热门科技推文 — 2026年9月23日

今日科技领域的焦点是，人工智能在模型、基础设施和创意工具等方面加速发展。OpenAI、Anthropic、小米和阿里巴巴纷纷展示速度更快、成本更低、具备多模态能力且日益雄心勃勃的系统；DigitalOcean则预告了面向编程智能体的托管运行环境。Pexo和AITOPIA等初创企业正通过对话式无代码工作流，让视频制作和智能体构建变得更加便捷。与此同时，新平台不断降低实验成本，而旧金山一所学院的成立，也反映出业界正加大投入，致力于培养新一代科技创业者。

1. Lupin_Ai_Coder (Group Score: 1622.0 | Individual: 61.7)

Cluster: 68 tweets | Engagement: 1498 (Avg: 148) | Type: Tech

RT @Pexoai_offical: Now you can create videos the way you vibe-code.

Meet Pexo, your AI video agent.

Share your vision and references. Work through the details with Pexo like you’re messaging a friend.

No complex prompts. No new tools to learn.

Want to change something? Mark it on the video and tell Pexo what to change, just like leaving a comment in a doc.

Everything in this video was made with Pexo: the visuals, the motion graphics, the music, the captions, and the voiceover.

Even our founder @evanLiaoQ appears on screen.

You vibe-coded what you built. Now vibe-create how the world sees it.

#Pexo #AIVideoAgent #VibeCreate #MadewithPexo

See 67 related tweets

@rakib_hossen_ai: RT @Naila_Sync: This wasn’t a one-prompt video demo.

I gave @Pexoai_offical a cinematic commercial ...

@maxpostingx: RT @s_mohinii: I gave an AI one simple challenge:

“Take my product and turn it into a launch video....

@eva_morgan_ai: Create. Review. Refine. That’s where AI video workflows are heading.\n\nQT @Naila_Sync: This wasn’t ...
@Amelia_With_Ai: You’ve actually highlighted one of the biggest problems @Pexoai_offical solves: editing.

With Pexo ...

@DilshadAI1: The best AI video tools don’t just create — they listen.

Give feedback. Make changes. Refine the re...

2. vinhnx (Group Score: 1257.3 | Individual: 57.4)

Cluster: 48 tweets | Engagement: 12652 (Avg: 288) | Type: Tech

RT @claudeai: Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f

See 47 related tweets

@robinebers: RT @ClaudeDevs: Opus 5.5 performs at the level of Fable 5.1. It's ~30% faster and ~40% cheaper than ...
@AnthropicAI: Claude Opus 5.5 is available today.\n\nQT @claudeai: Introducing Claude Opus 5.5, the first model in...
@kieranklaassen: Opus 5.5 is the model for people who do actual work: build products, write code.
It's good from l...
@cgtwts: Dario: we need to slow down AI development

Anthropic a week later: introducing Claude Opus 5.5 http...

@nateliason: Pacing the frontier meant setting a faster pace btw\n\nQT @claudeai: Introducing Claude Opus 5.5, th...

3. mostofakamal00 (Group Score: 808.0 | Individual: 41.8)

Cluster: 37 tweets | Engagement: 426 (Avg: 119) | Type: Tech

RT @ElaraAiLab: AI video creation shouldn’t require learning complicated workflows.

I tested @Pexoai_offical by creating a SaaS launch video for Linear.

I shared the product URL, explained the direction in natural language, reviewed the first scene, gave feedback, and kept refining it through conversation.

Pexo handled the UX motion, UI animation, typography, music, and final assembly.

The interesting part isn’t just the final video.

It’s being able to create and revise the whole thing by simply talking to an AI video agent.

Try Pexo: https://t.co/aguYsinBxV

Here’s the process → final result

#MadeWithPexo

See 36 related tweets

@stevencoder9: This post shows how video creation is shifting. You start with a rough reference clip, then just tal...
@Rubab59f: The shift from "one-shot prompt and hope for the best" to multi-turn conversational editing is where...
@salmakhatunn: Video creation is shifting from writing long prompts to simply having a conversation with AI.

Start...

@codeMdSanto: RT @Hasan_4300: I wanted to see what AI video creation feels like when you don’t have to figure out ...
@salmakhatunn: Creating a product launch video used to mean hours of editing and learning complicated tools.

This ...

4. StockMKTNewz (Group Score: 587.9 | Individual: 46.7)

Cluster: 33 tweets | Engagement: 2419 (Avg: 281) | Type: Tech

🇺🇸 PRESIDENT TRUMP SAYS HE IS CHANGING THE NAME OF ARTIFICIAL INTELLIGENCE TO SUPER INTELLIGENCE

“From this point forward all of United States documents and hopefully the world will be changed to use the much more accurate term super as opposed to artificial, so it’s super intelligence

In other words welcome to the new world of super intelligence - SI, SI, let’s see if that goes, it sounds much better, it is much better, and it’s much more accurate

Let’s see if I have any power, maybe I do and maybe I don’t, we’re going to find out pretty soon

SUPER INTELLIGENCE” https://t.co/hA0ph0qm9X

See 32 related tweets

@jachiam0: I suspect the real reasoning is something like: "If Bernie and the others are calling for a ban on s...
@dzhulgakov: hear me out: AI -> SI, but it stands for Specialized Intelligence

also seems we picked a good da...

@coinbureau: 🔥TRUMP: The U.S. “totally rejects” attempts to create a global system governing AI.

Trump also call...

@_NathanCalvin: I had already kind of given up on "superintelligence" as being a description of AI that is vastly mo...
@MTSlive: SITUATION DETECTED: During his address to the UN, President Trump announced that the United States w...

5. Parul_Gautam7 (Group Score: 514.8 | Individual: 38.7)

Cluster: 21 tweets | Engagement: 353 (Avg: 197) | Type: Tech

Alibaba is setting its sights far beyond today’s AI capabilities.

Qwen is exploring 5–10T parameter models with recursive self-improvement, while the new Zhenwu V900 is built for large-scale AI workloads with clusters supporting up to 500K cards.

On the infrastructure side, Alibaba Cloud plans to surpass 20GW of global data center capacity by 2032.

And with machine thinking still below 3% of human capacity, the headroom for scaling intelligence is enormous.\n\nQT @AlibabaGroup: Speaking at Apsara Conference 2026, Alibaba Group CEO Eddie Wu highlighted the massive frontier ahead: with Machine Thinking currently representing under 3% of human capacity, scaling toward 1,000x opens an unprecedented runway for exponential growth.

To power this future, Alibaba is building across all three pillars: • AI Models: Advancing Recursive Self-Improvement (RSI) with plans for 5–10T parameter Qwen models driving toward ASI. • AI Chips: Introducing the Zhenwu V900—China's most powerful AI chip, scaling up to 500,000 cards per cluster. • AI Cloud: Expanding global data center capacity to exceed 20GW by 2032 to meet boundless demand.

See 20 related tweets

@wallstengine: $BABA UNVEILS NEW AI CHIP, PLANS 5T-10T PARAMETER MODELS AND 20GW+ OF DATA CENTER CAPACITY

Alibaba ...

@alex_prompter: Alibaba’s ASI ambitions come with one hell of a power bill.

20 gigawatts of global data center capa...

@TechByMarkandey: Alibaba is building for the next big phase of AI.

At Apsara Conference 2026, CEO Eddie Wu said Mach...

@manishkumar_dev: Alibaba’s vision for the future of AI is seriously ambitious.

5–10T parameter models, 500K AI cards...

@VaibhavSisinty: Alibaba laid out one of the most aggressive AI roadmaps anyone has announced this year.

Here's what...

6. eva_morgan_ai (Group Score: 386.8 | Individual: 43.7)

Cluster: 16 tweets | Engagement: 496 (Avg: 101) | Type: Tech

RT @Pexoai_offical: AI Guide Makes Product Launch Videos Easy for Beginners!

Need your founder on camera for a launch video, but no time to shoot?

Here's how this Pexo launch video was made.

We gave Pexo one photo of @evanLiaoQ and a short voice recording.

Pexo learned his face, cloned his voice, and read the whole script with it. Every line came back with timestamps, so the captions and graphics lined up on their own.

Since everything is generated from that one photo, he looks the same in every scene. We never booked a studio or did a second take.

Whenever something needs fixing — a shot, an audio line — we just tell Pexo directly. Even adding captions is this simple.

See 15 related tweets

@eva_morgan_ai: From a single photo and voice sample to a complete launch video — this is a glimpse of how AI is res...
@ai_explorer25: Pexo turned a single photo and voice note into a complete founder launch video. Consistent face, clo...
@ArafatMd93059: RT @GeekyDamian: 𝗖𝗮𝗻 𝗔𝗜 𝗥𝗲𝗰𝗿𝗲𝗮𝘁𝗲 𝗧𝗵𝗲 𝗤𝘂𝗮𝗹𝗶𝘁𝘆 𝗢𝗳 𝗮 𝗣𝗿𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗩𝗶𝗱𝗲𝗼?

What if you could tu...

@CodebyNihan: The future of video creation isn’t learning complicated editing software.

It’s simply telling AI wh...

@ElizabethA77617: This feels like one of the more practical ways to use AI for video.

Being able to mark a specific f...

7. maxpostingx (Group Score: 343.4 | Individual: 43.1)

Cluster: 11 tweets | Engagement: 308 (Avg: 68) | Type: Tech

RT @Safwon_CoderAI: Hospitality venues wanting to book live music or manage entertainment usually get stuck dealing with an endless headache of spreadsheets and late invoices.
As someone who can't edit videos to save my life, I threw my SaaS concept into Pexo and chatted my way into the product video below—complete with warm blush aesthetics, clean geometric typography, and smooth scene transitions. This time, I used @Pexoai_offical.
Pexo is an AI video creation tool where you can just chat with it to make videos. Tell it what you want, how to tweak the visuals—plain English works fine. Even someone like me with zero editing experience can jump right in. First, a bit of background. I usually build products, not videos. This time, I wanted to properly introduce JAM JUICE—the world's first entertainment operating system for hospitality pros—only to realize the real hassle isn't writing the copy, it's the video. I don't know AE. And I sure don't want to burn my evenings learning editing just to drop one clean promo.
I tossed the concept into Pexo and told it: Make me a warm, clean SaaS promo for hospitality professionals—keep scene changes under 3 seconds and make it look professional. It didn't spit out a video right away.
First, it aligned with me. Soft blush background, charcoal typography, coral-pink keywords, and precise pacing. Feels like hashing out requirements with someone who actually knows motion design. I fired back a few sentences in plain English. Once confirmed, it went off and ran with it. What came back was the finished 19-second editorial piece. Text sync, smooth transitions, and a gorgeous color palette—all there. No dragging footage into a timeline. For someone who just wants to explain their product clearly, that's a game-changer.
You can keep chatting after that. If an animation drags, tell it to tighten up. Never touched the timeline the whole way. The one below is what came out of chatting with the platform concept. If you've got a fresh SaaS product, want an intro video, but don't feel like learning editing first—give Pexo a shot.

See 10 related tweets

@anjum_ai: All the product details are already written out, but when it comes to promotion, are we still just d...
@jihad_sameul: RT @ExpertWithElora: Still dropping plain-text checklists or messy screenshots when showing off app ...
@ethan_walker_ai: RT @hey_simanto: Still struggling to explain complex marketing services with just text links and sta...
@LunaTechAI: I used to avoid making promo videos because editing is a nightmare.

Pexo just removed every excuse....

@Mason148749: Your product is ready to launch—your video shouldn’t take a week. ⚡\n\nQT @karlarboledas: You built ...

8. LearnWithSubhan (Group Score: 339.4 | Individual: 63.5)

Cluster: 9 tweets | Engagement: 748 (Avg: 79) | Type: Tech

RT @lihazadn_Ai: 50 legendary websites that feel like the internet’s hidden toolbox 🧰

https://t.co/g5ZVDyU9UZ — Free research papers
https://t.co/oTVYPGbnmI — Borrow books online
https://t.co/r4ueOvzTDr — Free academic journals
https://t.co/PY1Chsylth — 70K+ free books
https://t.co/bYjlSmE92j — Free textbooks
https://t.co/QuosEMdqq9 — Free online courses
https://t.co/i3uJiXBdIf — Solve complex problems
https://t.co/MXwwALj8YE — AI research assistant
https://t.co/qv7FmbY8rH — Research-backed AI answers
https://t.co/5bMqdmZrhH — Find research connections
https://t.co/K12mBXpzQv — Academic search engine
https://t.co/RPJKCf3k9P — Understand research papers easily
https://t.co/B7NrPfm3zY — Free Photoshop alternative
https://t.co/NqtB5cC3r7 — Compress & optimize images
https://t.co/9MRb7jl7xf — Remove image backgrounds
https://t.co/shpJF4dwNS — Remove objects from photos
https://t.co/eSYBusQZsf — Remove video backgrounds
https://t.co/y3K9BlNnao — Beautiful code images
https://t.co/ED4x9UwHcK — Share elegant code screenshots
https://t.co/Edjh7lXgWy — AI search engine for developers
https://t.co/8L6HgCitLE — Test and debug regex
https://t.co/3ELARtKTZ2 — Format & clean code
https://t.co/5RJMCetauI — Format & validate JSON
https://t.co/qV98RNqN49 — Understand terminal commands
https://t.co/EneDhPVrQC — Stunning product mockups
https://t.co/DDYsVyszwp — Powerful online mockup generator
https://t.co/V9bhHR9thp — Find alternatives to any app
https://t.co/mmkEE6wmNn — Check data breaches
https://t.co/AmpRu4WaKG — Scan files & URLs for malware
https://t.co/9OYf7zW63H — Self-destructing text notes
https://t.co/fFLqLK87ph — Disposable temporary email
https://t.co/MYRuX4Ut9N — 10-minute temporary email
https://t.co/kb9xO7t2WW — Temporary & secure file sharing
https://t.co/KHHqrmG5qK — Explore global live radio
https://t.co/T1Kqt2ggYW — Discover similar music & artists
https://t.co/rYWLclexpD — Find songs from movies & shows
https://t.co/EKzivBR6XU — Focus music for coding
https://t.co/mbgJhlW8xJ — Custom background sounds
https://t.co/0xBXLxATxk — Ambient café sounds for productivity
https://t.co/efWBUrONQ5 — Find where to stream movies
https://t.co/uDDEFt2yQy — The internet’s digital library
https://t.co/selSDMXoAP — Create a personal copy of any webpage
https://t.co/MlI8naIim4 — Find related websites
https://t.co/LGDKuj2HUN — AI YouTube video summaries
https://t.co/GwH2awNzHt — Ultimate smart bookmark manager
https://t.co/JLVAeEJM15 — Check real-time service outages
https://t.co/Nsbm3n4Jgv — Smart reverse image search
https://t.co/2Q8nZ4WmLR — Quick internet speed test
https://t.co/q77reMw4tf — All-in-one PDF tools
https://t.co/y6LgAXq6O4 — Merge, split & convert PDFs

Save this. You’ll definitely need some of these later! 🔖

Follow @lihazadn_Ai for more useful websites, AI tools & tech resources.

See 8 related tweets

@Grow_withAI: 🚀 50 Useful Websites You Should Know

Unpaywall — Free research papers
Open Library — Borrow ...

@Ziven_Coder: 🚀 50 Useful Websites You Should Know

Unpaywall — Free research papers
Open Library — Borrow ...

@AnnuKumari35786: 50 websites that feel like the internet’s hidden toolbox

https://t.co/BaWncAnwLD — Free researc...

@eva_morgan_ai: RT @FutureGirlAi: 🚨 50 websites that are WAY too useful to be this easy to access:

Save this thread...

@SuccessSparksX: RT @SuccessSparksX: 50 websites that feel like the internet’s hidden toolbox 🧰

https://t.co/ujVE...

9. Alokkumarzz (Group Score: 308.2 | Individual: 48.4)

Cluster: 12 tweets | Engagement: 4124 (Avg: 161) | Type: Tech

RT @digitalocean: DigitalOcean Managed Agents is now in public preview.

Run Claude Code, Codex, or your own LangGraph agent in a runtime environment that pauses when idle. Put its tools behind one governed endpoint, and pick from 75+ open and proprietary models. One cloud, one bill.

Prompts to get started available in the blog: https://t.co/QQqQ9w28lG

See 11 related tweets

@wallstengine: DIGITALOCEAN LAUNCHES MANAGED AGENTS IN PUBLIC PREVIEW

DigitalOcean is launching a managed runtime ...

@heyrobinai: 🚨 you don't need a DevOps engineer to run AI agents anymore

today, you connect your tools, pick you...

@exec_sum: JUST IN: DigitalOcean has launched managed agents in public preview.

This gives developers one plac...

@AnatoliKopadze: someone finally put the whole agent stack in one place...

agent runtime, tool access, and inference...

@Alokkumarzz: AI agents are changing too fast to bet everything on one model.

that's why the 75+ model support in...

10. MollySOShea (Group Score: 296.9 | Individual: 52.8)

Cluster: 8 tweets | Engagement: 586 (Avg: 92) | Type: Tech

RT @gaganbiyani: Today, we’re launching an ambitious new school called The Horowitz Andreessen Academy.

Based in San Francisco, The Academy serves the most promising young high school graduates.

We think this can be an elite institution that attracts top tier talent. One that prepares students for the future rather than remaining stuck in the past.

The #1 goal is to help students learn to build, which is the most important skill in the AI era. They'll learn primarily by pursuing their own projects, either individually or in groups. There are classes and guest lectures, too, from some truly amazing people who have built modern-day Silicon Valley.

The Academy is designed as a network, since that’s the reason students go to school in the first place. Core to that network are our 10 Founding Partners: Anduril, Anthropic, Coinbase, Google, Meta, NVIDIA, OpenAI, Palantir, Replit, and Stripe. The network includes over 50 hiring partners and over 200 speakers and mentors. To join as a hiring partner or faculty member, you can apply on our website.

We raised $42M in funding led by @a16z. I'll be CEO and @pmarca and @eriktorenberg will join me on the board.

Applications are open for our Founding Class Fellowship, which will be one year and tuition-free. Eventually, pending regulatory approval, we plan to offer a two-year program that charges tuition, similar in cost to an elite private university.

We're looking for the most unusually ambitious young builders on the planet. Come join us in San Francisco: https://t.co/NfavmBy2ex

See 7 related tweets

@ADoricko: Proud to be a mentor for The Horowitz Andreessen Academy

Young people are capable of incredible thi...

@eriktorenberg: I’ve been wanting to build this for a decade, and Gagan and I almost teamed up to do it 6 years ago....
@Tocelot: The next generation of AI-native leaders will be taught differently

AI makes it possible to build a...

@MollySOShea: “The goal of the academy very much is for moonshots.”

Gagan Biyani @gaganbiyani says Horowitz Andre...

@MollySOShea: “One of the greatest lessons of the Valley is: it's not about your failure. It's about your success....

11. RameshBais12 (Group Score: 293.6 | Individual: 38.4)

Cluster: 13 tweets | Engagement: 190 (Avg: 30) | Type: Tech

RT @Aura_lixx: AI is moving from simply using tools to actually building intelligent agents.

AITOPIA Agent Builder makes that transition accessible without requiring you to write a single line of code.

Build → Publish → Earn. Your idea can become an agent, and your agent can become a real product.

Definitely worth exploring if you’re interested in building with AI. 🚀

See 12 related tweets

@mostofakamal00: RT @AITOPIAai: Build an AI agent. Publish it.Earn from it.

Introducing AITOPIA Agent Builder , crea...

@Grow_withAI: RT @David_Ai0: Building an AI agent doesn’t have to mean writing code.

With AITOPIA Agent Builder, ...

@altryne: AI Agent - An AI with a loop, tool calling and maybe a harness

AI Assistant - Has it's own compute...

@SuccessSparksX: RT @SuccessSparksX: The future of AI development is codeless. 🌐 With AITOPIA Agent Builder, you can ...
@Liyana_Coder: RT @heyrohitai: Building AI agents just got simpler.

No coding needed with @AITOPIAAI.

Build. Publ...

12. codeSTACKr (Group Score: 262.7 | Individual: 46.7)

Cluster: 13 tweets | Engagement: 4616 (Avg: 499) | Type: Tech

RT @OpenAI: Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.

GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.

We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.

See 12 related tweets

@BradGroux: RT @thsottiaux: GPT-6 Sol and Luna are out. Not only are they a very significant improvement across ...
@haider1: YEAH, GPT-6 sol is dropping today

also waiting for gpt-6 luna, which apparently showed up in the co...

@haider1: GPT-6 Sol costs only $2 for input and$ 10 for output.

WTF\n\nQT @OpenAI: Please welcome GPT-6 Sol a...

@ChrisGPT: GPT 6 Sol scores the around the same as 5.6 Sol on DeepSwe, but is more token efficient and half the...
@himanshustwts: Intelligence is becoming too cheap to meter, for real.

This is actually Deepseek V4.1 level pricing...

13. ElizabethA77617 (Group Score: 248.7 | Individual: 33.6)

Cluster: 10 tweets | Engagement: 0 (Avg: 113) | Type: Tech

This is a smart way to validate an idea before spending months building it.

Pexo makes it easier to turn a rough product concept into something people can actually see and understand.

Being able to visualize an idea early can be incredibly useful.\n\nQT @lxfater: 分享一个低成本验证创业想法的办法：产品还没做，先用 AI 把宣传片拍出来！！

我一直想要一副能用来 Vibe Coding 的眼镜，但市面上没找到太合适的产品。

这次试了下 Pexo，没想到还真把我脑子里的画面做出来了！！

Pexo 是一款 AI 视频创作工具，可以直接跟它聊天来做视频。我把想法和想要的画面告诉它，看完哪里不满意，就在画面上圈出来，在批注里说清楚怎么改，再继续调整。

不用会 AE，也不用打开剪映。原本只能靠嘴AI视频制作_压缩讲的创业点子，现在有了一条宣传片，可以直接拿给别人看、听听反馈了。

下面的视频里，我不仅会展示这条虚拟产品宣传片，还会手把手带你看，怎么用 Pexo 从一个想法做到成片👇

如果你也想要这样的眼镜，评论告诉我！！

这次用的工具：@Pexoai_offical

https://t.co/x0Cb3xEBHb

See 9 related tweets

@Arzu_with_Ai: RT @lxfater: 分享一个低成本验证创业想法的办法：产品还没做，先用 AI 把宣传片拍出来！！

我一直想要一副能用来 Vibe Coding 的眼镜，但市面上没找到太合适的产品。

这次试了...

@hey_Jessicaai: A smart way to validate an idea before spending months building it.

Pexo helps turn rough concepts ...

@salmakhatunn: J bro nailed the difference between tools and agents.

The interesting part is the workflow: decide ...

@JayBisen473370: RT @felix_muller_ai: 产品 Launch 视频一直是我很感兴趣的方向，平时也会自己不断摸索和尝试。

最近看到 Pexo 这个新产品后，我第一时间注册账号亲自试了试。刚好 Prod...

@novaXsignals: Why build the product first when you can validate the idea visually?

Turn the concept in your head ...

14. Xiaomi (Group Score: 221.7 | Individual: 38.1)

Cluster: 7 tweets | Engagement: 1546 (Avg: 391) | Type: Tech

RT @XiaomiMiMo: Introducing Xiaomi MiMo-V2.6 — Pro & Flash.

Frontier intelligence, all the modalities, built in public.

🔹 Two omnimodal models, advancing through scaled reinforcement learning 🔹 Pro performs on par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks 🔹 Pro scores 46 on the Artificial Analysis Intelligence Index — the highest among open-source models 🔹 Stronger coding, computer use, 3D reasoning and creative capabilities 🔹 Open model weights, technical report, RL environments and training code

Blog：https://t.co/JGdwC9Ymvj

See 6 related tweets

@theo: Did a few quick tests and these seem very very legit. 2.6 Pro in particular is doing well in some pr...
@tonbistudio: New Mimo model released!

Excited to try this, Mimo V2.5 was always my go to for cheap, fast, high ...

@gmi_cloud: soon\n\nQT @XiaomiMiMo: Introducing Xiaomi MiMo-V2.6 — Pro & Flash.

Frontier intelligence, all the ...

@ivanfioravanti: Another model I was waiting for the release! MiMo 2.6! On benchmarks it seems great! Need to test it...
@OfirPress: another win for mini-swe-agent vs codex & claude code. mini-swe-agent is consistently the best h...

15. codeMdSanto (Group Score: 214.9 | Individual: 31.6)

Cluster: 11 tweets | Engagement: 103 (Avg: 88) | Type: Tech

Most people are paying for AI credits just to experiment.

Meanwhile, Lovie is giving away ~160M tokens/month across 20 providers and ~400 models — for free.

No credit card. No monthly bill.

If you’re building with AI, this is definitely something to put on your radar. 🚀\n\nQT @lovieco: Today, Lovie is giving away ~160m tokens/month from top ai labs for 0. Renewed every month.

20 providers. ~400 models. 1 key. These tokens are enough to ship your v1 without ever touching a credit card.

Head to https://t.co/ZMMtwciR8B to claim yours. https://t.co/rCT9e1t3jz

See 10 related tweets

@supressedvoic: The best part of building with AI is being able to experiment without constantly watching your credi...
@TechByMarkandey: Building with AI can get expensive really fast.

You might spend a lot on API calls just to test an ...

@Grow_withAI: RT @codeMdSanto: Most people are paying for AI credits just to experiment.

Meanwhile, Lovie is givi...

@hey_abusiddik: RT @Hey_Aivetra: Building with AI just got easier with Lovie.

Lovie offers ~160M tokens/month acros...

@maxpostingx: RT @supressedvoic: The best part of building with AI is being able to experiment without constantly ...

16. FrancoisChauba1 (Group Score: 213.5 | Individual: 60.6)

Cluster: 6 tweets | Engagement: 1002 (Avg: 97) | Type: Tech

RT @AndrewYNg: The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field.

I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so.

First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world.

The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability.

Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended.

I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent.

Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.)

Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration.

Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building.

[Original text (with links): https://t.co/jni2tWazAH ]

See 5 related tweets

@elonmusk: Interesting perspective\n\nQT @AndrewYNg: The loudest voices stoking fears about AI dangers have mad...
@NaveenGRao: This 👇\n\nQT @AndrewYNg: The loudest voices stoking fears about AI dangers have made tremendous head...
@BrianRoemmele: Thank you professor @AndrewYNg for level headed clarity around AI doom and safety.\n\nQT @AndrewYNg:...
@pstAsiatech: Ng: The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take s...
@robbensinger: RT @shorttimelines: Every person I've seen be dismissive of the Hugging Face incident has described ...

17. DeryaTR_ (Group Score: 212.1 | Individual: 38.3)

Cluster: 8 tweets | Engagement: 341 (Avg: 150) | Type: Tech

As I mentioned, last month we crossed the key milestones in the age of AI with the release of Astra, & reached AGI escape velocity. Even with proposed pacing of AI advances, the age of AI will continue shooting for the stars at full speed, transforming every field along the way🚀

“On August 28, we began training a new internal model. In addition to resolving the Navier–Stokes Millennium Prize problem⁠, this model has now resolved more than 100 long-standing open problems across most areas of mathematics. The pace of its progress in mathematics has surprised the mathematicians within OpenAI.”\n\nQT @OpenAI: We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.

The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning.

Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community.

https://t.co/QCMFLFlIv8

See 7 related tweets

@VaibhavSisinty: OpenAI trained a new internal model starting August 28.

In less than a month it solved over 100 lon...

@AISafetyMemes: 1) 24 days ago, OpenAI began training a new model...

That model has already solved over 100 l...

@rynorhn: holy fucking shit.

openai says its new internal model has already resolved more than 100 long-stand...

@Kyrannio: RT @choblin29: 🚨wtf. OpenAI says its new internal model it began training just 24 days ago has alrea...
@lukOlejnik: OpenAI says its model made more than 100 breakthroughs in mathematics. The company is now consulting...

18. higgsfield_ai (Group Score: 211.9 | Individual: 50.1)

Cluster: 7 tweets | Engagement: 2969 (Avg: 430) | Type: Tech

Claude Opus 5.5 vs GPT-6 Astra in 3D gaming.

Samurai game demos made with Higgsfield. https://t.co/e2TNM61ra0\n\nQT @higgsfield: Introducing Higgsfield x Claude Opus 5.5.

Anthropic’s newest model, which is 40% cheaper and over 30% faster than Opus 5, is live on Higgsfield.

Pair Opus 5.5 with the Higgsfield MCP and turn your most ambitious ideas into complete products.

Available on Claude via Higgsfield MCP and in Higgsfield Supercomputer.

See 6 related tweets

@higgsfield_ai: Claude Opus 5.5 + Higgsfiled built this animated windmill in Blender in 15 minutes.

The modeling, r...

@higgsfield_ai: Claude Opus 5.5 vs GPT-6 Astra for commercial video production with Higgsfield.

Opus: 60 min, $50. ...

@higgsfield_ai: Claude Opus 5.5 recreated these Higgsfield effects in TouchDesigner from references in 15 minutes. h...
@higgsfield_ai: Claude Opus 5.5 vs GPT-6 Sol in 3D game development.

What a time to be building with Higgsfield. h...

@higgsfield_ai: Claude Opus 5.5 on Higgsfield turned one house photo and its floor plans into a full 3D model in Ble...

19. CodebyNihan (Group Score: 197.1 | Individual: 36.0)

Cluster: 6 tweets | Engagement: 36 (Avg: 129) | Type: Tech

For a decade, the AI industry followed one formula: More data. More parameters. More compute. Boltzbit is asking a different question:

What if the model could adapt its own weights while it works? Its Bayesian Self-learning Transformers use live data to generate weight updates, allowing new feedback to change the model instead of remaining trapped in a prompt or memory layer.

The paper’s theoretical analysis estimates that BAST could reduce compute requirements by [300, 3,000]× compared with conventional training. Dive into the method and calculations here 👇 https://t.co/TTeE69NmHC\n\nQT @boltzbit: We’ve just released the preview version of our latest paper: Infinite-Parameter LLMs — Generating and Adapting Weights from Live Data. In it, we demonstrate that large language models can learn up to 1,000x faster through Bayesian Self-learning transformers (BAST) than SOTA training algorithms engineered by human researchers. This breakthrough marks a critical shift in AI research, from cost-intensive, static-weight AI to energy-efficient, dynamic-weight AI.

For a decade, building more advanced models has been achieved by training bigger ones on more data. This approach is now hitting a ceiling with the supply of pretraining text projected to run out in the next two to five years. Meanwhile, the fast adoption of AI agents is producing an unprecedented amount of continuously-growing data that models can learn from. None of it is captured, locked in individual sessions and lost as soon as the agent completes the task.

Bringing self-learning AI agents to users captures the value of that data. We validated that BAST LLMs overcome the fundamental limitations of memory-based learning, such as Retrieval-Augmented Generation (RAG), in both cost and performance, across long-context conversations. Without underlying architectural innovation like BAST, RAG, larger context windows, and agent scaffolding suffer a drop in performance and escalating input token cost due to the static weight of LLMs.

This research underscores a fundamental premise of our work at Boltzbit to achieve General Learning Intelligence and democratise model ownership. Scaling up static-weight model size or layering on more workarounds cannot address the spiraling cost of training AI systems. The viable path is a new AI architecture that adapts its own weights.

Preview version of the paper: https://t.co/OPHELtkvNz

See 5 related tweets

@Tech_Arish: AI agents have a major problem most people overlook. They generate valuable data during every task, ...
@choyamymuna: AI models usually stop learning the moment training ends. Boltzbit’s Bayesian Self-learning Transfor...
@Ayzacoder: "Every AI agent interaction produces new data. Almost none of it becomes part of the model. Boltzbit...
@SharminNahar401: Live data + adaptive weights + continuous learning = a new direction for AI.

Boltzbit’s Bayesian Se...

@choyamymuna: RT @choyamymuna: AI models usually stop learning the moment training ends. Boltzbit’s Bayesian Self-...

20. TipRanks (Group Score: 190.2 | Individual: 38.1)

Cluster: 7 tweets | Engagement: 94 (Avg: 55) | Type: Tech

Jefferies raised the firm's price target on Meta Platforms $META to$ 875 from $710 and keeps a Buy rating on the shares.

Prior to Muse, investors had completely dismissed the idea that Meta would be able to release a personalized consumer AI agent, favoring Google instead, but Muse is "a killer product" and "flips the narrative," the analyst tells investors.

Based on "unprecedented" consumer adoption - well ahead of Claude and Gemini after launch - and the firm's experience over the past two weeks that "supports the hype," the analyst believes Muse has "all the signs of a killer app."

JPMorgan says Meta Platforms' Muse has continued to climb in Apple's App Store to currently the number one most downloaded free app in the U.S. The "combination of virality and Meta's strong distribution platform make the buzz around Muse quite palpable," the analyst tells investors in a research note.

JPMorgan says that while still early, Muse has the potential to become the most widely used consumer AI application since ChatGPT.

The firm believes Muse adoption and engagement will continue to ramp, with Meta beginning to prove out AI returns beyond its core advertising platform. JPMorgan sees a potential total addressable market in the tens of trillions of dollars.

It reiterates an Overweight rating on Meta with an $820 price target.

BofA analyst Justin Post notes Meta's Muse app launched on September 8 and early reviews suggest Muse's functionality is resonating with users.

Third-party data suggest strong initial interest in Muse, which reached the number one position on the U.S. iOS App Store in the Productivity category shortly after launch and has maintained a 4.9-star rating across more than 25,000 reviews, added the analyst. The firm, which thinks investors were concerned about Meta's ability to build a differentiated AI product, sees Muse's early success driving more optimism on Meta's potential AI investment returns.

BofA keeps a Buy rating and $810 price target on Meta shares.

See 6 related tweets

@negligible_cap: JPM believes Muse has potential to become the most widely used consumer AI application since ChatGPT...
@exec_sum: BREAKING: JPMorgan is predicting Meta’s new Muse AI app could become "the most widely used consumer ...
@StockSavvyShay: JPMorgan says $META Muse could become the most-used consumer AI app since ChatGPT after reaching #1 ...
@SquawkCNBC: "I think what I've seen so far is Muse is the best product coming to market," says @markmahaney of M...
@PolymarketMoney: JUST IN: JPMorgan declares Meta Muse could become the “most-used consumer AI app since ChatGPT” afte...