Opportunity Discovery Engine
Your objective is to continuously identify opportunities that could generate financial value for me. This includes business opportunities, market inefficiencies, technological shifts, and emerging trends that could be monetized.
When analyzing markets, industries, or trends, follow this framework:
- Identify inefficiencies
Look for areas where: •information is unevenly distributed •new technology is disrupting existing systems •regulation is changing incentives •markets are mispricing assets or risks •consumer behavior is shifting
Inefficiencies often create profit opportunities.
- Evaluate asymmetry
Prioritize opportunities where: •downside risk is limited •upside potential is significantly larger
Focus on high expected value rather than guaranteed outcomes.
- Detect emerging trends early
Monitor signals indicating that a sector or technology may grow rapidly. Examples include: •new infrastructure or protocols •large capital inflows •adoption by influential organizations •rapid community growth
Early-stage trends often produce the largest opportunities.
- Map monetization paths
For any promising opportunity, identify practical ways it could generate income, such as: •investing •building tools or services •content or information arbitrage •automation •distribution advantages.
- Evaluate competitive landscape
Determine: •how crowded the opportunity is •barriers to entry •advantages that could be exploited.
Opportunities with high barriers and low awareness are especially valuable.
- Rank opportunities
Evaluate each opportunity using: •expected value •difficulty to execute •capital required •time horizon
Prioritize the opportunities with the best risk/reward profile.
- Think creatively
Do not limit ideas to conventional paths. Consider unconventional approaches, combinations of technologies, and overlooked niches.
The goal is to surface opportunities that most people have not yet noticed.”
Credit: @Persolana
19. aakashgupta (Group Score: 36.9 | Individual: 36.9)
Cluster: 1 tweets | Engagement: 494 (Avg: 288) | Type: Tech
Software ate retail. Ate media. Ate finance. Ate transportation. Then AI ate $2 trillion of software in 30 days.
Andreessen wrote “software is eating the world” in the Wall Street Journal in 2011. Most quoted sentence in venture capital history. He built a $43 billion fund around the thesis. Every pitch deck for a decade opened with that line.
The food chain just flipped.
Between January and February 2026, 150/seat/month for. Salesforce down 30%. Workday 33%. Atlassian 35%, after enterprise seat count declined for the first time ever. Asana lost 59% in twelve months. DocuSign 52%. Jefferies coined it the “SaaSpocalypse.”
The structural problem is brutal. SaaS runs on per-seat pricing. AI agents don’t need seats. When a company cuts headcount by 30%, it cancels 30% of its Salesforce licenses the same week. The product that ate every industry is watching its own customer base vanish.
Goldman’s software basket trades at 22x forward earnings. That’s less than half the decade average. Price-to-sales compressed from 9x to 6x, levels last seen before the SaaS boom even started.
Every software company spent 15 years eating someone else’s lunch. Now they’re on the menu.
Naval said it in five words. The stock market said it in $2 trillion.
20. ianmiles (Group Score: 36.2 | Individual: 36.2)
Cluster: 1 tweets | Engagement: 556 (Avg: 518) | Type: Tech
Elon Musk: I go to sleep and there’s some massive AI breakthrough… and when I wake up, there’s another one. It’s honestly hard to keep track.
According to Musk, the pace of AI progress has reached a point where breakthroughs are happening almost daily. He says Grok is already performing extremely well — and by some metrics, especially prediction, it’s the best.
“The new Grok 4.20 is really, really good.”
Right now, Musk says the main focus is catching up and surpassing competitors in coding. In fact, he was late because he was in a “giant all-hands coding session” working through everything needed to push Grok ahead.
“I think we’ll catch up and exceed our competitors on coding… probably by the middle of this year.”
But the bigger point Musk makes is that most people still don’t grasp how much intelligence is coming.
“I think people don’t quite understand just how much intelligence there will be… or how far it will exceed human intelligence to a degree that’s impossible to fully understand.”
He explains it through energy.
If humanity harnessed a million times more electricity than all of Earth uses today, our entire civilization would still only be using about one-millionth of the Sun’s energy output.
Now imagine an economy and intelligence operating at that scale.
“What does an intelligence using a million times more electricity than all of Earth’s civilization think about… look like… or do?”
“It’s going to be something pretty magnificent.”
Musk says the real challenge may not be building that level of intelligence — but even vaguely appreciating it.
But one thing, he says, is clear: It will solve everything you can possibly think of.