科技热推 - 2026年6月30日

科技每日简报 2026年6月30日

Today's top tech conversations are led by @kieranklaassen, whose post about 'RT @bcherny: As engineering, p...' garnered the highest engagement. Key themes trending across the top stories include model, models, product, deepseek, already. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. kieranklaassen (Group Score: 362.3 | Individual: 62.8)

Cluster: 9 tweets | Engagement: 3902 (Avg: 324) | Type: Tech

RT @bcherny: As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:

Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship
Builder: quickly turns a prototype/idea into production-grade product/infra
Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance
Grower: takes a product that has been built and iterates on it to improve Product-Market Fit
Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales

Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS.

A healthy team needs a mix of these, depending on the product:

A product that is new and pre-PMF needs people that are strong at 1+2+3
A product that is growing and has found PMF needs 2+3+4 and some 5
A product that has strong PMF needs 3+4+5 and some 2

Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?

See 8 related tweets

@dani_avila7: 100% agree with Boris, but…

These archetypes only work inside orgs that are already technically and...

@yrechtman: Obviously yes. https://t.co/eHSbNDf7qx\n\nQT @bcherny: As engineering, product, design, DS, etc. mel...
@pmarca: Interesting.\n\nQT @bcherny: As engineering, product, design, DS, etc. melt into a new kind of role,...
@SIGKITTEN: 2 kind of roles still gonna be relevant: the Grower and the Shower\n\nQT @bcherny: As engineering, p...
@dejavucoder: sweeper be like https://t.co/iItbgmX8Tb\n\nQT @bcherny: As engineering, product, design, DS, etc. me...

2. PolymarketMoney (Group Score: 318.0 | Individual: 41.7)

Cluster: 13 tweets | Engagement: 3025 (Avg: 311) | Type: Tech

BREAKING: Anthropic CEO says open-source AI is moving down a “very dangerous path.”

See 12 related tweets

@RnaudBertrand: In other words: "please, dear legislators, ban my competition"\n\nQT @coinbureau: 🚨ANTHROPIC CEO: OP...
@heyshrutimishra: Anthropic CEO Dario Amodei told lawmakers that open-source AI is moving down a “very dangerous path....
@paoloardoino: Centralized big-tech AI is starting a warpath against open-source AI models. The excuse: safety. The...
@WesRoth: A 2023 warning from Anthropic CEO Dario Amodei about openly released AI models is resurfacing as ope...
@sairahul1: RT @rewind02: Anthropic CEO Dario Amodei:

"We might be six to twelve months away from AI doing mos...

3. VaibhavSisinty (Group Score: 295.8 | Individual: 32.3)

Cluster: 11 tweets | Engagement: 133 (Avg: 225) | Type: Tech

This is actually wild. 🤯

Five of the best open weight AI models in the world. One subscription. $9.99 a month.

It is called Cline Pass. And it just changed how I think about AI subscriptions.

GLM-5.2, DeepSeek, Kimi, Qwen, MiniMax. All of them. One place. No switching between APIs.

No managing multiple keys. No surprise bills.

You get 2 to 5x discounted access compared to standard API pricing.

The models are already matching and beating closed models on coding benchmarks.

And now you can run them inside Cline CLI and IDE directly.

For developers this is the unlock. Your entire coding workflow powered by the best open weight models in the world at a fraction of the cost.

Sign up via Cline today and it drops to $1.99 for the first month.\n\nQT @cline: We’ve been impressed with GLM-5.2 and so are introducing a$ 9.99/month subscription to give you 2-5x discounted access to it and other open weight models like DeepSeek, Kimi, MiniMax, Mimo, Qwen.

Use it on Cline CLI & IDE with $1.99 special promo if sign up via: npm i -g cline https://t.co/25tCruyG7R

See 10 related tweets

@alexcooldev: Most of my coding doesn't need a frontier model on every turn.

Burning Opus or GPT-5.5 tokens on ro...

@Scobleizer: Open weights are becoming daily drivers for coding work. Models like Glm 5.2 and Kimi k2.7 code hit ...
@Hesamation: many devs are already moving to open source models like GLM 5.2 for coding, this is the easiest way ...
@heyrobinai: i stopped tracking which open model is newest and best.. and got hours back every week..

GLM 5.2, K...

@dr_cintas: ClinePass just launched. $9.99/month for curated access to the best open-weights coding models insid...

4. BullTheoryio (Group Score: 237.0 | Individual: 33.7)

Cluster: 9 tweets | Engagement: 422 (Avg: 1190) | Type: Tech

🚨 SOUTH KOREA JUST UNVEILED $650 BILLION IN AI AND CHIP SPENDING.

Samsung and SK Hynix announced a combined investment plan that could top 1,000 trillion won ($650 BILLION) over the next decade, alongside President Lee Jae Myung's push for balanced regional growth.

Here's the breakdown:

Samsung and SK Hynix announce 800 trillion won($518 B) investment in four semiconductor fabrication plants.
81 trillion won ($52.45B) goes into HBM packaging facilities.
30 trillion won ($19.42B) over 15 years for next-generation memory tech.
Seoul's DRAM production is targeted to double in 5 years.
18.4 gigawatts of AI data center capacity planned by 2035, starting with 8.4 gigawatts by 2029 from SK Group, GS Group, and Naver.

Samsung's Chairman Lee Jae-yong said current capacity can't meet demand, with new investments also going into robotics, biotech, batteries, and chip substrates across multiple regions.

Despite the scale of the announcement, both stocks fell. Samsung dropped -5.3%, SK Hynix fell -3.4%.

A single semiconductor fab alone costs at least 60 trillion won ($40 billion) and takes years to build, so the actual supply increase from this plan won't materialize for years, while investors have to absorb the financing cost now.

That's also why the Korea Exchange pulled a planned options launch tied to these stocks, the market is pricing in execution and funding risk over the headline number itself, especially with South Korea's market already down 10% in the past week on broader AI valuation fears.

See 8 related tweets

@cryptorover: 🚨 SOUTH KOREA JUST ANNOUNCED A $518 BILLION AI CHIP MEGA-PROJECT.

Samsung and SK Hynix are building...

@rohanpaul_ai: Samsung and SK Hynix could announce as much as $1.3T of investment over 10 years on Monday.

Samsung...

@cryptorover: 🚨 SOUTH KOREA’S MARKET JUST SENT A MAJOR WARNING SIGNAL.

KOSPI erased nearly ₩100 trillion today.

...

@business: South Korean firms including Samsung and SK Hynix will spend at least 1,350 trillion won ($880 billi...
@VaibhavSisinty: The two companies that make the memory inside every NVIDIA GPU on earth just announced $518 billion ...

5. imsroch (Group Score: 224.3 | Individual: 49.8)

Cluster: 7 tweets | Engagement: 1342 (Avg: 171) | Type: Tech

RT @cursor_ai: Introducing Cursor for iOS.

Build from anywhere by launching always-on cloud agents. Or remotely control agents running on your computer from the app.

Composer 2.5 is 75% off in the app now through July 5. https://t.co/dFxQyrgmBb

See 6 related tweets

@Baconbrix: I want to try building a mobile app with this 👀\n\nQT @cursor_ai: Introducing Cursor for iOS.

Build...

@elonmusk: Cursor for iOS!\n\nQT @cursor_ai: Introducing Cursor for iOS.

Build from anywhere by launching alwa...

@VaibhavSisinty: Cursor just launched on iOS. And it changes what coding from your phone actually means. 🤯

You can n...

@kentcdodds: I've been using this for about a month and it's it's been really nice to be able to take my agent wi...
@kieranklaassen: Have been using this for the last month every day! Can’t do without it anymore!

Just simple, funct...

6. FirstSquawk (Group Score: 217.3 | Individual: 42.7)

Cluster: 9 tweets | Engagement: 281 (Avg: 63) | Type: Tech

South Korea has unveiled a $520 billion semiconductor manufacturing project, aiming to build the world's largest chipmaking cluster in partnership with Samsung Electronics and SK Hynix to strengthen its global leadership in semiconductors.

The mega project will establish multiple advanced chip fabrication plants, research centers, and supporting industries, creating a complete semiconductor ecosystem while boosting production of AI, memory, and advanced logic chips.

The government will provide infrastructure, power, water, tax incentives, and regulatory support, with the cluster expected to attract hundreds of suppliers and create thousands of high-skilled jobs over the coming decades.

The initiative is designed to enhance supply chain resilience, maintain South Korea's competitiveness against rivals such as the United States, China, and Taiwan, and meet the rapidly growing global demand for AI and next-generation semiconductor technologies.

See 8 related tweets

@business: Samsung and SK Hynix will build two chip plants each in a $518 billion plan aimed at cementing South...
@WesRoth: South Korea unveiled a massive $576 billion strategy to strengthen its leadership in AI and semicond...
@wallstengine: South Korea 🇰🇷 unveiled a $576B+ AI and chip investment push

Samsung and SK Hynix will invest ~$518...

@business: South Korea is orchestrating investments of at least $880 billion from companies including Samsung E...
@Reuters: South Korea rolled out sweeping chip and AI mega-projects, as President Lee Jae Myung pledged to cem...

7. firstadopter (Group Score: 188.5 | Individual: 37.7)

Cluster: 7 tweets | Engagement: 67 (Avg: 90) | Type: Tech

If our AI policy is going to be driven by what Jamie Dimon fearmongers to Scott Bessent, then god help us.

JPMorgan's incentives may differ from the U.S.\n\nQT @firstadopter: This is how dumb our government is.

China already has advanced models that can find exploits.
By banning Mythos and GPT-5.6, the government denies the general public and companies the ability to DEFEND themselves in cybersecurity.
Haphazard policy, which casts doubt on whether future models will be available, is DRIVING our allies and the rest of the world toward building on non-U.S. models.
The uncertainty may hurt leading U.S. AI companies by limiting their ability to invest aggressively in better models down the line. The business model breaks down if Anthropic and OpenAI can't sell their upcoming models to the world.

If you want a 30-day rational vetting process with clear, transparent rules with no one-off micromanagement on who gets access, fine. Get it done. But this current system is absolute, insane idiocy.

WSJ: "“It is incentivizing companies across the globe to use cheaper but very capable Chinese open-weight models, while at the same time undermining the U.S. AI industry,” said Niels Provos, a researcher who led security teams at Google and Stripe. “I don’t understand it.”

WSJ: "Security researchers said that a new AI model, released this month by China’s Zhipu AI, also known as https://t.co/RkDD1NhZEz, can match the latest U.S. models when it comes to finding security bugs"

See 6 related tweets

@firstadopter: This is how dumb our government is.

China already has advanced models that can find exploits. 2....

@WesRoth: The U.S. wants to prevent China from accessing its strongest models.

But if American models remain ...

@pmarca: Concerning.\n\nQT @quxiaoyin: The worst case scenario for USA AI: 1. Chinese open sources keep gaini...
@VaibhavSisinty: The moment enterprises figure out they can self host a Chinese open weight model, train on their own...
@DataChaz: china is open-sourcing Mythos https://t.co/494VedVcdc\n\nQT @Polymarket: JUST IN: A new Chinese AI m...

8. BullTheoryio (Group Score: 159.6 | Individual: 48.1)

Cluster: 5 tweets | Engagement: 5199 (Avg: 1190) | Type: Tech

🚨 SAMSUNG, SK HYNIX, AND MICRON ARE GETTING SUED FOR ENGINEERING THE MEMORY CHIP SHORTAGE.

The lawsuit, filed June 25 in California, accuses the three companies of using their pivot to AI memory chips as cover to cut production of regular DRAM, the memory used in everyday laptops and phones.

DRAM prices have risen roughly 500-700% over the past four years. Micron reportedly shut down its consumer DRAM brand, Crucial, at the most profitable price point in its history, a move the lawsuit calls economically irrational unless it was coordinated.

The lawsuit points directly to Apple's recent price hikes on iPads and Macs as evidence the damage is already reaching consumers.

This isn't the first time.

Between 1998 and 2002, Samsung, Hynix, Micron, Infineon, and Elpida ran an actual price fixing cartel, confirmed by US federal prosecutors.

Samsung paid a $300 million criminal fine, Hynix paid$ 185 million, and Infineon paid $160 million, with several executives serving real prison time, sentences ranging from 4 to 14 months.

The new lawsuit alleges Samsung and SK Hynix later rehired and promoted some of those same convicted executives into senior roles.

Together, the three companies control the vast majority of global DRAM supply today, and building a single new DRAM factory costs $15 to$ 20 billion and takes years, making it nearly impossible for new competitors to break in and undercut them.

That's the core problem this lawsuit is targeting.

Three companies with total control over a market everyone depends on, the same companies already convicted once before, now facing the same accusation again while prices keep climbing and ordinary buyers have nowhere else to turn.

Jefferies doesn't expect relief anytime soon. Prices are forecast to climb another 40-50% next quarter, then a further 30-40% on top of that the quarter after, meaning prices could roughly double by year end.

2027 is expected to bring another 40-45% increase on top of that, with no real normalization expected until 2028.

See 4 related tweets

@BullTheoryio: BREAKING: $MU is down -7.77% today, wiping out$ 100 billion from its market cap.

Micron is one of t...

@shanaka86: JUST IN: The AI boom may have handed the chip industry the perfect cover for its oldest crime. Three...
@zephyr_z9: Lol This is stupid No cleanroom available to expand capacity\n\nQT @Pirat_Nation: Samsung, SK Hynix,...
@techdaily24: Samsung, SK Hynix, Micron Sued in US over Memory Price Fixing https://t.co/OtqdDjHdFC #technology #T...

9. heyshrutimishra (Group Score: 139.5 | Individual: 27.9)

Cluster: 7 tweets | Engagement: 14 (Avg: 34) | Type: Tech

RT @heyshrutimishra: DeepSeek open-sourced something yesterday that should not be free.

An 85% speed upgrade. Same model. Zero new training. Already running in production on millions of users. And now available to every developer on earth as a free download.

The US-China AI race is usually covered as a model benchmarks competition. Country A releases a model that scores higher on MMLU. Country B releases a reasoning model that beats it on math. The scorecard changes every few weeks and everyone moves on.

DSpark is not that story. DSpark is about who controls the cost of inference at scale, which is the actual competition that matters in 2026.

Here's the problem DSpark solves.

Every large language model generates text one token at a time. Each token requires a full forward pass through the entire model. You feel this as latency. The bigger the model, the slower every single response.

The standard fix is speculative decoding. A small draft model guesses several tokens ahead, then the full model verifies the batch in one pass instead of running one by one. If the guesses are good, you accept them all. If not, you correct and move on. The output is mathematically identical to running the full model normally. The speed is not.

The problem with existing speculative decoding is that it forces a tradeoff. Parallel drafters generate fast but lose token-to-token coherence, so acceptance rates decay badly on long sequences. Sequential drafters preserve quality but are slower to generate. Every system before DSpark picked one side of this tradeoff.

DeepSeek picked neither. They combined a parallel backbone with a lightweight Markov head that injects token dependencies at near-zero cost. The result is a drafter that generates fast and stays coherent. Against Eagle3, the prior state-of-the-art, accepted token length improves by 26 to 31 percent across Qwen3 model families. Against DFlash, it improves by 16 to 18 percent.

Then they solved the second problem: verification waste.

Under heavy server load, verifying long candidate sequences is expensive. If most of those candidates are going to be rejected anyway, the GPU time spent verifying them is pure waste. DSpark adds a confidence prediction head that estimates rejection probability before verification runs, then dynamically trims the verification length per request based on current server load. When traffic is light and GPUs are idle, verify more. When traffic is heavy, verify less and stay fast.

In production on DeepSeek-V4 Flash, per-user generation speed increased by 60 to 85 percent compared to the MTP-1 baseline. On V4 Pro, 57 to 78 percent. Under high-concurrency conditions with strict latency targets, aggregate throughput improved by 51 to 400 percent depending on server load.

This is already running on every user hitting DeepSeek's API today.

The paper was co-authored by DeepSeek founder Liang Wenfeng alongside researchers from Peking University. They then open-sourced the entire training and evaluation stack, called DeepSpec, under an MIT license. The V4 Pro DSpark checkpoint is on Hugging Face. The training code is on GitHub. The framework already works on Qwen3 and Gemma model families, not just DeepSeek's own models.

The pattern here is not accidental. Release the model. Then release the infrastructure that makes it faster. Then release the code so the entire community can replicate it. This is how you build a lead that compounds.

Every major Western lab is focused on the next model release. DeepSeek shipped an 85 percent speed improvement to a model that already existed.

See 6 related tweets

@Analyticsindiam: NEW: DeepSeek has launched DSpark, a new speculative decoding system that significantly speeds up li...
@QuixiAI: RT @yuhasbeentaken: After spending the last 6 hours testing DSpark and reading the full paper, here’...
@teortaxesTex: > the improvement over DFlash: > +20% acceptance length > +14% throughput > avg 127 tok/...
@VaibhavSisinty: DeepSeek just made their AI 5x faster. Without changing the model at all.

It is called DSpark. And ...

@Techmeme: DeepSeek details DSpark, a speculative decoding framework for its V4 models, saying it speeds up AI ...

10. sydneyrunkle (Group Score: 138.3 | Individual: 35.3)

Cluster: 6 tweets | Engagement: 60 (Avg: 37) | Type: Tech

we just released dynamic subagents, which let your agent programmatically orchestrate subagents in a code interpreter.

this lets agents do work at scale that tool calls can't reliably handle, like processing hundreds of documents or analyzing thousands of datapoints!\n\nQT @sydneyrunkle: https://t.co/WoyfH0rF7U

See 5 related tweets

@hwchase17: RT @LangChain: Deep Agents now supports dynamic subagents. Instead of invoking subagents with tool c...
@hwchase17: RT @sydneyrunkle: we just released dynamic subagents, which let your agent programmatically orchestr...
@LangChain: RT @neildahlke: The thing that breaks production agents usually isn't reasoning. It's volume. Hundre...
@sydneyrunkle: RT @huntlovell: We're very excited about dynamic subagents! Go read this great writeup on what they ...
@sydneyrunkle: i ran a bunch of benchmarks this week to confirm dynamic subagent value. a funny quote from one agen...

11. WhaleInsider (Group Score: 138.2 | Individual: 32.1)

Cluster: 6 tweets | Engagement: 3126 (Avg: 1074) | Type: Tech

JUST IN: 🇨🇳 Ex Meta PM and AI founder Xiaoyin Qu says “American and European enterprises will ditch OpenAI and anthropic and adopt Chinese models.” https://t.co/yiIriM84ed

See 5 related tweets

@zephyr_z9: Meanwhile, Anthropic & OpenAI's combined ARR crossed 100B this month This is delusional bullshi...
@IamEmily2050: Again, what actually will happens is the opposite. OpenAI and Anthropic will capture the majority of...
@WesRoth: Xiaoyin Qu predicts that American and European enterprises will increasingly replace OpenAI and Anth...
@JordanSchachtel: RT @innovationcncl: “We now have a Chinese open-weight model that is as good as the currently availa...
@RnaudBertrand: RT @quxiaoyin: American and European enterprises will ditch OpenAI and anthropic and adopt Chinese m...

12. scaling01 (Group Score: 138.0 | Individual: 44.0)

Cluster: 4 tweets | Engagement: 1415 (Avg: 355) | Type: Tech

Sounds like we will get the 1.5T model in July and the 2T model in August\n\nQT @elonmusk: In addition to their excellent and unique training data, the Cursor team is also making major engineering contributions to v9 SFT & RL. It’s an honor and a pleasure to work with them.

For this 1.5T run, Cursor data was added in supplemental training, which is not quite as good as having it in initial training.

The 2T run that started a few weeks ago has greatly improved data in scope & scale in almost every arena, and many upgrades/fixes to the training recipe. That will finish in late July for August release.

See 3 related tweets

@WesRoth: A newer 2-trillion-parameter model began training several weeks ago.

Musk says it includes:

🔹Much ...

@techdevnotes: Elon on Grok 1.5T and 2T models\n\nQT @elonmusk: In addition to their excellent and unique training ...
@martin_casado: RT @elonmusk: In addition to their excellent and unique training data, the Cursor team is also makin...

13. zephyr_z9 (Group Score: 136.2 | Individual: 37.0)

Cluster: 7 tweets | Engagement: 812 (Avg: 178) | Type: Tech

But a memory tourist told me CXMT will flood the market with cheap DRAM 😱😱😱\n\nQT @jukan05: Even Chinese companies are signing LTAs now lol

RTRS:

China’s CXMT has signed a $3 billion LTA with Tencent.
CXMT is also in talks with other Chinese internet companies, including Alibaba Cloud, ByteDance, and Xiaomi.
As of Q1, CXMT’s DDR5 yields still lagged behind Western peers.
CXMT currently operates two 12-inch DRAM fabs in Hefei and one fab in Beijing, with total wafer capacity of around 300k wafers per month.

See 6 related tweets

@teortaxesTex: > - As of Q1, CXMT’s DDR5 yields still lagged behind Western peers. > Hefei, Beijing The funny...
@jukan05: KIS’ comment on Reuters’ report regarding CXMT’s LTA:

“For several reasons, it does not seem easy f...

@rohanpaul_ai: RT @rohanpaul_ai: FT: Apple is asking Washington for permission to buy DRAM from CXMT, a blacklisted...
@zephyr_z9: RT @jukan05: Even Chinese companies are signing LTAs now lol

RTRS:

China’s CXMT has signed a $3 ...
@WesRoth: China’s CXMT has reportedly secured a nearly $3 billion memory-chip supply agreement with Tencent.

...

14. teortaxesTex (Group Score: 130.2 | Individual: 32.4)

Cluster: 5 tweets | Engagement: 137 (Avg: 64) | Type: Tech

> China’s AI playbook: kill OpenAI and anthropic with free great models I have literally never seen any Chinese engineer or AI figure profess the belief that they can kill OpenAI, Anthropic, Google, or even, like, Perplexity. They're quite clear-eyed about this game. https://t.co/C36ekvqGSk\n\nQT @quxiaoyin: China’s AI playbook: kill OpenAI and anthropic with free great models. Make it free. Then use cheap electricity to export compute as well. Currently the blocker is chip but Hauwei would catch up soon. Imagine a world where instead of paying hundreds of billions to OpenAI and anthropic, you pay almost zero to similar level of intelligence with cheap cheap inference. What’s gonna happen?

See 4 related tweets

@cryptopunk7213: the chinese labs really are relentless with accelerating open source ai. deepseek is doing everythin...
@TeksEdge: Normally I would agree but we’ve been seeing Chinese closed source models proliferate recently and n...
@cgtwts: Ex Meta PM and AI founder Xiaoyin Qu:

“American and European enterprises will ditch OpenAI and An...

@RnaudBertrand: RT @quxiaoyin: China’s AI playbook: kill OpenAI and anthropic with free great models. Make it free. ...

15. Hadley (Group Score: 128.5 | Individual: 48.9)

Cluster: 4 tweets | Engagement: 1660 (Avg: 190) | Type: Tech

RT @rauchg: Learn to ship. Shipping is a skill distinct from coding. Shipping is designing, coding, QAing, story-telling, teaching, marketing, selling, pivoting, iterating…

It used to be that coding dominated in importance because of coding ability scarcity. AI will push you to go further.

See 3 related tweets

@pmarca: Learn to ship.\n\nQT @rauchg: Learn to ship. Shipping is a skill distinct from coding. Shipping is d...
@brian_armstrong: True.

A lot of learning to ship is getting over the fear of what others will think. And choosing th...

@sxmawl: Let it rip…\n\nQT @rauchg: Learn to ship. Shipping is a skill distinct from coding. Shipping is desi...

16. VaibhavSisinty (Group Score: 119.8 | Individual: 28.6)

Cluster: 6 tweets | Engagement: 183 (Avg: 225) | Type: Tech

Meta just built a device that reads your thoughts and turns them into text. With just a helmet. 🤯

It is called Brain2Qwerty v2. Published in Nature today.

Here is how it works.

Your brain fires specific signals every time you think about typing a word.

Meta built an AI that watches those signals from outside your skull and decodes what you are thinking.

They trained it on 22,000 sentences from 9 people. Each wore the helmet for 10 hours while typing.

The best performer hit 78% word accuracy. Just from brain waves alone. But here is what most people are missing.

This is not a party trick. This is built for people who had a stroke or a brain injury and lost the ability to speak or move their hands.

People who currently have no way to communicate with the world. This gives them a voice back. Without a single surgery.

Meta open sourced the full code today so every neuroscience lab in the world can build on it.

To be frank, the gap between science fiction and science fact just got a lot smaller.\n\nQT @AIatMeta: We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.

Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.

We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.

🧵👇

See 5 related tweets

@MTSlive: SITUATION DETECTED: Meta has released Brain2Qwerty v2, a non-invasive brain-to-text decoder capable ...
@MTSlive: Meta AI researcher watching Brain2Qwerty decode the first direct brain to timeline gem post https://...
@Polymarket: JUST IN: Meta unveils Brain2Qwerty v2, a non-invasive brain-to-text system that can turn raw brain s...
@coinbureau: ⚡️NEW: META UNVEILS AI THAT TURNS BRAIN SIGNALS INTO TEXT

$META introduced Brain2Qwerty v2, an AI s...

@PolymarketMoney: JUST IN: $META unveils Brain2Qwerty v2, an AI system that translates brain signals into text in real...

17. chandrarsrikant (Group Score: 119.6 | Individual: 36.4)

Cluster: 4 tweets | Engagement: 87 (Avg: 258) | Type: Tech

🚨Persistent's Nagarro deal unnerves market, hinges on seamless integration, margin turnaround

The street believes Pune- based Persistent is buying scale at the cost of near-term margin dilution. Nagarro's adjusted EBITDA margin of 13.9 percent trails Persistent's 18.4 percent, while the acquisition would add about $1.14 billion in revenue but also leave the company with roughly$ 1.6 billion in acquisition financing.

Consequently, shares of Persistent Systems fell over 11 percent, on June 29, after the company announced its acquisition of German firm Nagarro over the weekend, even as analysts broadly endorsed the strategic rationale behind the deal.

The acquisition will create a digital engineering behemoth with nearly $2.9 billion in annual revenue and more than 46,000 employees across over 40 countries, significantly expanding Persistent's European footprint and moving it closer to its stated ambition of becoming a$ 5 billion revenue company by FY31.

By @shaw_reshab and @debanganaghosh4

https://t.co/RYsVAOr2Cv

See 3 related tweets

@chandrarsrikant: Persistent mega M and A - Street view https://t.co/NGQ077OxQS\n\nQT @chandrarsrikant: 🚨Persistent's ...
@chandrarsrikant: Persistent shares down by nearly 8 percent in early trades ...\n\nQT @chandrarsrikant: Persistent Sy...
@shaw_reshab: RT @moneycontrolcom: #Business | Persistent’s Nagarro deal unnerves market, hinges on seamless integ...

18. DivesTech (Group Score: 108.8 | Individual: 26.5)

Cluster: 5 tweets | Engagement: 87 (Avg: 109) | Type: Tech

Great to be on @CNBCMorningCall with @MorganLBrennan to kick the week off discussing AI and tech market jitters and our view this creates buying opportunities to own the tech winners 🏆🐂🍿🧱📺🎯👇@CNBC https://t.co/aWFABuqAGN\n\nQT @CNBCMorningCall: .@DivesTech says hyperscalers are oversold, memory remains in a multiyear AI supercycle, and Big Tech is poised to rebound in the second half as AI monetization and earnings become the next major catalysts.

$MSFT$ META $GOOGL$ ORCL $MU #AI

https://t.co/CIWjtzK6Hy

See 4 related tweets

@DivesTech: Another big week ahead for the markets and tech stocks @CNBCMorningCall @MorganLBrennan @barbara_dor...
@DivesTech: AI monetization will be a focus area in 2Q earnings for the tech sector @CNBCMorningCall @MorganLBr...
@CNBCMorningCall: .@DivesTech says hyperscalers are oversold, memory remains in a multiyear AI supercycle, and Big Tec...
@MorganLBrennan: Time to buy back into the hyperscalers?\n\nQT @DivesTech: AI monetization will be a focus area in 2Q...

19. pvncher (Group Score: 108.0 | Individual: 50.7)

Cluster: 3 tweets | Engagement: 2161 (Avg: 296) | Type: Tech

Vibe coding is gonna hit new levels of insanity\n\nQT @AIatMeta: We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.

We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.

🧵👇

See 2 related tweets

@MTSlive: SITUATION EXPLAINED: Meta just introduced Brain2Qwerty v2.

• Brain2Qwerty v2 is the highest-perform...

@omarsar0: Highly recommended reading.

What an impressive use of LLMs and deep learning.

Achieves "real-tim...

20. VadimStrizheus (Group Score: 107.7 | Individual: 36.8)

Cluster: 3 tweets | Engagement: 45 (Avg: 66) | Type: Tech

A one person $100K/month operation just became possible with one MCP:

find any business asking "why aren't we in ChatGPT"
that's your opening. run CrowdReply MCP.
it audits their AI visibility in minutes
finds every gap. writes the content. implements automatically.
charge $10K/month. close 10 clients. done.

no staff. no tools. no manual work.

just one MCP replacing what agencies charge $50K/year to do.

the category is brand new. the window is wide open.

bookmark this.👇\n\nQT @Crowdreply_io: Today we're introducing the CrowdReply MCP.

The first ever MCP that analyzes and ranks your website in AI search.

Simply talk to it and it'll find where you're missing, then goes in and handles the implementation. https://t.co/ijZIa0KIc0

See 2 related tweets

@shiri_shh: A $20K/month business just became a one person operation:

DM any business and ask "does your bra...

@cgtwts: No way this is real.

A $50K/month business with zero employees:

DM local businesses: “Do you sh...