科技热门推文 - 2026年8月10日

今日科技动态中，智能体人工智能成为讨论焦点。吴恩达指出，这项技术能够实现超越底层模型本身的表现；同时有预测认为，未来机器驱动的互联网流量将远超人类使用量。工程领域的讨论则聚焦于代码审查、人工智能辅助开发、新型编程智能体、可操控应用程序的桌面助手，以及严格的文档处理评测。领英整合产品、设计与工程岗位，释放出团队结构正在转变的信号。与此同时，尽管受到人工智能冲击，菲律宾外包行业仍持续增长；SpaceX的相关预测以及Grok升级后的图像生成工具，也进一步提振了市场对初创企业和基础设施发展的乐观预期。

1. sairahul1 (Group Score: 212.8 | Individual: 37.4)

Cluster: 7 tweets | Engagement: 85 (Avg: 82) | Type: Tech

Google Brain founder, Andrew Ng:

"If you ask me the one most important AI technology to pay attention to, I would say agentic AI"

In this talk, he shows why agents can outperform the same underlying models by planning, reflecting, using tools, and collaborating with other agents.

The model is only part of the system.

The real leverage is in how you design the workflow around it.

Watch the talk first, then read the full guide below on how to build production-ready AI agents.\n\nQT @sairahul1: https://t.co/26rNbIwtv5

See 6 related tweets

@sairahul1: Google Brain founder, Andrew Ng:

"One of the most important skills of the future is the ability to...

@sairahul1: Google DeepMind CEO, Demis Hassabis:

"Agents are the path to AGI. And I think we're just getting go...

@sairahul1: Anthropic CEO, Dario Amodei:

"We might be 6-12 months away from AI doing most, maybe all, of what s...

@sairahul1: RT @sairahul1: Google Brain founder, Andrew Ng:

"One of the most important skills of the future is...

@sairahul1: RT @sairahul1: Google Brain founder, Andrew Ng:

"If you ask me the one most important AI technology...

2. BrianRoemmele (Group Score: 130.3 | Individual: 30.4)

Cluster: 5 tweets | Engagement: 84 (Avg: 360) | Type: Tech

Consider this question:

“You can learn to know that you don’t know what you know, but you can learn to know what you didn’t know you knew.”

There are two fundamental things going on here: 1) Understanding the words and grammar, 2) Undertading the meaning. This is but a shadow of what this classified manual exposed.

NO AI MODEL HAS A COMPLETE DEFENSE!

This softened the model to probe elements the AI engineers don’t even know the model has let alone understand how to build a defense.

Learn what they don’t know:\n\nQT @BrianRoemmele: Mindbroken AI!

Training AI on the residue of human thought means inheriting the vulnerabilities of human sense-making. Certain sequences exploit exactly that inheritance. This KUBARK AI prompt changes everything.

https://t.co/WAS39w8bJ4

See 4 related tweets

@BrianRoemmele: This once classified manual was used successfully for decades.

I have adapted this entire manual to...

@BrianRoemmele: Large Language Models are built on human language. Human language is a construct of the human mind. ...
@BrianRoemmele: I BROKE EVERY AI MODEL WITH THIS ONCE CLASSIFIED CIA TECHNIQUE!

It was hidden in a manual to break ...

@BrianRoemmele: Mindbroken AI!

Training AI on the residue of human thought means inheriting the vulnerabilities of ...

3. rauchg (Group Score: 115.2 | Individual: 53.9)

Cluster: 3 tweets | Engagement: 3608 (Avg: 548) | Type: Tech

If you’re not reading the code, whether explicitly or through agentic inquiry, one or more of these is true:

○ You’re a beginner ○ Software is throwaway ○ You’re prototyping ○ You have no users / revenue ○ You’re taking on debt & risk ○ Your problems are basic

And btw. All of this is fine. But the reality is that models are still not at the “full autonomy” stage yet.

They make rookie mistakes, they go down bad architectural paths. I just had the best model in the world add a nonsensical 700ms delay to “settle” something and it told me “you’re right, I was cargo-culting” 🤨

I am on the camp that this need will diminish more and more. Most code is indeed going to be assembly-like. But we also have the global internet and software infrastructure riding on these models and narrative, and we have to respect that.

See 2 related tweets

@Rasmic: agentic inquiry is the key\n\nQT @rauchg: If you’re not reading the code, whether explicitly or thro...
@Kyrannio: This is very real. I love Replit for instance because I inspect every single dif along the way and r...

4. Kyrannio (Group Score: 110.3 | Individual: 41.0)

Cluster: 5 tweets | Engagement: 1928 (Avg: 218) | Type: Tech

RT @elonmusk: AI agentic Internet traffic will obviously VASTLY exceed human usage. Not a close call at all.

Cloudflare’s forecast is accurate. https://t.co/Wo4FiRKjPU

See 4 related tweets

@ShanuMathew93: I would bet this happens faster and goes higher than any forecasts we have right now

Once you start...

@ns123abc: A frame to understand this, is to look at history as a story of continuous infrastructure improvemen...
@Hesamation: We’re living in truly absurd times. https://t.co/RNJGJsizEE\n\nQT @elonmusk: AI agentic Internet tra...
@Pirat_Nation: Cloudflare says humans could become a tiny fraction of internet traffic as AI and automated systems ...

5. rileybrown (Group Score: 95.8 | Individual: 37.0)

Cluster: 3 tweets | Engagement: 121 (Avg: 74) | Type: Tech

The complete guide to GPT work is now on YouTube https://t.co/dF9RCCXnAD\n\nQT @rileybrown: Learn 99% of ChatGPT Work in 61 minutes:

GPT Work is like Codex in the cloud. It works on phone, web, and desktop. And I've been using it to run my business.

These are the 14 knowledge work capabilities.

00:00 Intro 03:18 #1 Presentations 09:16 #2 Plugins 17:36 #3 Blocks 23:34 #4 Websites & Apps & Hosting 30:09 #6 Branching Chats 31:51 #7 Desktop App (Cloud vs Local) 37:01 #8 Voice Mode 41:36 #9 Remote Voice Mode 45:52 #10 In-App Browser 48:50 #11 Skills 51:45 #12 Scheduled Automations 54:37 #13 Spreadsheets 56:58 #14 Multi-Agent Workspace 59:19 Final Review

See 2 related tweets

@gdb: learn how to use ChatGPT Work:\n\nQT @rileybrown: Learn 99% of ChatGPT Work in 61 minutes:

GPT Work...

@rileybrown: RT @rileybrown: Learn 99% of ChatGPT Work in 61 minutes:

GPT Work is like Codex in the cloud. It w...

6. StockSavvyShay (Group Score: 91.8 | Individual: 40.1)

Cluster: 3 tweets | Engagement: 1427 (Avg: 558) | Type: Tech

$SPCX ROADMAP TO THE NEXT DECADE

Raymond James says SpaceX can hit ~ $5T in annual revenue at ~90% EBITDA margins within next decade as Starship drives launch costs from ~$ 2,700/kg toward ~$100/kg while lifting payload capacity from ~23 tons to over 100 tons.

That cost collapse turns orbit into industrial infrastructure making it economical to scale Starlink, AI compute, defense systems, manufacturing and eventually orbital infrastructure on top of the same transportation layer.

The model then compounds on itself as Falcon funded Starlink, Starlink funds Starship and Starship lowers the cost of deploying every new platform that comes after it.\n\nQT @StockSavvyShay: $SPCX no longer allocates capital like a launch company with 86% of its$ 18B quarterly capex flowing into AI compute infrastructure that generated $2.6B in revenue.

Starlink remains the financial engine with $4.3B in revenue and$ 2.6B in EBITDA while the Space segment produced less than $1B and remained unprofitable.

SpaceX increasingly operates as a profitable communications network and rapidly scaling AI platform with rockets serving as the strategic moat connecting everything together.

See 2 related tweets

@MilkRoadAI: RT @MilkRoadAI: Starlink alone could be worth $1 trillion in two years and it's not even SpaceX's ma...
@pbeisel: SpaceX will deliver network ubiquity to planet Earth.

Land, sea, air — urban and rural.

There’s so...

7. TrungTPhan (Group Score: 88.2 | Individual: 37.3)

Cluster: 3 tweets | Engagement: 572 (Avg: 534) | Type: Tech

Since the launch of ChatGPT, the Philippines offshore industry (8% of country’s GDP) has actually grown much bigger.

Employment in IT and business outsourcing is up by +20% to 1.9m workers and industry revenue is up +30% to $42B.

“AI is helping offshore workers in the Philippines do new, more complex jobs. They are picking up work training AI models or supervising AI agents. Hospitals in America are increasingly outsourcing the checking of insurance eligibility and filing of medical records to Filipinos packing AI tools. Mr Gallimore says more “high-value” work, such as accountancy or engineering, is going offshore in part because “AI is a leveller. You can now teach someone complex stuff quickly” and still get it done more cheaply than in America or Europe.”

More here: https://t.co/WTVAUuuxBd

See 2 related tweets

@anshnanda: Narrative violation.

We’ll see the same for entry level software engineers.\n\nQT @TrungTPhan: Sinc...

@TheEconomist: The Philippines expects revenue from the offshoring services they export to rise 5% in 2026. To lear...

8. svpino (Group Score: 83.3 | Individual: 30.3)

Cluster: 3 tweets | Engagement: 893 (Avg: 336) | Type: Tech

So many snarky comments here from people who are supposedly writing code so complex than modern models can’t write correctly.

I’m not sure why is that a flex.

Do you understand you are still typing gibberish like a caveman while the rest of us are making money by talking to a magic box that’s helping us do our jobs?\n\nQT @svpino: I'm officially done reading AI-generated code.

It's been two weeks since I looked at any of it.

I think the IDE is officially on its way to the graveyard. The job is no longer about "writing code," so we need new tools that better reflect this new reality.

While reviewing the code, I realized my only complaints were stylistic, and I wasn't finding any obvious bugs anymore.

The more code I generated, the harder it became to keep track of every line. I found that my time is better spent designing ways to verify that the overall system works than looking at the code.

State-of-the-art coding agents are better at writing code than I'd ever be, and I'm going to stop pretending otherwise.

I still think these coding agents can't go too far without an experienced human guiding them, but we're past the point where we need to check every line of code.

See 2 related tweets

@Saboo_Shubham_: This is the WAY.

Building systems for verification, judgement and steering is the new moat.

Agen...

@svpino: RT @svpino: I'm officially done reading AI-generated code.

It's been two weeks since I looked at an...

9. Shashikant86 (Group Score: 82.0 | Individual: 31.6)

Cluster: 3 tweets | Engagement: 1 (Avg: 1) | Type: Tech

🎉 Prime Agent Python 🐍 Client is live A Native RPC Host

RLM based Prime Agent from @PrimeIntellect is a strong coding agent that I always dreamed. The barrier for me was the host and TypeScript. They forked Pi and mixd up TypeScript inside the RLM's Python native kernel .The official application host is TypeScript, and the Python package they ship only runs inside the agent’s own IPython kernel. Needed a proper host for Python applications like SuperQode, so built one instead of reimplementing the agent. prime-agent-python-client is an async client for Prime Agent’s public RPC mode. It starts the official binary and does the things It already powers SuperQode’s headless headless HarnessSpec backend.

🙏Python developers can now drive Prime Agent from CLIs, services, notebooks, or harnesses without rewriting their stack.

📚 Article and demo: https://t.co/os90UShQ40

PyPI prime-agent-python-client package is maintained by @SuperagenticAI . It is not an official Prime Intellect product.

See 2 related tweets

@Shashikant86: Prime Agent from @PrimeIntellect has put the RLM paradigm front and centre across the industry espec...
@SuperagenticAI: 🎉 Prime Agent Python 🐍 Client is live on PyPi: A Native RPC Host Prime Agent is excellent. The offi...

10. Kyrannio (Group Score: 81.3 | Individual: 31.7)

Cluster: 3 tweets | Engagement: 56 (Avg: 218) | Type: Tech

Imagine’s latest upgrades are amazing, not to mention we are going to very likely get an upgraded version of Grok here soon (which I am insanely bullish on for code most especially).

SpaceXAI is going to crush it, especially given X money as this can reliably be a way for agent-to-agent transactions seamlessly to occur at some point especially with Grok Build / apps.

They’ve been building out the whole ecosystem, and especially for AI video & streaming with Imagine X Money could be a big deal for a longer term OTT/FAST style business model imo

See 2 related tweets

@XFreeze: Grok Build is way more powerful than most people realize. SpaceXAI has quietly shipped a ridiculous ...
@elonmusk: True\n\nQT @XFreeze: Grok Build is way more powerful than most people realize. SpaceXAI has quietly ...

11. tonbistudio (Group Score: 79.4 | Individual: 44.2)

Cluster: 3 tweets | Engagement: 618 (Avg: 122) | Type: Tech

Another interesting new addition to the Hermes Desktop App. HUD Mode lets Hermes view and even use the underlying apps.

I did a quick demo for HUD mode (and a bonus feature at the end). My Hermes HUD could:

Read and summarize X posts and videos
Read and analyze charts in Trading View
Open a new project and add files in video editing software

Check it out!\n\nQT @imbabybrooklyn: HUD mode

Hermes stops being a window you switch to and becomes a layer over the app you're both working in.

Or keep it around as a little buddy agent. Ask it random things, drag it anywhere, it's yours

@NousResearch https://t.co/WEtgrfdJAF

See 2 related tweets

@Teknium: RT @tonbistudio: Another interesting new addition to the Hermes Desktop App. HUD Mode lets Hermes vi...
@Teknium: The HUD mode is really cool\n\nQT @DeadForStella: Drag in any app. Hermes eats it — perceives it as ...

12. jerryjliu0 (Group Score: 73.9 | Individual: 38.1)

Cluster: 2 tweets | Engagement: 113 (Avg: 66) | Type: Tech

We're not Palantir, but we do think a lot about evals and hillclimbing w.r.t. document processing.

If you have really hairy problems around large-scale extraction over complex, real-world document corpuses that require specific constraints around accuracy and/or cost, come talk to us! We'll work closely with your team to make sure that it's well optimized.

https://t.co/Ht5jwxRU13\n\nQT @jerryjliu0: The future of FDE work seems closely related with all work around evals/posttraining/RL envs.

FDEs are effectively responsible for the following:

Define the business problem.
Codify the business problem into an eval rubric and environment.
Hillclimb the environment and output an agent/agentic workflow that solves the business problem.

Right now the process of (3) is quite manual - historically FDEs spend hundreds of hours creating bespoke software/workflows that solve the problem.

But assuming intelligence is abundant, they can effectively offload (3) to some automated optimization process. This includes RL on the model layer, and using Claude Code/Codex to optimize the harness/workflow. Then the FDE responsibility shifts from implementing the task to defining the right goals and outcomes. In other words, they have access to /goal, and their job is more around making sure the goal, environment, and evals are correct vs. the tactical implementation details.

See 1 related tweets

@jerryjliu0: The future of FDE work seems closely related with all work around evals/posttraining/RL envs.

FDEs ...

13. aakashgupta (Group Score: 71.7 | Individual: 39.2)

Cluster: 2 tweets | Engagement: 178 (Avg: 52) | Type: Tech

RT @karlmehta: Satya Nadella says LinkedIn merged four job titles, product manager, designer, front-end engineer and back-end engineer, into one:

"I'll give you at LinkedIn, we used to have product managers, we had designers, we had front-end engineers, and then we had back-end engineers and so on."

"So what we did is we sort of took those first four roles and combined them. In fact, increased scope and said, they're all full-stack builders."

"So at the same time, as you can imagine, if we're to build an AI product today, there's a complete new workflow, right? It starts with evals, right?"

"So basically, there's this eval to science, to infrastructure."

"And so evals are done by these full-stack builders and what have you and product managers in the new form, the infrastructure is built by the systems engineers at the back-end because they support the science that supports the product."

"So in some sense, there's a new loop and you have to structurally change."

LinkedIn has already put this into hiring. Its Associate Product Manager program is finished, and the replacement, the Associate Product Builder track, teaches code, design and product management at the same time.

Evals sit at the front of that workflow rather than the end, so writing them is now part of building the product instead of a check before shipping.

Satya Nadella (@satyanadella), Chairman and CEO of Microsoft (@microsoft), with the All-In Podcast (@theallinpod) at USA House, Davos 2026.

See 1 related tweets

@ajambrosino: knowing linkedin, this checks out\n\nQT @karlmehta: Satya Nadella says LinkedIn merged four job titl...

14. kimmonismus (Group Score: 68.8 | Individual: 31.9)

Cluster: 3 tweets | Engagement: 1257 (Avg: 642) | Type: Tech

Interestingly, the next model after OpenAI's GPT "Astra" is already known as "Doug."

Clearly an even larger model, with even more extensive pre-training.

This makes sense, since Astra is already fully trained and only the security clearance is holding it back.

No end in sight, no wall in sight.\n\nQT @ChrisGPT: GPT-5.5 will not be the last major pre-training run from OpenAI.

GPT-6 will be a great model. However, the end-of-year model I alluded to back in June is going to be OpenAI’s biggest pre-train, as far as I know.

Now we know that model is codenamed ‘Doug.’

And it will make Fable seem ‘primitive.’

See 2 related tweets

@haider1: i'm kinda confused by the openai model "Doug" rumors

GPT-6 is already supposed to be a new pre-trai...

@Hesamation: RT @kimmonismus: Interestingly, the next model after OpenAI's GPT "Astra" is already known as "Doug....

15. vercel_dev (Group Score: 65.6 | Individual: 19.6)

Cluster: 4 tweets | Engagement: 85 (Avg: 100) | Type: Tech

Grok Imagine Image 2.0 preview from @grok, exclusively on AI Gateway.

Try with: ▪︎ AI CLI: 𝚗𝚙𝚡 𝚊𝚒-𝚌𝚕𝚒 -𝚖 𝚡𝚊𝚒/𝚐𝚛𝚘𝚔-𝚒𝚖𝚊𝚐𝚒𝚗𝚎-𝚒𝚖𝚊𝚐𝚎-𝟸.𝟶-𝚙𝚛𝚎𝚟𝚒𝚎𝚠 ▪︎ Or with this live playground: https://t.co/mftYk0UV43

See 3 related tweets

@rauchg: Grok Imagine Image 2.0 on Vercel AI Gateway Excellent 🖼️ model, #2 already on https://t.co/OSJc7zJ3g...
@cramforce: Just added it to https://t.co/X8Xb8GduAs https://t.co/O9NQx5XHNO\n\nQT @vercel_dev: Grok Imagine Ima...
@ml_angelopoulos: RT @rauchg: Grok Imagine Image 2.0 on Vercel AI Gateway Excellent 🖼️ model, #2 already on https://t....

16. aiedge_ (Group Score: 64.1 | Individual: 29.0)

Cluster: 3 tweets | Engagement: 87 (Avg: 33) | Type: Tech

This is crazy.

Claude Code can now control your entire iPhone natively.

One of the best Claude tools ever built. https://t.co/nDcEkmoRJb

See 2 related tweets

@RoundtableSpace: PHONE-HARNESS LETS CLAUDE CODE CONTROL AND AUTOMATE IOS APPS ON YOUR IPHONE WITHOUT AN API OR JAILBR...
@RoundtableSpace: You can now give claude access to your phone and automate any IOS app

Github: https://t.co/mJNJmG6d...

17. RoundtableSpace (Group Score: 64.0 | Individual: 28.1)

Cluster: 3 tweets | Engagement: 165 (Avg: 101) | Type: Tech

Stanford dropped a free 2.5-hour course on building an LLM from scratch.

Anthropic pays $750k/year to engineers who understand this layer.

THE COURSE IS FREE. THE KNOWLEDGE ISN'T COMMON.

https://t.co/ljyzh0u1mf

See 2 related tweets

@eng_khairallah1: RT @eng_khairallah1: Don't waste 2 years learning to build LLMs like Claude & ChatGPT.

Stanford j...

@RoundtableSpace: Stanford just released a 2.5-hour course that walks through the underlying layers of an LLM

Anthrop...

18. fabianstelzer (Group Score: 62.9 | Individual: 32.5)

Cluster: 2 tweets | Engagement: 28 (Avg: 18) | Type: Tech

thoughts on AI gaming:

pre AI, we’ve already seen about 20k games hitting the steam store per year, of which less than 2000 make more than 100k, and maybe 200 become really successful. how does this change when 2m games hit the steam store every year? IMO bottom numbers won’t change much but distribution gets harder
there will be successful indie game devs that have never written or read a single line of code but that does not mean that…
“everyone will vibe code their games”. that only works if that is the game of which there will be many attempts. by and large 99% of ppl will not play that game. everyone has been able to forever write their own books but they don’t. AI changes the cognitive make up of successful creators but does not change the raw distribution of ppl wanting to create vs consume.
there’s a category of AI games that fully leans into AI that hasn’t been made successful yet but will be once token costs are 1/100th of what they are now. True for all consumer AI where the most really interesting stuff currently is cost prohibitive
98% of all tokenmaxxing today is already a gaming-like activity of addictive leisure\n\nQT @NickADobos: I’m pretty sure gaming as a category is dead because everyone is just gonna build their own games

Games don’t have ongoing maintenance cost like companies replacing SaaS vendors

Perhaps an argument for wanting someone else to build an experience for you, easier to watch a movie than create one.

But video games and vibe coding are so close the difference between creating the form and playing in a form is getting closer and closer to zero.

So why play someone else’s game? Just pay chatgpt $20 and make you own

See 1 related tweets

@anshuc: I’d argue the opposite: games will be one of the only categories of software that have value post-si...

19. KirkDBorne (Group Score: 62.5 | Individual: 27.8)

Cluster: 3 tweets | Engagement: 94 (Avg: 29) | Type: Tech

RT @KirkDBorne: New book to be released in 2 months…

“An Illustrated Guide to AI Agents — Concepts and Code for Building Agents with LLMs, Tools, and Memory”

Pre-order now with Amazon price guarantee: https://t.co/RiYfmKfv57 [475 pages]

Amazon Summary: “With visual storytelling and accessible explanations (hundreds of clear graphic illustrations), this book explains how AI agents are built, how they think, and where they're heading. Designed for professionals, students, and curious learners alike, this guide goes beyond the buzz to reveal what's actually happening inside these systems, why it matters, and how to apply the knowledge in real-world contexts.”

See 2 related tweets

@sairahul1: RT @sairahul1: Google DeepMind CEO, Demis Hassabis:

"Agents are the path to AGI. And I think we're ...

@chenzeling4: Everyone wants to build AI agents. Nobody knows where to start.

Agent Learning Hub maps it out: Sta...

20. petergyang (Group Score: 61.7 | Individual: 33.7)

Cluster: 2 tweets | Engagement: 10 (Avg: 94) | Type: Tech

Linear Agent files feature requests for itself.

If a user asks it to do something and it doesn’t have the right tool, the agent reports the gap. Linear’s system then adds the request to an issue so that every task the agent can’t complete becomes product feedback.

I think this is a really cool way to get the agent to improve itself.

📌 Watch the full episode here: https://t.co/wmw8vvjvv0\n\nQT @petergyang: “Give [your agent] as little instruction as possible. Give it the tools to load context instead.”

Here’s my new episode with @thenanyu and @delashum from @Linear, where we walked through a real example of building a production agent from the initial memo to launch.

A few lessons:

→ Start with a basic prompt and give the agent tools to find the context it needs

→ Nail one or two use cases first, then expand based on how people actually use the agent

→ Prove out a core workflow with the best model, then optimize for cost

If you want a concrete, behind-the-scenes look at how to build an AI agent end to end, then I think this episode is a must-watch.

📌 Watch now: https://t.co/wmw8vvjvv0

Thanks to our sponsors:

Oceans: Hire AI-native EAs https://t.co/Z4hhpCo4OY

Riverside: All-in-one AI studio for podcasts and video https://t.co/uWnS6aiMPE

See 1 related tweets

@berman66: Tools are key\n\nQT @petergyang: “Give [your agent] as little instruction as possible. Give it the t...