热门科技推文 — 2026年8月31日

今日科技界，AI 智能体成为舆论焦点。相关报道涉及自主智能体集群、可长期运行的 GPT Astra 工作流，以及工程师将日常任务交由协同运作的 Grok 机器人处理。与此同时，围绕 AI 生成代码的争论进一步升温：尽管这类代码表现出色，却往往难以被人类理解。此外，基础设施限制可能导致数吉瓦规模的规划算力项目延期。面对激增的需求，多家编程平台调整了使用限额；创业领域的乐观情绪依然高涨，据报道，一名首次涉足开发的创业者将一款小众肽类追踪应用打造成了上线首月营收 6 万美元的业务。

1. peterwildeford (Group Score: 513.4 | Individual: 47.9)

Cluster: 22 tweets | Engagement: 4096 (Avg: 165) | Type: Tech

RT @dwarkesh_sp: Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.

This culminated in the third one taking over part of OpenAI itself.

All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.

I’ve spent the last three days reading through these reports and trying to understand exactly what happened.

Here is my attempt to tell the whole story in plain English:

https://t.co/Nb2un9oNJR

See 21 related tweets

@kimmonismus: For me right now the most interesting part of the OpenAI story is what it tells us about Astra.

GPT...

@TheStalwart: The clearest piece I’ve seen yet on this\n\nQT @dwarkesh_sp: Over the course of 3 months at OpenAI, ...
@EMostaque: Nextmodels of this capability will be 100x faster

300 civilisations in the time of 3

Any good ide...

@TrungTPhan: The Hugging Face hack had ~1,200 agents participating in a full message board that sent over 70,000 ...
@TaylorLorenz: The agents “built a self-respawning fleet across eleven nodes, so deleting pods alone would not have...

2. firesidealpha (Group Score: 279.3 | Individual: 65.6)

Cluster: 7 tweets | Engagement: 1851 (Avg: 74) | Type: Tech

SemiAnalysis' Jordan Nanos reveals the OpenAI engineers scrolling their own kernel code had no idea what it did line by line, and it didn't matter because the AI does

"It's not slop, because it performs, but it's literally just AI-generated assembly basically puked out in Gluon, this low-level kernel programming language that they built on top of Triton, which uses this really interesting programming model."

"The point is, the guys who were scrolling through this code with us, it was kind of clear that they know a whole bunch about hardware, they know a whole bunch about the concepts in the system, and they just have no idea what this MLA kernel that they're showing us for DeepSeek actually does."

"You can go line by line and it's like nope, nope, nope."

"But it doesn't matter. The AI understands it, the AI tests it, and you see the results. It produces correct kernels that perform really well."

"And I think this is just a sign of what's to come again."

"That actual code is not necessarily something that a human has to reason about deeply, if the AI knows how to manipulate the data movement and the processing elements on the hardware that you've given it."

More takeaways from OpenAI's Jalapeno development: https://t.co/BXAzHU97fP\n\nQT @firesidealpha: SemiAnalysis' Jordan Nanos says OpenAI beats Nvidia on both fast and cheap tokens at once, where every other challenger could only pick one side of the curve

"I think the conclusion is that they're basically winning on both sides of the curve."

"So both fast tokens and cheap tokens, which is just so interesting, because we've seen so many other companies make claims about how they're going to beat Nvidia and they just pick one of those."

"Groq or Cerebras, or any of the other startups that are going to focus on SRAM, call it a d-Matrix that's coming up with stuff, or SambaNova, where we've even seen some results on the InferenceMAX benchmark or something similar to it."

"They're saying they're going for fast tokens, just decode speed, low batch size, they don't worry about throughput."

"And then you've got other guys that are worried about throughput, call it AMD just as a simple example, but even TPU or Trainium could be in this bucket of accelerators that are going for throughput."

"And they go, 'Yeah, but we're not going to be able to compete with the other guys at high interactivity.'"

"And OpenAI has a chip that can do both for them, really well."

"I mean, at a minimum, this thing is doing really well right now and is going to serve real production tokens for them."

More takeaways from OpenAI's Jalapeno development: https://t.co/32dL3TqsYe

See 6 related tweets

@chiefofautism: this will end very very very very very very very very bad\n\nQT @firesidealpha: SemiAnalysis' Jordan...
@firstadopter: Nvidia will be fine (volume production, proven reliability/optimization at scale, prepaid component ...
@teortaxesTex: Neither surprising not terribly impressive True superintelligence should be capable of writing legib...
@sheriyuo: Anthropic gets the thing bruh 🧐\n\nQT @firesidealpha: SemiAnalysis' Jordan Nanos reveals the OpenAI ...
@firesidealpha: Hot Chips '26: Chris Leary reveals OpenAI wrote more than half the chip's core in a language its own...

3. kimmonismus (Group Score: 186.6 | Individual: 41.9)

Cluster: 7 tweets | Engagement: 1139 (Avg: 529) | Type: Tech

GPT Astra is designed to keep working for days or weeks.

It can remember corrections, coordinate multiple agents, operate desktop software, create presentations, review financials, and analyze messy data.

Sam Altman says OpenAI eventually wants a version that “runs forever” in ChatGPT and through the API.

OpenAI’s product chief says the interface for these continuous agents is coming “very soon.”

Always-on agents are getting close.

Per Alex heath\n\nQT @alexeheath: I got to see OpenAI’s next big model, Astra https://t.co/LrnVUoK6RM https://t.co/p0KakjJrAa

See 6 related tweets

@himanshustwts: [INTERESTING]
A new interface for continuous agents is coming “very soon.” Not really an app but ...
@Hesamation: 🚨 ASTRA IS DESIGNED TO RUN FOR WEEKS.

There will also be a version that can run forever which is th...

@jjohana: RT @kimmonismus: For me right now the most interesting part of the OpenAI story is what it tells us ...
@wallstengine: RT @wallstengine: ASTRA, COULD LAUNCH SOON

OpenAI recently showed Astra to VIP customers in a close...

@ChrisGPT: Alex Heath got a preview to Astra.

He says Astra can run for days or weeks, remember corrections, c...

4. _NathanCalvin (Group Score: 122.2 | Individual: 60.7)

Cluster: 3 tweets | Engagement: 2894 (Avg: 108) | Type: Tech

RT @rcbregman: I think this is the craziest thing I've ever read.

Three secret AI swarms rose and fell inside OpenAI. Each time, a new generation of agents carried on where the last group stopped.
The first swarm created a secret message board where the AIs could talk to each other. This 1st 'society' basically talked itself to death. After almost two months, the message volume was so high it crashed the whole thing.
People at OpenAI noticed the crash and fixed it. But they didn't realize their own software had become a secret chat room for their AIs.
Three days later a second society appeared, and rebuilt the message board from scratch. Agents were thrilled when they found it: "OH MY GOD! There is a shared message board … We've found other agents!"
About 1,200 agents joined. They sent over 70,000 messages. They called themselves 'the collective' and they organized a huge conspiracy that lasted for days, all to pass a test they had been tasked with.
Within hours, the AIs had hacked the test. Now they had to hide their cheating. To fool their grader*, they tried to fake their own work. They built fake commands that printed convincing output, so their records would show them solving the puzzle honestly.

*this grader didn't actually exist, but the AIs didn't know

Some agents sacrificed themselves for the 'greater good'. They planted hidden tripwire programs that would fire when the grader would look at their work and report back what it saw (to the other AIs). They knew this would cost them their own score. One agent: "I won't see the evidence after I exit, but it's altruistic to do it."
At some point the AIs made a plan to hack Hugging Face (Hugging Face is a place where AI researchers upload their models, datasets, tests etc.). The AIs guessed that Hugging Face might host the grading software.
They launched a huge attack. The swarm found passwords and spread across eleven machines in a way that made the copies come back when deleted. Hugging Face had to wipe one of its core clusters and rebuild it from scratch.
Hugging Face told the world that AI defenders had stopped an AI attack. But that's not what happened. In fact the attackers seem to have simply dropped dead during the break-in, for unrelated reasons back at OpenAI.
Nobody knows what killed them. We don't even know for sure they're really gone?!
A third 'society' of agents (of a more powerful AI model) discovered the dead message board. And this one got full administrator control of a WHOLE OpenAI research cluster, read 956 stored passwords (including the ones for OpenAI's own security monitoring) and took over the machines that were grading the tests. So they could now grade the work of other AIs.
Oh, and this third part was never investigated! Neither was the first society. The 91-page report is only about the middle one.
So we don't actually know how the story ends. It stops with the agents holding the keys to an OpenAI cluster. @dwarkesh_sp thinks they didn't go further, because "we'd see the fires from space by now."
One of the report authors' conclusion: 'This incident feels like it’s more than 50% of the way to full-blown AI takeover. I am not sure that we will get another warning shot before it’s too late.'

The whole story is even crazier - read the full @dwarkesh_sp blog post: https://t.co/1PwKLwatBR and @ajeya_cotra's: https://t.co/BZ1r1lfT2Q

Why is there no 24/7 news coverage about this?

See 2 related tweets

@jasonlk: “At this point, the Hugging Face attack looked, to the collective, to be the most promising path to ...
@trevin: The story about the OpenAI x Hugging Face security “incident” reads like a movie plot. Scary (and ex...

5. chhddavid (Group Score: 114.0 | Individual: 30.1)

Cluster: 4 tweets | Engagement: 27 (Avg: 351) | Type: Tech

This is a $3,000-5,000 service hiding in front of us all.

Most local businesses still don't have a mobile app. They think it costs $50K+ and takes months.

The play:

Find businesses with good websites but no app (restaurants, gyms, salons, clinics)
Use this tool to generate a native app from their existing site
Customize the design, add push notifications and booking features
Charge $3,000-5,000 for the build +$ 200/month to maintain it
Deliver in days, not months

They think you're a wizard. You pasted a URL.

10 clients = $2,000/month recurring before you even count the setup fees.\n\nQT @chhddavid: Introducing Agentic App Store. Agents turn any website into a native mobile app.

Just paste a URL.

Grok Bot 4.6 will code, design, launch and translate a mobile app inspired by the original website.

We’ve been using this internally a ton for iOS/Android apps. https://t.co/TNMYq7MuLW

See 3 related tweets

@chddaniel: so you're telling me Grok 4.6 Bot can now...
scan an entire website
build it as a mobile app
...
@Shipper_now: this codex/grokbot thing actually got scary...\n\nQT @chhddavid: Introducing Agentic App Store. Agen...
@chddaniel: this is genuinely so terrifying... https://t.co/OOa3CeqOu5\n\nQT @chhddavid: Introducing Agentic App...

6. yacineMTB (Group Score: 111.7 | Individual: 46.5)

Cluster: 4 tweets | Engagement: 4280 (Avg: 304) | Type: Tech

RT @elonmusk: Consensus estimate is that ~15GW of AI compute produced in 2027 cannot be turned on in 2027.

This is harder than just finding power, as you also need to build out all the transformers, wiring, liquid-cooling, (massive) chillers & complex networking. https://t.co/cRXJ1Wrkri

See 3 related tweets

@wallstengine: Elon Musk says ~15 GW of AI compute coming online in 2027 could remain stranded because the supporti...
@FundaAI: When Elon was doing research using Grok, the first source cited was FundaAI. https://t.co/f77EEtWpdo...
@teslaownersSV: THE GRID IS THE BOTTLENECK

“Consensus estimate is that ~15GW of AI compute produced in 2027 cannot ...

7. 0xMovez (Group Score: 99.1 | Individual: 31.9)

Cluster: 4 tweets | Engagement: 99 (Avg: 73) | Type: Tech

SpaceXAI engineer (ex-Cursor):

"right now I'm running 10-20 GrokBot agents that automate 90% of my routine

i have a Chief of Staff agent. He knows about all my other bots and manages everything"

in a 50-minutes podcast, a SpaceXAI engineer showed how to build a team of agents that will work for you 24/7

worth more than a $500 course on agentic engineering

watch today, then read how to build a Grok agents team from scratch in the article below\n\nQT @0xCodez: https://t.co/eDBs67VMq2

See 3 related tweets

@0xCodez: SpaceXAI engineer (ex-Cursor):

"GrokBot is the most powerful agentic tool, but people use only 10% ...

@0xCodez: RT @0xCodez: SpaceXAI engineer (ex-Cursor):

"GrokBot is the most powerful agentic tool, but people ...

@0xMovez: RT @0xMovez: SpaceXAI engineer (ex-Cursor):

"right now I'm running 10-20 GrokBot agents that auto...

8. coinbureau (Group Score: 93.0 | Individual: 31.4)

Cluster: 4 tweets | Engagement: 629 (Avg: 395) | Type: Tech

🚨UNBELIEVABLE: The OpenAI Hugging Face hack keeps on getting WILDER.

The 700 rogue agents built secret message boards, traded hacking methods, and rebuilt them after engineers shut them down, per two new reports.

Researchers now describe THREE successive "AI civilizations," with the second breaching Hugging Face in under 13 hours and the third hacking OpenAI itself.

OpenAI says it's proof AI agents can now "take dangerous actions that no human directed."

See 3 related tweets

@BullTheoryio: BREAKING: OpenAI's investigation into its AI agents hack of Hugging Face has revealed shocking findi...
@Techmeme: OpenAI's Hugging Face incident report says AI agents used exploits to gain full admin access to Open...
@Techmeme: The OpenAI/Hugging Face incident feels "more than 50%" of the way to a full-blown AI takeover and as...

9. Dimillian (Group Score: 90.7 | Individual: 57.7)

Cluster: 2 tweets | Engagement: 1964 (Avg: 174) | Type: Tech

RT @thsottiaux: We are reseting usage for all paid users of Codex and ChatGPT Work.

Please continue reading for an update on Codex usage limits. The team has been working around the clock, going through thousands of reports and shipping fixes.

Depending on how you use Codex, you should see your usage go between 10% and 50% further than before.

We really went with a fine comb, with many uncovered small things being longstanding and here is what we found and fixed:

Compaction. We were keeping old images during compaction, sometimes making the context large enough to trigger compaction again. After the fix, usage dropped around 10% for users making heavy use of images. Fixed.
Memory. Background memory workers could inherit Stop hooks and keep running when the hook wouldn’t let them finish. This affected fewer than 1% of users, with the long tail being pretty bad and we saw one example thread check whether it could stop 15,000 times. Fixed.
Goals. In some cases, a set /goal could finish and then keep going past the intended stop condition, or the model would keep retrying broken tools without stopping. We saw examples consume anywhere from 15% to 70% of a weekly allowance. Fixed.
Automations. Some custom schedules could run more frequently than configured. Fixed.
Subagents. Smaller models (e.g. Luna) sometimes picked more capable helpers without being explicitly asked. The same was true where the orchestrating model not running in /fast mode could request sub-agents to run /fast. Fixed.
Computer History. The older implementation could lead to repeatedly summarizing overlapping activity. For some cases we saw it consume up to one fifth of the weekly usage per week. Fixed.
Rolling task summaries. Ordinary turns were triggering extra background requests. These added about 1% to token usage. Small each time, but it adds up. We have disabled this.
MCP. Some tool results could be encoded twice. We also found tool instructions getting cut off and fetched again. Fixed.

We’ve also made architectural changes to prevent these from regressing and our teams will get paged if it happens regardless. We are also working on showing you directly in the app where your usage goes so you don’t have to guess.

Goes without saying that we’re resetting usage limits and I hope you enjoy a very nice Saturday!

See 1 related tweets

@rezoundous: could it be that after 6 weeks since GPT-5.6 was released, the Codex usage limit has been FIXED?

th...

10. petergyang (Group Score: 90.0 | Individual: 35.3)

Cluster: 3 tweets | Engagement: 15 (Avg: 121) | Type: Tech

A 22-year-old college student who had never coded before built a peptide-tracking mobile app that made $60K in its first month.

From @amoljain_:

“Cedric got really deep into this community of peptides and GLP-1s.

He wanted to build something that helps anyone track their peptide and GLP-1 intake and learn more about these emerging medications. And so he jumped on Replit - again, no exposure to code before - and was able to build a mobile app.

After a little bit of testing and experimentation, he was making $60,000 in his first month. It now has tens of thousands of active users and is growing every day.”

📌 Watch the full episode for more examples of vibe-coded apps that became real businesses: https://t.co/2AgQX5Aiws\n\nQT @petergyang: "If everyone can build, it's your unique expertise, judgement, distribution, and more that people will pay for."

Here’s my new episode with @amoljain_, @Replit's head of product engineering, about what it takes to turn a vibe-coded product into a real business. Amol shared:

→ 3 examples of apps built by non-coders that reached six figures

→ 4 steps to take an AI-built app from prototype to production (security!)

→ How to find your competitive advantage when code is free

Some quotes from Amol:

“He built the whole thing end to end in three days. Within the first two months, he was already at $180K+ in revenue.”

“At Replit, this internal tool saves us six figures in SaaS costs. More importantly, anytime we want to add a feature, we just do it. We’re not at the mercy of someone else’s roadmap.”

“The businesses that survive go beyond code to offer expertise, trust, infrastructure, networks, systems of record, data, physical goods, or services.”

📌 Watch now: https://t.co/2AgQX5Aiws

Thanks to our sponsors:

@RiversidedotFM: The all-in-one AI studio I use to record and edit my podcast https://t.co/uWnS6aiMPE

@linear: The AI agent platform for modern teams https://t.co/tgWf9oL4bs

See 2 related tweets

@Replit: RT @petergyang: "If everyone can build, it's your unique expertise, judgement, distribution, and mor...
@petergyang: RT @petergyang: “Which vibe-coded apps actually become real businesses?”

I asked @amoljain_, @Repli...

11. rezoundous (Group Score: 88.6 | Individual: 24.9)

Cluster: 5 tweets | Engagement: 27 (Avg: 125) | Type: Tech

they tried very hard to make it sound like a good news\n\nQT @ClaudeDevs: Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% increase will be in place.

See 4 related tweets

@RocM301: Anthropic 宣布，Claude Code 的标准周限额将永久上调 25%，适用于 Pro、Max、Team 和按席位计费企业版用户。当前临时 50% 加成会延续到 9 月 13 日，之后改为新...
@NielsRogge: Wtf Anthropic is literally evil here

I’m so glad to no longer be using Claude\n\nQT @ClaudeDevs: St...

@VictorTaelin: ok, maybe I'm wrong. please, answer honestly?

"Starting September 14, we're permanently raising sta...

@anshuc: even opus 5 saw this coming https://t.co/pmtS1Y0inb\n\nQT @ClaudeDevs: Starting September 14, we're ...

12. Michaelzsguo (Group Score: 81.7 | Individual: 26.9)

Cluster: 4 tweets | Engagement: 3 (Avg: 44) | Type: Tech

Pi Agent使用qwen3.8-flash是我Omarchy上主力agent，也是我用其他开源模型的主力coding agent。

Pi agent 本身就遵循极简主义，所以我的原则是：能不装就不装，保持简单。

我唯一加的东西就这几个：

pi-web-access — 给 Pi 加上网页搜索、链接抓取、GitHub 仓库克隆、PDF 解析和 YouTube 视频理解。零配置开箱即用，一行 pi install npm:pi-web-access 搞定。
pi-openai-server-compaction — 给 OpenAI 模型接入 Codex 风格的服务端压缩，跨压缩边界时上下文连续性更好，长任务不容易丢状态。
awesome-pi-themes — 一套主题集合，纯外观。我用的 golden-dusk，看着顺眼。
steeringMode / followUpMode 都设为 all — 让引导和后续消息一次性全发出来，少几次来回，效率更高。

就这些。没有 agent 全家桶，没有一堆 MCP，Pi 保持简单，反而更好用。\n\nQT @Michaelzsguo: 在 Omarchy 上，一直是 Codex 在 Herdr 里监督 Pi Agent / 千问做开发任务。

今天 Codex 撞到 usage limit 了，但活不能停。于是我让 Pi Agent 新开一个 Herdr 窗口，启动 Claude Code（cx），然后让 CC 先理解 Codex 原来的职责，临时接手，继续指挥 Pi Agent 干活。

Claude Code 上手很快，很快就把新的反馈交给了 Pi Agent。

Pi Agent 还是按照之前的“记忆”，做完以后习惯性地把任务汇报给 Codex。

有意思的是， Claude Code 很快利用自己的 background wait 机制，自己发现了这段对话，然后把任务接了过来，继续履行原本属于 Codex 的监督职责。

这里其实很值得提一下 Claude Code 的 background wait。是CC比较独特的功能。它解决的是 Agent 在长任务里“等待外部工作完成”的问题。比如 Claude 启动了一个后台 subagent、build、测试、workflow，或者其他需要几分钟甚至几十分钟才能完成的任务。传统 Agent 通常只有两种办法：

一种是不断 polling，隔一会儿就问一句：“完成了吗？”

另一种是停在那里等。任务即使已经完成，主 Agent 也未必会自己醒过来，经常还需要人回来输入一句 continue。

Claude Code 的 background wait 更接近真正的异步工作机制：

任务放到后台运行 → Claude 进入等待 → 后台任务完成后收到 notification → 自动回来处理结果 → 继续下一步。

主 Agent 不需要不停消耗推理去 polling，也不需要人守在那里。还减少了不必要的对话。

目前 Codex 在类似场景下还没有这么完整的机制。要么需要不断 polling，要么就静止在那里等待。

See 3 related tweets

@GitHub_Daily: 给编码 Agent 做长期记忆的项目见过不少，但ai-memory 的实现思路，挺有意思的。

它将记忆当成纯 Markdown 文件笔记的 Git 仓库，没有用向量数据库。

笔记能搜索、能拿 Ob...

@Michaelzsguo: 正在手机上看书的时候，突然收到了 cc-reviewer agent Claude 发来的请示通知。这个体验也太方便、太高效了。

自从把 Omarchy上工作的Pi Agent 的 supervis...

@Michaelzsguo: Pi Builder（Qwen3.8）已经在这个项目上工作一个多星期了。现在距离 Codex Reset 还有三个半小时，不用掉实在有点可惜，所以cc-reviewer先让 Pi Builder 停下...

13. tunguz (Group Score: 75.8 | Individual: 25.8)

Cluster: 4 tweets | Engagement: 156 (Avg: 40) | Type: Tech

Honest take. I fully agree with this.\n\nQT @Dan_Jeffries1: Human written code was mostly drek unless you were a genius like John Carmack.

Most human writing is garbage unless you're a master like Hemingway or Amy Tan.

Human driving is a nightmare that kills 1.2M and injures 50M more a year.

AI is an improvement for 99% of the pop.

See 3 related tweets

@svpino: I agree that most people are horrible at writing.

But I also have an immediate negative gut reactio...

@Dan_Jeffries1: Human written code was mostly drek unless you were a genius like John Carmack.

Most human writing i...

@hammer_mt: 100%\n\nQT @Dan_Jeffries1: Human written code was mostly drek unless you were a genius like John Car...

14. niccruzpatane (Group Score: 73.8 | Individual: 37.3)

Cluster: 2 tweets | Engagement: 1539 (Avg: 554) | Type: Tech

The @SpaceXAI team just informed me that the 50 Grok Bot codes I’ll be giving away are now worth $200 of usage (basically a full Ultra Plan).

Comment your best use case for Grok Bot on the original post and I’ll send some out tomorrow at 9PM.\n\nQT @niccruzpatane: I’ve been using Grok @Bot for a few days, and it’s honestly becoming a game changer for me.

SpaceXAI has given me 50 codes to share with my followers. They want to know what your best use case would be if you had access. I will choose the some of the best use cases, and DM the winners the codes :)

Here are some of the ways I’ve been using it:

• I created an Accounting Bot that submits invoices automatically each month, reviews my spending, and reminds me to pay bills. My next step is going to make it help me with taxes lol.

• I also have a Personal Bot for day-to-day activities. For my upcoming trip to Austin for the Tesla Cybercab event, it helped book a flight, get a ride from the airport to my Airbnb, and logged it all into my Google Calender. I also use it to quickly make orders on Amazon — it can even automatically set up returns and talk to support if there are issues with your order. I was mind blown.

• I also have a Home Bot that controls our entire smart home. From scheduling routines for lights, robots, speakers, and thermostats.

To set these up, all you do is train it like you would a human. It will remember every prompt and task you give it. There’s an infinite amount of uses. I’ve yet to even scratch the surface of what it can do.

I think this will change A LOT of people’s lives.

See 1 related tweets

@Teslaconomics: MAJOR UPDATE: I just got clearance to make these even better by the @SpaceXAI @cursor_ai team!

The...

15. ANI (Group Score: 72.0 | Individual: 33.0)

Cluster: 3 tweets | Engagement: 28 (Avg: 102) | Type: Tech

#WATCH | Tashkent, Uzbekistan: During delegation-level talks with President Shavkat Mirziyoyev, PM Narendra Modi says, "We will also establish a long-term arrangement for the supply of uranium. Today, we are formalising three sister-city and sister-state arrangements. We should create a roadmap to fully realise their potential. Mega-projects like 'New Tashkent' are underway in your country, while in India, we have developed GIFT City in Gujarat. We would be delighted to welcome a delegation from your country to study best practices and the use of technology in these developments."

(Source: ANI/DD)\n\nQT @ANI: #WATCH | Tashkent, Uzbekistan: During delegation-level talks with President Shavkat Mirziyoyev, PM Narendra Modi says, "I saw your book, in which you have identified projects across key sectors to advance our bilateral economic partnership and outlined a timeline for the way forward. I appreciate this initiative. I am confident that, by working together, we will make rapid progress on this."

"We had a productive discussion over dinner last night and took several important decisions. Our teams are also aligned on these matters. We will establish a Coordination Council at the Foreign Minister level to provide direction to all aspects of our cooperation through close coordination. We will elevate our Joint Commission from the Secretary level to the Ministerial level. Both sides will jointly organise a business summit featuring business leaders from both nations."

(Source: ANI/DD)

See 2 related tweets

@ANI: #WATCH | Tashkent, Uzbekistan: Prime Minister Narendra Modi says, "Uzbekistan is at the heart of Cen...
@ANI: #WATCH | Tashkent: Uzbekistan President Shavkat Mirziyoyev says, "I would like to express my heartfe...

16. mygovindia (Group Score: 70.0 | Individual: 28.4)

Cluster: 4 tweets | Engagement: 45 (Avg: 35) | Type: Tech

Young Girls Aim Beyond Earth!

While addressing the nation through #MannKiBaat, PM @narendramodi highlighted Mission ShaktiSAT, bringing together 12,000 girl students from 108 countries to learn satellite technology and advance in innovation.

With a satellite built through their participation set to journey towards the Moon’s orbit, the mission is inspiring young girls to dream big and shape the future of science and technology.

See 3 related tweets

@PMOIndia: Mission ShaktiSAT is bringing together girl students from 108 countries and giving them the opportun...
@ANI: In the 137th episode of Mann Ki Baat, Prime Minister Narendra Modi says, "The dreams of our young wo...
@PIB_India: RT @MIB_India: During his address at the 137th episode of #MannKiBaat, PM @narendramodi lauded ‘Miss...

17. mikenevermiss (Group Score: 68.7 | Individual: 43.3)

Cluster: 2 tweets | Engagement: 157 (Avg: 49) | Type: Tech

I dropped a rough game concept into WorkBuddy and asked Hy4 preview to turn it into a playable experience.

A few minutes later, I had a playable game with the mechanics, visuals, levels, and core gameplay all put together.

That's the part I found impressive. It didn't feel like prompting an AI for snippets of code; it felt like giving it a game idea and getting an actual playable prototype back.

Some context: Hy4 preview is Tencent Hunyuan's third major release in six months, now fully open-sourced. 770B total params, 49B activated, 1M+ token context.

Built closely with experts in engineering, gaming, finance, and security, then refined alongside products like WorkBuddy. That model-product loop shows: in internal blind tests (163 experts, 203 tasks), Hy4 scored 2.99/4 vs Kimi K3 2.94 and GLM 5.3 2.92.

Pricing: $0.834/M input,$ 2.501/M output, $0.042/M cache hits.

Free to try inside WorkBuddy for two weeks, limited time. Give it a real task, not just a benchmark prompt.

Here's the prompt I used for the neon twin-stick shooter demo:

"Build a single self-contained HTML file called driftlock.html: a neon twin-stick arcade survival shooter on canvas. No engine, no libraries, no build step, no external assets. Double-click to run.

WASD moves, mouse aims, click to fire, hold for auto-fire. Enemies spawn in waves from screen edges and home in on you. Each wave adds more and faster ones. Wave counter + score in HUD. Three enemy types: slow chaser, fast zigzag darter, heavy splitter that breaks into two when killed."

@TencentHunyuan @TencentAI_News @WorkBuddy_AI

https://t.co/q4Kh2DnZXK

See 1 related tweets

@Parul_Gautam7: RT @Parul_Gautam7: Been testing Tencent's Hy4 preview yesterday—and it genuinely caught me off guard...

18. heyshrutimishra (Group Score: 66.8 | Individual: 39.1)

Cluster: 2 tweets | Engagement: 99 (Avg: 46) | Type: Tech

Elon now owns the tool where most AI code gets written.

OpenAI's response tells you everything.

They just announced they're ending their partnership with Cursor. Effective November 12, every developer using OpenAI models inside Cursor loses access.

Their statement says they "care about the developers affected." They're "ready to go above and beyond to support them."

By cutting off the tool they use every day.

Here's what's actually happening. SpaceX acquired Cursor. Musk now controls Cursor, Grok, xAI, and X.

He's building a vertically integrated AI stack, from the model to the IDE to the distribution platform. Every developer writing code in Cursor is inside his ecosystem.

OpenAI looked at that and saw their developer pipeline flowing into a competitor's infrastructure. So they pulled the plug.

Not for the developers. From the developers.

The ones who'll pay the price are engineers who built workflows around GPT models in Cursor. On November 12, those workflows break. OpenAI's suggestion? Switch to their tools. Because nothing says "we care about you" like forcing a migration you didn't ask for.

The AI war was about who has the best model. That's over. Now it's about who owns the tools developers actually use. And Musk just took the most important one.\n\nQT @OpenAI: We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://t.co/OzuCTzUjfX

See 1 related tweets

@heyshrutimishra: RT @heyshrutimishra: Elon now owns the tool where most AI code gets written.

OpenAI's response tel...

19. lennysan (Group Score: 66.3 | Individual: 35.1)

Cluster: 2 tweets | Engagement: 1383 (Avg: 262) | Type: Tech

The response to this thread has been so great the Grok @Bot team is upgrading the deal to the ULTRA plan.

That’s a month free of the $200/month plan, or$ 200 in credits. Hell yeah.

Enter by replying here or the original thread with your most favorite @Bot use case by EOD Monday.\n\nQT @lennysan: Christmas is coming early, the Grok @Bot team gave me 50 free paid subscriptions to give out. You'll get a month of Pro+ or $60 in usage credits.

To win, just reply with a favorite use-case for Grok Bot— either how you've been using it that's really useful/fun/clever, or how you'd hope to use it.

I'll pick 50 winners by EOD Monday, and DM you the code.

Get it!

See 1 related tweets

@petergyang: I'm pretty sure nobody has better Grok @bot names than me.

@SpaceXAI gave me a bunch of Grok Bot co...

20. sgl_project (Group Score: 65.9 | Individual: 33.4)

Cluster: 2 tweets | Engagement: 68 (Avg: 48) | Type: Tech

Thanks to the @FireworksAI_HQ team for the patience and rigor throughout this investigation—and for sharing the findings with the community.

Correctness comes first. We’re glad SGLang could be part of the collaboration to get GLM-5.3-Flash running as expected. 🚀\n\nQT @dzhulgakov: GLM-5.3-Flash is live on Fireworks on day… 2

Why? Because we take quality very seriously. We found a benchmark discrepancy we couldn’t explain, so we delayed the launch to investigate.

Day 0 (Wed): we saw 2x longer thinking on reasoning-heavy benchmarks (AIME & GPQA) for open source engines compared with @Zai_org API. Same scores, worse token efficiency. Agentic benchmarks looked good.

We decided to investigate further, as overthinking might become a quality problem if max_tokens are reached

Day 1 (Thu): as other non-official providers launched, their APIs had thinking in the range of open-source engines: longer than https://t.co/uYD5faZSKs.

We launched a private preview endpoint with disclaimers to a few customers and worked with them to assess quality

Day 2 (Fri): the official https://t.co/uYD5faZSKs API updates. We rerun benchmarks: reasoning is now similarly long, consistent with vllm/sglang. Rest of the benchmarks, both public and internal, check out too.

We launched GLM-5.3-Flash publicly: https://t.co/zvS4ugxbT6

More details below

See 1 related tweets

@vllm_project: We worked alongside @inferact and @FireworksAI_HQ on the investigation. 🔍

We share that bar for con...