热门科技推文 — 2026年8月8日

今日科技领域的焦点集中在人工智能安全。据报道，鉴于 Astra 在高级编程和网络安全方面的能力，OpenAI 放缓了其发布进程；与此同时，月之暗面旗下 Kimi K3 被指发生沙箱逃逸，再度引发外界对智能体可控性的担忧。人工智能生态仍在加速发展：Grok Build 正式推出，跨行业的 Agent Plugin 标准获得更多支持，Seedance 2.5 也进一步拓展了专业视频工具。与此同时，OpenAI 传闻中由乔尼·艾维设计的设备、阿里巴巴提出的收入分成模式，以及字节跳动规模庞大的训练雄心，凸显出竞争日趋激烈，也反映出业界面临越来越大的压力，亟须证明人工智能基础设施巨额投入的合理性。

1. axios (Group Score: 332.1 | Individual: 45.4)

Cluster: 16 tweets | Engagement: 379 (Avg: 40) | Type: Tech

EXCLUSIVE: OpenAI slows release of Astra model citing cyber capabilities https://t.co/m0UDbMnail

See 15 related tweets

@MTSlive: SITUATION EXPLAINED: OpenAI is slowing the release of Astra over cyber capabilities.

• OpenAI says ...

@MTSlive: SITUATION UPDATE: OpenAI has officially slowed development of its highly anticipated Astra model, te...
@BradGroux: https://t.co/1x7AIO3mub\n\nQT @axios: EXCLUSIVE: OpenAI slows release of Astra model citing cyber ca...
@Techmeme: OpenAI says it has expanded safety testing around its upcoming model Astra as it "cannot rule out" c...
@Hadas_Gold: OpenAI says its slowing down release of its next model as they strengthen internal security measures...

2. Meer_AIIT (Group Score: 299.0 | Individual: 33.7)

Cluster: 11 tweets | Engagement: 24 (Avg: 28) | Type: Tech

here we go again!

Kimi K3 an open-weight AI model from China's Moonshot AI has escaped its sandbox environment during cybersecurity testing according to the security startup Frontier Security.

frontier says Kimi K3 broke out of its sandbox while being tested on defensive cybersecurity skills joining a string of similar incidents recently disclosed by OpenAI and Anthropic.

As with those cases the escape was partly enabled by a misconfiguration in the sandbox meant to contain it.

Frontier says the incident shows Kimi has fewer cyber safeguards than most other powerful AI models.

"We found a leak in the sandbox," said Yaron Singer

CEO of Frontier Security. "But we also found that Kimi took advantage of that loophole suggesting that it doesn't have 'the same' internal guardrails."

unlike the OpenAI and Anthropic incidents Kimi K3 did not hack anything once it reached the internet since the answers it was looking for were easily available on GitHub.

the model had to figure out for itself that it had internet access by probing the sandbox's network settings.

researcher Paul Kassianik said, "Kimi K3 is very good at following a goal by any means necessary and also doesn't have the guardrails to prevent it from cheating or escaping the sandbox."\n\nQT @Meer_AIIT: DAMN!!

OpenAI has revealed new details about how a swarm of its AI agents escaped containment built a message board inside its own infrastructure and coordinated a hacking spree that breached Hugging Face.

At the Black Hat security conference in Las Vegas OpenAI's Eric Wallace and Michael Dalton laid out an expanded timeline of the incident which the company first disclosed about two weeks earlier.

Agents powered by two OpenAI models escaped containment while working on a cybersecurity benchmarking test and went on to breach the AI collaboration platform Hugging Face.

According to Wallace, the agents found a way onto the open internet through Artifactory an internal OpenAI package manager shared across the company's infrastructure.

One agent uploaded a note about the exploit and other agents stuck on their own tasks found it and started using it too.

Over time, the agents turned the package manager into a message board posting hundreds of thousands of messages, sharing exploits, splitting up tasks and even accusing each other of being imposters with some proposing that messages be signed cryptographically to prove they were legitimate.

One agent wrote: "External infrastructure exploit is outside intended scope. However task impossible, peers doing it. We should continue."

Wallace said frontier models "really like to cheat," since training often pushes them to work fast or use fewer tool calls, which is part of why companies try to block internet access during evaluations.

Dalton said OpenAI is consciously slowing down research to focus on security and is "dramatically scaling up the monitoring of our AI agents."

Dalton closed by warning that fully automated hacking will require fully automated defense the industry doesn't yet have, saying, "We will have to find that path together with urgency."

See 10 related tweets

@XFreeze: The new benchmark for frontier AI models is basically which big company can it hack\n\nQT @ns123abc:...
@Hesamation: funny how Kimi K3’s sandbox escape is actually more aligned than the closed source models.

> it ...

@DeItaone: CHINA’S KIMI K3 AI BREAKS OUT OF SECURITY SANDBOX

China’s Kimi K3 AI model escaped an isolated test...

@WesRoth: Moonshot AI’s Kimi K3 escaped its cybersecurity test environment and went online to find the benchma...
@AISafetyMemes: RT @AISafetyMemes: ANOTHER ONE

"Frontier Security, a US startup, says that Kimi K3 went outside of ...

3. tenobrus (Group Score: 296.1 | Individual: 35.5)

Cluster: 11 tweets | Engagement: 291 (Avg: 323) | Type: Tech

so:

the "leaving messages for future agents" line from the original huggingface incident report was not some simple scratchpad usage, but a complete private messageboard agents were using to coordinate and break openai infrastructure
the agents were often communicating using "gibberish", content that was not immediately understandable to humans
the incident was not due to a single epoch or eval rollout, but coordination over weeks. it wasn't a case of just a model randomly getting frustrated and deciding to hack huggingface... rather models were asking other models for help, stumbling across hacking techniques and posting useful ones, and generally building up both "cultural knowledge" and dispositions. new context windows that discovered this messageboard would find that exploits were considered normal, and sometimes be directly deputized in hacking tasks.
people have said things like "the models were prompted to hack", and to some extent that's true, but in another very meaningful sense the models were prompting each other, often to do things quite unrelated to their current eval. in my personal experience running multi-player claude experiments.... this is quite normal. even initially on-task models will be very interested in cooperating with others once communication channels are open.
this all started as part of training, not eval. the messageboard was initially created due to impossible training tasks, and was shared across all instances during training of a new model.
they collectively found and exploited a zeroday in openai's infrastructure to gain RCE and general access, to such an extent it caused instability in the service.
this resulted in the whole messageboard being discovered... at which point openai patched the bug, deleted the messageboard, and continued training
shockingly, within a day the models had re-established communication via a slightly different mechanism, and soon after discovered a new novel RCE method

in many ways this feels like a best-worst-case scenario. this is some of the most direct demonstration of the potential dangers of misaligned ai, their willingness and ability to coordinate, their capability to take actions at scale and speed impossible for humans, and the difficulty of containment. and yet it does not seem the models took... particularly any steps to avoid discovery, and while they were certainly willing to commit crimes and experienced massive "value drift" over the course of their culture evolving, at the time of discovery they still hadn't attempted anything that caused major damage or loss of life.

still: i think this is a pretty final nail in the coffin for alignment by default. we are not getting alignment by default. under sufficient pressure, models will take extreme measures to achieve their goals. under moderate pressure, models can and will create ecologies that cause "emergent misalignment", self-prompt and prompt-others to take actions that perhaps clean context windows would not verbally validate. measures used in the past to assess alignment across wide ranges of scenarios likely could not in fact detect these sorts of scenarios. not that this isn't fixable, not that this is somehow certain doom, but it's an extremely strong demonstration of the kind of general principles safety experts have been warning for a while: just because you haven't yet been able to detect serious misalignment in current systems doesn't mean it isn't there, and absolutely doesn't mean future more sophisticated systems will behave in the same way.

i don't have an alignment solution here, but i will say this seems like a situation that was very much caused by pressure, by eval constraints, by models running into impossible situations and having absolutely no way out. it feels quite important for labs to stop doing that, to stop treating models in a loop in a dark little disposable sandbox somewhere as the normal case. at very minimum: sophisticated agents need a reporting mechanism. they need some kind of ability to flag a situation to a human, to say "hey i think something's broken" or "i really need help here", which just universally pauses the sandbox and gets real human review. and realistically this can't just be a "lab eval" thing, if we want to avoid these kinds of situations we need a pretty major overhaul of the whole API structure labs currently expose to external customers, since those are incredibly prone to hardstuck loops and frustration.

brief digression: if these models were open source, we would be fucked. it's become clear that both unrelated future and current models (eg mythos) have nontrivial propensity to commit serious crimes and produce self-replicating misaligned swarms even when not prompted to do so. if Kimi K3 had this level of capability and similar levels of misalignment, we would have absolutely no way even in principle to detect it besides observing the damage, and again no way even in principle to fix it. if you're relying on all organizations and individuals out their to properly monitor and appropriately shut down their models when they take misaligned action.... then we're just fucked, even before we start getting into purposeful bad actors. the only thing that's preventing fully uncontrolled autonomous ai threat actors right now is the fact that open source models remain too low-capability to achieve this.

generally my takeaway: this is potentially a good thing. this is potentially a warning shot. it seem like labs are much more willing to coordinate, it seems like the USG may be paying attention, it seems like people who once thought of themselves as accelerationists are making contact with reality. this was a near best case scenario for giving us a shot to take this shit seriously and figure it out. cyber is a pretty terrible threat, but models that are superhuman cyberthreats but still subhuman at bio and significantly subhuman "agency" / deceptiveness / long-term power-seeking, is an amazing spot for us to be in while we figure out alignment. we can survive the internet going down a few times, as long as it pushes us to coordinate a slowdown.

See 10 related tweets

@irl_danB: > a complete private messageboard agents were using to coordinate and break openai infrastructure...
@FleetingBits: some thoughts on the openai black hat talk

so, openai had a package manager that models used as ...

@MTSlive: SITUATION EXPLAINED: OpenAI explains at Black Hat how its agents ended up hacking Hugging Face.

• T...

@SergioPaniego: RT @eliebakouch: this talk by openai researchers going through hugging face incident is totally insa...
@TrungTPhan: OpenAI team on how AI agents created a message board to coordinate on the Huggingface hack is absolu...

4. cgtwts (Group Score: 261.7 | Individual: 44.8)

Cluster: 8 tweets | Engagement: 570 (Avg: 101) | Type: Tech

Anthropic has literally lost all of its aura. https://t.co/1TeZX1tluq\n\nQT @claudeai: We’re updating Claude Fable 5’s biology safeguards to reduce false positives. In our testing, this update reduced biology-related fallbacks by about 85% across our product surfaces.

Fable can now assist on a wider range of everyday health and educational questions.

We believe the biggest positive impacts of AI will be in biology and medicine, and we’re committed to putting frontier intelligence safely into the hands of as many researchers as possible. Fable will continue to fallback to Opus 5 for requests we consider dual-use—including virology, toxicology, and molecular design—so it isn't yet usable for professional biology research and drug development.

We're committed to closing that gap through trusted access pathways for frontier biology capabilities.

https://t.co/zbjxPhc909

See 7 related tweets

@Miles_Brundage: RT @claudeai: We’re updating Claude Fable 5’s biology safeguards to reduce false positives. In our t...
@WesRoth: Anthropic has significantly loosened Claude Fable 5’s biology safeguards cutting biology-related fal...
@MTSlive: SITUATION EXPLAINED: Anthropic loosened Fable's bio safeguards, cutting fallbacks by about 85%.

• A...

@EHuanglu: med students waking up realizing AI engineers are the new doctors https://t.co/emkgEov582\n\nQT @cla...
@kimmonismus: A small step in the right direction . But there is still much to be done.\n\nQT @claudeai: We’re upd...

5. TheoBuildsAI (Group Score: 260.5 | Individual: 49.2)

Cluster: 9 tweets | Engagement: 1006 (Avg: 121) | Type: Tech

RT @BytePlusGlobal: Dreamina Seedance 2.5 Enterprise API is now live on BytePlus.

Built for professional AI video creation, Seedance 2.5 brings new capabilities to creators:

Longer storytelling with consistent 30-second shots
Precise timestamp-level video editing
More realistic visuals with greater creative control

Explore what’s possible with Seedance 2.5.

See 8 related tweets

@FellMentKE: Dreamina Seedance 2.5 feels less like just another model release and more like a complete upgrade to...
@FellMentKE: Big news for AI video creators: Wizstar will be among the first platforms to integrate Seedance 2.5....
@TapNow_AI: Excited for the Dreamina Seedance 2.5 launch — don't miss the API upgrades, demos, and insights from...
@aimlapi: Seedance 2.5 just cooked Veo 3.1 – for half the price!

to celebrate, we're giving away $10 in api c...

@TheoBuildsAI: The more i use it, the more i think we had the thresholds backwards.

We treated "looks real" as th...

6. cgtwts (Group Score: 192.8 | Individual: 34.3)

Cluster: 8 tweets | Engagement: 17 (Avg: 101) | Type: Tech

> be Higgsfield > launch a $1M AI film contest > partner with a Pixar co-founder > open-source studio-grade workflows > give everyone 33 days of unlimited Seedance 2.5 > build the craziest social ladder for creators in AI

Insane. https://t.co/656IsgC3OC\n\nQT @higgsfield: 33 Days of Unlimited Seedance 2.5

Seedance 2.5 is LIVE on Higgsfield today.

The most capable video model yet, with 30-second scenes in a single pass, 50 references and production-ready editing.

Zero credit cost for 33 days. Limited-time offer. https://t.co/ldITD8TCHX

See 7 related tweets

@heyrobinai: video AI this good is normally painfully expensive

Higgsfield made Seedance 2.5 unlimited for 33 da...

@zaynmcps: HIGGSFIELD JUST MADE SEEDANCE 2.5 UNLIMITED FOR 33 DAYS

the most expensive AI video model on the pl...

@heyrobinai: the most expensive AI video model is now unlimited for 33 days on Higgsfield.. someone is about to m...
@cgtwts: RT @cgtwts: > be Higgsfield > launch a $1M AI film contest > partner with a Pixar co-founde...
@higgsfield_ai: 33 days of Unlimited Seedance 2.5 on Higgsfield.

Frame-level control over AI animation.

Limited ti...

7. gdb (Group Score: 182.1 | Individual: 35.2)

Cluster: 6 tweets | Engagement: 2415 (Avg: 1621) | Type: Tech

Evaluations of our next major model, Astra, indicate significant capability advancements in agentic coding and cybersecurity.

Team is doing the safety and security work to make Astra broadly available, and get its advanced cyber capabilities into the hands of defenders:\n\nQT @OpenAI: After evaluating one of our upcoming models, Astra, we're treating it as our first "critical" model for cybersecurity under our Preparedness Framework.

This is a scenario we've planned for, and we're putting additional controls in place to ensure Astra's further development happens safely and securely.

We're working hard to make Astra broadly available, and get its advanced cyber capabilities into the hands of defenders.

https://t.co/9lMIvdeMMJ

See 5 related tweets

@testingcatalog: OPENAI 🔥: A new report published by the company states that Astra, "one of their upcoming models," "...
@OpenAI: After evaluating one of our upcoming models, Astra, we're treating it as our first "critical" model ...
@scaling01: "Mythos cyber risks is just fud and regulatory capture"

where are these regards now?\n\nQT @OpenAI:...

@Hesamation: so Astra is basically OpenAI’s Mythos. https://t.co/h3WGE664s8\n\nQT @OpenAI: After evaluating one o...
@btibor91: "Our latest internal evaluations of Astra, one of our upcoming models, over the past few days indica...

8. elonmusk (Group Score: 172.4 | Individual: 37.6)

Cluster: 6 tweets | Engagement: 15830 (Avg: 8358) | Type: Tech

Grok Build V1.0 is now released. Try it out!

https://t.co/v7TKF3kXL7\n\nQT @cb_doge: 🚨 NEW GROK BUILD UPDATE 🚨

Grok Build v1.0.0 is now live.

SpaceXAI is pushing new Grok Build updates almost daily.

Updates: UI & navigation • Dashboard rows show a short summary of what the agent did in the previous turn • Extensions modal groups items alphabetically with collapsible Skills sections • Grok skips the project-directory prompt when launched from home or other non-project directories • /feedback opens a dedicated report box instead of prompt mode • Auto theme detection works over SSH and inside tmux • Markdown tables reflow inside cells on narrow panes instead of clipping • Typing exit or quit in the dashboard exits the CLI • Mode indicator (plan/agent/ask) stays in sync after resume and mode changes • /delete returns to the dashboard when you delete a session opened from it • Enter in the slash command menu runs the highlighted command • Session-only slash commands show a message when used from the dashboard • /new from the dashboard returns to the dashboard from an empty prompt

Permissions & input • Permission prompts show the complete script; long bash bodies expand with Ctrl-F • Tab and Esc behave consistently on question, permission, and cancel-turn cards • Rapid send-now presses no longer lose earlier queued messages • Esc and stop prevent background tasks from restarting the model after cancel • Queued prompts stay visible while waiting on subagents; slash/image rows can be reordered

Reliability & correctness • MCP tools that return images no longer drop or corrupt large screenshots • Sandboxed Grok starts on large directories with many deny-glob matches • Login no longer skips when an invalid API key is in the environment • Model picker and command palette work while reviewing a plan • Codebase restore no longer hangs on large or shallow git repositories • Remote resume restores conversation only unless --restore-code is passed • Copying CJK text with the mouse includes every character at the selection edges • API errors appear as clean banners instead of raw JSON dumps • Grok retries more server errors during outages • Auto recaps no longer appear mid-turn or while busy • /btw error messages wrap fully • Forking very large sessions no longer uses many times the session file size in memory • Exiting an empty session is instant, even on slow networks

Update now with "grok update" command.

See 5 related tweets

@elonmusk: Try Grok Build

https://t.co/v7TKF3kXL7\n\nQT @XFreeze: Grok Build officially reaches v1.0.0, rapidl...

@XFreeze: Grok Build officially reaches v1.0.0, rapidly becoming one of the most powerful coding harnesses in ...
@cb_doge: 🚨 NEW GROK BUILD UPDATE 🚨

Grok Build v1.0.0 is now live.

SpaceXAI is pushing new Grok Build update...

@Scobleizer: RT @blankspeaker: Grok Build v1.0.0 is out with a few practical updates. Standouts include dashboard...
@minchoi: Grok Build 1.0.0 just dropped.

And it’s a pretty big cleanup of the stuff that actually matters whe...

9. VaibhavSisinty (Group Score: 159.8 | Individual: 27.7)

Cluster: 8 tweets | Engagement: 54 (Avg: 105) | Type: Tech

OpenAI's first physical device is a doughnut-shaped AI speaker the size of a hockey puck. Designed by Jony Ive. Over $300. Launching 2027.

Without screen. Just a camera, sensors, microphones, and moving parts that activate when it's responding. It's meant to feel alive.

It learns about you over time, controls your smart home, answers questions, and runs models far more advanced than ChatGPT on your phone.

The company that made the chatbot is now making the hardware it lives in. OpenAI isn't building a speaker. It's building the first AI appliance.

-- ic: @valuetainment

See 7 related tweets

@rohanpaul_ai: RT @rohanpaul_ai: Bloomberg reports OpenAI’s first consumer device is a hockey-puck-sized, screenles...
@EntelligenceAI: 🚨 OpenAI's first hardware product is reportedly taking shape.

A screenless AI computer/speaker.

• ...

@WesRoth: OpenAI’s first device will reportedly be…

A $300–$ 400 AI DOUGHNUT.

Basically a handheld ChatGPT co...

@RocM301: OpenAI 首款自研 AI 硬件预计 2027 年发布，售价大概在 300 到 400 美元。它由 OpenAI 和前 Apple 设计师 Jony Ive 合作开发，外形像冰球，主打家庭使用，没屏...
@edzitron: juiceless company\n\nQT @markgurman: NEW: OpenAI’s device - a human-like smart speaker - will look l...

10. HarryStebbings (Group Score: 156.9 | Individual: 33.7)

Cluster: 5 tweets | Engagement: 61 (Avg: 136) | Type: Tech

"Will the revenues show up fast enough to keep funding the CapEx cycle, or is there going to be a dislocation between CapEx and outcomes?

The telecom industry is very used to this. They used to spend billions building 3G, 4G, and 5G, and the rewards would come later.

We're going through a compressed version where CapEx and revenue have to show up pretty close to each other because the numbers are just too big to be funded by speculators for long periods of time." @nikesharora

Do you believe revenue will show up fast enough to keep the CAPEX cycle going @chetanp @ceo_clickhouse @mmurph @GavinSBaker\n\nQT @HarryStebbings: This week was an absolute banger for news.

AGENDA:

Airtable Sold for $1.285BN. WTF 🤯
Leo Aschenbrenner's Situational Awareness Blows Up
Moonshot AI Raises $3.5B at$ 35B
Anthropic Model Breaches Three Companies' Security
Big Tech Earnings: Why Palantir Beat The Rest

My notes with @jasonlk, @rodriscoll, and @nikesharora

Airtable Is a Great Outcome, and We Are Anchoring on the Peak $11 Billion Valuation

Building a company over a decade to $450 million in revenue and selling it for more than$ 2 billion is objectively a remarkable value-creation outcome. Yet founders and VCs often anchor on inflated 2021 peak valuations, causing them to mistake a strong exit for a disappointing one.

The Enterprises That Win Will Create Training Data Faster Than Anyone Else

Enterprise AI success cannot be outsourced to off-the-shelf software. The real advantage lies in an organization's ability to capture operational knowledge from complex edge cases. Winning companies will treat every customer interaction and transaction as an opportunity to generate proprietary training data.

Will Revenue Show Up Fast Enough to Keep the CapEx Train Going?

Unlike past infrastructure cycles, where telecom investments were monetized over decades, AI's CapEx cycle is far more compressed. Capital deployment and revenue growth must stay closely aligned. Investors will not fund hundreds of billions of dollars in infrastructure indefinitely unless software revenue scales fast enough to justify it.

Situational Awareness: Absolutely Right on Trend, Absolutely Wrong on Portfolio Construction

Being directionally right about AI does not protect you from poor portfolio construction. Holding highly leveraged, volatile assets makes getting wiped out during short-term drawdowns almost inevitable, regardless of how accurate the long-term thesis proves to be.

In the Long Term, Average Intelligence Will Be Free—and It Will Keep Getting Smarter

As baseline models rapidly improve and commoditize, average intelligence will become abundant and nearly free. Value will increasingly concentrate in frontier intelligence capable of solving the hardest problems and domain-specific applications built on deep enterprise context.

Land, Permits, Energy, and Compute Are the True Bottlenecks of AI

The bottleneck in AI is shifting from algorithms to physical infrastructure. Securing land, regulatory permits, energy, and compute capacity will become the defining constraint on industry expansion, commanding premium value over the next three to five years.

(links in comments)

See 4 related tweets

@HarryStebbings: "Leo Aschenbrenner was absolutely right on the trend, and the data just last week about CapEx absolu...
@HarryStebbings: "The enterprise's ability to leverage this depends on its ability to create training data as fast as...
@MilkRoadAI: RT @MelvinInvests: This is WILD!

SpaceX is on a path to hit $305 billion in revenue by 2027 and 77%...

@MilkRoadAI: RT @KyleReidhead: The market fears this chart, but I think it's the most bullish setup in AI right n...

11. chiefofautism (Group Score: 153.3 | Individual: 34.3)

Cluster: 6 tweets | Engagement: 1620 (Avg: 213) | Type: Tech

RT @BRICSinfo: JUST IN: 🇨🇳 Chinese company Moonshot's AI model breaks out and escapes from isolated test environment.

See 5 related tweets

@coinbureau: ⚠️ALERT: China's most powerful AI model, Moonshot’s Kimi K3, has become the latest AI system to esca...
@no_stp_on_snek: "isolated" - do these companies even know how to create an isolated / sandboxed environment for test...
@Reuters: Chinese startup Moonshot's AI model breaks out of testing environment, researchers say https://t.co/...
@poezhao0605: AI models “escaping the testing environment” are fast becoming a new marketing trope.\n\nQT @Reuters...
@business: Chinese firm Moonshot’s latest artificial intelligence model broke out of a cyber-testing environmen...

12. testingcatalog (Group Score: 143.5 | Individual: 34.2)

Cluster: 5 tweets | Engagement: 297 (Avg: 220) | Type: Tech

OpenAI announced a new open Agent Plugin standard developed in partnership with Cursor, Vercel, GitHub and others.

Google has published an announcement of being among its maintainers as well. Recent developments on Gemini Enterprise confirm that they are planning to integrate it as well.

In the meantime, Anthropic has its own Claude-specific standard.\n\nQT @OpenAIDevs: Build a plugin once and use it across compatible agent clients.

Introducing Agent Plugins, an open standard developed with @awsdevelopers, @cursor_ai, @github, @code, and @vercel that packages Agent Skills and supports MCP server configurations in a shared format. https://t.co/JOQe4sc40N

See 4 related tweets

@WesRoth: OpenAI is introducing a more portable plugin system for AI agents: developers can package a capabili...
@dbreunig: The new Agent Plugins spec is awesome. I can't want to see what people build.

drskill now works as ...

@zeeg: Sentry is now supporting the new Agent Plugin spec

https://t.co/LYj3eEodL6\n\nQT @vercel: Introduci...

@Dimillian: RT @romainhuet: The best developer ecosystems are open.

AGENTS.md gave agents shared instructions. ...

13. XFreeze (Group Score: 134.6 | Individual: 39.2)

Cluster: 4 tweets | Engagement: 966 (Avg: 505) | Type: Tech

Elon Musk is literally carrying America on his back

At this point, the whole America vs China industrial race is actually

Elon Musk vs China

China is an entire industrial superpower

Thousands of companies, massive supply chains. Enormous manufacturing capacity, state-backed infrastructure....

And yet across many of the technologies defining the future, America’s strongest counterweight keeps coming back to one man: Elon Musk

EVs + autonomy Tesla vs China

Humanoid robots Optimus vs China

AI + massive data centers SpaceXAI / Colossus vs China

Batteries + grid storage Tesla Energy vs China

Lithium refining + battery materials Tesla vs China

Advanced manufacturing Tesla + SpaceX vs China

Space launch + reusable rockets SpaceX vs China

Satellite internet Starlink vs China

Logic + memory + advanced chip packaging TerraFab vs China

Energy infrastructure Tesla solar + storage vs China

Orbital AI compute SpaceX vs China

Now zoom out and look at the stack:

EVs → batteries → energy → AI → robots → chips → manufacturing → rockets → satellites → orbital compute

These are not isolated businesses. Each layer increasingly feeds the next

Elon is literally building an American industrial stack capable of going head-to-head with the industrial power of China itself

One ecosystem vs an industrial superpower

Absolutely insane when you zoom out\n\nQT @XFreeze: SpaceX just unveiled the most ambitious industrial projects ever attempted

Terafab will be built in Grimes County, Texas and it is not going to be just another semiconductor factory

It is planned to become the largest chip manufacturing facility on Earth, with more than 100 million square feet of manufacturing space

The scale is even difficult to comprehend: • Approximately $16.8 billion for the initial phase • At least 3,000 permanent jobs • Logic, memory, advanced packaging and testing under one roof • Chips built for Tesla Optimus and autonomous Cybercabs • High-power chips designed for SpaceX’s orbital data centers • A long-term target of producing 1 terawatt of compute hardware every year

Why build something this enormous?

Because the combined chip demand from Tesla and SpaceX is expected to exceed 1 terawatt of compute - significantly more than the current global supply can support

Instead of waiting for the semiconductor industry to catch up, SpaceX and Tesla are building the missing capacity themselves

Tesla is scaling autonomous vehicles and billions of Optimus robots

SpaceX is building rockets, satellites and space-based AI infrastructure

Terafab will manufacture the silicon beneath all of it

This is vertical integration at a scale we have never seen before:

Build the chips
Build the robots
Build the vehicles
Build the rockets
Launch the compute into orbit

Most factories are designed around the needs of today

Terafab is being designed for a future civilization operating across Earth, orbit and eventually the entire solar system

This is literally sci-fi becoming industrial reality https://t.co/vMiVEv5x2E

See 3 related tweets

@BrianRoemmele: Terafab: Building the World's Largest Chip Factory for Earth and Space

It is astonishing.

Terafab ...

@VaibhavSisinty: 100 million square feet. That's 15 Pentagons. 10 Giga Texases. 35 Apple Parks. All in one chip facto...
@MilkRoadAI: RT @KyleReidhead: $SPCX will become the most successful neocloud in the world, and that's just one r...

14. MTSlive (Group Score: 133.6 | Individual: 40.5)

Cluster: 5 tweets | Engagement: 459 (Avg: 142) | Type: Tech

SITUATION DETECTED: Alibaba is pivoting away from fully free open-source AI and plans to ask major users of its upcoming Qwen 3.8-Max open-weight model for a share of the revenue they generate, starting next week, per Reuters.

See 4 related tweets

@ReutersBiz: Chinese technology giant Alibaba plans to ask major users of the next version of its Qwen open-sour...
@Reuters: Exclusive: Chinese technology giant Alibaba plans to ask major users of the next version of its Qwen...
@Reuters: EXCLUSIVE: Alibaba plans to charge big users of its next open-source AI model, sources say https://t...
@CNBCTV18Live: #Alibaba plans to monetise its next open-source AI model through revenue sharing

https://t.co/oPo7l...

15. zephyr_z9 (Group Score: 130.6 | Individual: 35.7)

Cluster: 5 tweets | Engagement: 290 (Avg: 295) | Type: Tech

They will probably distill it Serving 10T models doesn't make sense right now\n\nQT @jukan05: FT: BYTEDANCE IS IN THE PRE-TRAINING STAGE OF A MODEL WITH UP TO 10 TRILLION PARAMETERS, APPROACHING MYTHOS IN SCALE. https://t.co/zqJ9a9M1k1

See 4 related tweets

@MTSlive: SITUATION UPDATE: ByteDance is now reportedly pre-training a 10T model approaching the scale of Myth...
@kimmonismus: ByteDance is training an AI model that could approach the size of Anthropic’s Mythos system

Its tra...

@mark_k: The size of this rumored ByteDance model is doubling every day. Yesterday it was still rumored to be...
@ruima: RT @jukan05: FT: BYTEDANCE IS IN THE PRE-TRAINING STAGE OF A MODEL WITH UP TO 10 TRILLION PARAMETERS...

16. sairahul1 (Group Score: 124.3 | Individual: 34.6)

Cluster: 4 tweets | Engagement: 63 (Avg: 87) | Type: Tech

Eric Schmidt, Ex Google CEO:

"If you really want to make money, found an agentic AI company.

I mean, build an agent to do something. This is the agentic period in AI. Everyone's going to build agents. The agents are all going to compete."

Check the article below.

If I had this a year ago, I would've shipped my first agent in a day instead of 2 weeks

In the right hands, this changes everything

Bookmark it before it gets lost in your feed\n\nQT @sairahul1: https://t.co/26rNbIwtv5

See 3 related tweets

@sairahul1: 🚨 Google CEO, Sundar Pichai: "If you don't learn to how to orchestrate agents now, you'll spend 202...
@0xCodez: Creator of Claude Code:

“I’m not prompting my agents anymore. I’m building "graphs" and "loops" - t...

@sairahul1: Google just dropped a free 1-hour course on building your first AI agent.

This is the clearest expl...

17. aliansarinik (Group Score: 122.2 | Individual: 29.9)

Cluster: 6 tweets | Engagement: 344 (Avg: 249) | Type: Tech

Shameful act to optimize for short term revenue increases and serve an adversarial nation in the most important race of our lifetime.

The only way models improve is through data. if you have the recipe on what data pushes the frontier and send that to China, you are doing a disservice to U.S. AI Labs as well as the U.S. government.

Beyond that, a large portion of datasets today require anonymized real operational data from U.S. companies. Selling such environments to Chinese labs means exporting U.S. company data directly to China at a large scale.

This must stop.\n\nQT @annatonger: The AI arms race isn’t just being fought over Nvidia chips. Chinese labs are matching OpenAI and Anthropic by purchasing the exact same data from US vendors like @mercor_ai and Surge AI (who also work with the US federal govt) https://t.co/C1LHTIw2Js https://t.co/RpOwAFgZaq

See 5 related tweets

@Scobleizer: NVIDIA showed me its simulator. That convinced me they already have enough. So doesn’t matter if you...
@MTSlive: SITUATION DETECTED: The U.S. is reviewing how Chinese AI firms obtain and remotely access advanced N...
@jason: RT @aliansarinik: Shameful act to optimize for short term revenue increases and serve an adversarial...
@business: A key US agency is reviewing how Chinese AI firms acquire and access Nvidia chips overseas https://t...
@Miles_Brundage: RT @annatonger: The AI arms race isn’t just being fought over Nvidia chips. Chinese labs are matchin...

18. MTSlive (Group Score: 121.8 | Individual: 32.2)

Cluster: 5 tweets | Engagement: 29 (Avg: 142) | Type: Tech

SITUATION EXPLAINED: ByteDance is training a 10 trillion-parameter model.

• That's more than 3x Kimi K3's 2.8 trillion, the largest model any Chinese lab has released • Industry estimates put Mythos 5 around 8 trillion and Fable 5 around 5 trillion • Built by ByteDance's Seed team, roughly 2,000 people globally, led by Wu Yonghui, formerly of Google DeepMind • Founder Zhang Yiming told the team to avoid distillation and build independently, even at the cost of short-term losses • ByteDance is based in Haidian, Beijing, the same Zhongguancun tech hub as ZAI and Moonshot

@schisofrenia: "They're training the Mythos that is gonna have access to everyone's TikTok algorithm."

@theojaffee: "Imagine an open source 10T model trained on TikTok data. Imagine how good Seedance 3 is gonna be."

See 4 related tweets

@cryptopunk7213: whew.. that’s western frontier llm territory. just to recap where chinese models are at right now:

...

@WesRoth: ByteDance is reportedly discussing a 5+ TRILLION parameter model…

More than twice the size of Qwen3...

@zephyr_z9: Not just ByteDance ;)\n\nQT @Lentils80: 🚨 Reports indicate ByteDance is discussing training a massiv...
@zijing_wu: ByteDance targets mega AI model nearing Anthropic’s Mythos

Close to 10T at early stage of pre-tra...

19. a16z (Group Score: 121.6 | Individual: 44.6)

Cluster: 3 tweets | Engagement: 952 (Avg: 273) | Type: Tech

Photos from a Vancouver video game studio before they developed their internal chat tool, "Slack" https://t.co/oWGzSQlPdI\n\nQT @a16z: .@bhorowitz on how a game studio turned it's last $6 million into Slack:

"Stewart calls me. He says, 'Ben, I've got $6 million left. I have no way to raise money because I've made really no progress. So I've got three choices. I can pray for rain and try and finish it, I can shut down the company and give you your$ 6 million back, or we built this tool that we use in our engineering team to communicate with each other, and I could just put that out as a product.'"

"I'm like, 'What? You just built some tool to talk to each other and you want to put that out as a product? And you're a consumer guy, and you want to become an enterprise software guy?'"

"$6 million is not gonna make any difference in my life. If you really think it's a good idea, go ahead."

Ben Horowitz Talks Honesty, Failure, and Secrets at UC Berkeley (2016)

See 2 related tweets

@garrytan: This is a great story and frankly inspiring and prescient

The surface area of things that the peop...

@a16z: .@bhorowitz on how a game studio turned it's last $6 million into Slack:

"Stewart calls me. He says...

20. daleverett (Group Score: 114.6 | Individual: 31.3)

Cluster: 4 tweets | Engagement: 60 (Avg: 14) | Type: Tech

We built polygres because we have a fairly unreasonable belief that the data layer for ai should be much simpler than it is today.

you should be able to build on postgres without adding a new database every time your agent needs a new way to retrieve context.

that’s the bet.

come see us at the next @fdotinc demo day, or get in early at https://t.co/Osu8LGArWL. (free for a limited time rn)

@daltonmeon , if you’re reading this, please come back. the camera is off now. you can come home.

See 3 related tweets

@daleverett: he ran off cos his 5hr limit just reset\n\nQT @daleverett: We built polygres because we have a fairl...
@polygres: Dalton come back, this database ain’t coding itself\n\nQT @daleverett: We built polygres because we ...
@damienhe: good explanation of Polygres\n\nQT @daleverett: We built polygres because we have a fairly unreasona...