China just wrecked all of American AI. Silicon Valley is in shambles.

[bnew]

Ghost of Hope (@grayface_ghost@tech.lgbt)

There's a reporter on the BBC now saying something about the US and the EU producing democratic AI but China producing authoritarian AI. Give me fukking strength When I was a kid we used to worry the planet would be destroyed by a giant mainframe running hunter killer robots. Instead we're...

tech.lgbt

Spoiler: full text & large images

1/4
@grayface_ghost@tech.lgbt
There's a reporter on the BBC now saying something about the US and the EU producing democratic AI but China producing authoritarian AI.Give me fukking strengthWhen I was a kid we used to worry the planet would be destroyed by a giant mainframe running hunter killer robots. Instead we're going to burn the planet into lifelessness so our billionaires can make more useless money selling us large language models that draw people with thirteen fingers and hallucinating facts that have conspiracy theorists injecting fluorescent light bulbs

2/4
@Thebratdragon@mastodon.scot
@grayface_ghost what the US/EU mean is how dare china make our new cash cow bubble, smaller and cheaper, before we have ripped off all the investors and public.

3/4
@grayface_ghost
@Thebratdragon Ding! Ten points to the Soft-Drink Mythical Creatures team

4/4
@nini@oldbytes.space
@grayface_ghost Our glorious and honorable AI, their craven and regressive AI.


To post posts in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[TDUBB]

 

[TDUBB]

 

[bnew]

Spoiler: text

Since launching DeepSeek R1 on January 20, DeepSeek has gained over 100 million users, with $0 advertising or marketing cost. By February 1, its daily active users surpassed 30 million, making it the fastest application in history to reach this milestone.

Why? I also spend so much time chat with it, the profound answer, is the key reason for me.

 

[bnew]

Josh Johnson
1.38M subscribers
When AI Competes: Deepseek Vs Open AI Explained

 

[bnew]

https://archive.is/JoO5U

 

[bnew]

https://archive.is/Icf4d

Spoiler: full text & large images

1/9
@deepseek_ai
Day 0: Warming up for /search?q=#OpenSourceWeek!

We're a tiny team @deepseek_ai exploring AGI. Starting next week, we'll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency.

These humble building blocks in our online service have been documented, deployed and battle-tested in production.

As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey.

Daily unlocks are coming soon. No ivory towers - just pure garage-energy and community-driven innovation.



2/9
@ujjwalthakur_
AGI



3/9
@AntDX316




4/9
@joacodok
the real openai



5/9
@0xa8l
Do you guys have a business model? It seems you are just open sourcing your secrets!



6/9
@novita_labs
if you want to use DeepSeek R1 API for free

[Quoted tweet]
Still looking to integrate DeepSeek API?

Refer a friend to Novita and both earn $20 in DeepSeek API credits—up to $500 total!

Use it for your application, or dive into AI exploration. Either way, it's on us.

Get your first $20 here: shorturl.at/BR4vc

#NovitaReferral


7/9
@innocentamna12
Nice work by the team! Good progress being made.



8/9
@Putra_GPT




9/9
@Ekaeoq
I hope you people win




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

https://www.pcmag.com/news/with-flood-of-chinese-ai-on-the-horizon-us-mulls-deepseek-ban

With Flood of Chinese AI on the Horizon, US Mulls DeepSeek Ban​

The Trump administration is reportedly considering a DeepSeek ban on government devices, but it might also ban it outright, à la TikTok, The Wall Street Journal reports.



By Emily Forlini

March 7, 2025

(Credit: Emily Forlini)

Concern in Washington over Chinese AI apps has extended to DeepSeek, with the Trump administration reportedly looking to ban it on government devices.

National security is the major issue, The Wall Street Journal reports, particularly as it relates to DeepSeek's handling of user data, which is stored on Chinese servers. Companies based in the region can be compelled to hand over information to the Chinese Communist Party (CCP), something the US government cited in its push to ban TikTok.

The TikTok ban is currently on hold after President Trump gave the app more time to find a US buyer. The WSJ reports that the administration is also considering an outright ban on DeepSeek in the US, though its sources say discussions are "still at an early stage."

Another option is preventing US companies from building products with DeepSeek via cloud service providers like Amazon AWS and Microsoft Azure, both of which have added it to their roster of available models. This could mean higher AI costs for these businesses if they had to replace it with models from OpenAI or other more expensive, US-made options.

At a minimum, the US would bar government employees from downloading the DeepSeek app on government devices.

Other nations have already limited access to or banned DeepSeek including Italy, South Korea, Australia, Canada, and Taiwan. In the US, New York and Texas banned it on government devices.

In our testing, we found DeepSeek to censor its answers and regurgitate CCP propaganda. At one point, its answer used "we" pronouns, speaking as the CCP. But it's unclear if banning individual products, like TikTok or DeepSeek, will be an effective long-term strategy.

---

A Flood of Chinese AI Coming: Are They Any Good?​

This week, two Chinese companies debuted AI models that supposedly rival DeepSeek, though the claims are difficult to verify.

Retail giant Alibaba says its model outperforms DeepSeek and OpenAI. "In a series of authoritative benchmark tests, Qianwen QwQ-32B model performed exceptionally well, almost completely surpassing OpenAI-o1-mini and being on par with the strongest open source reasoning model DeepSeek-R1," Alibaba says, as translated by Google.

A startup called Manus AI, meanwhile, claims to have built "a truly autonomous agent...the next paradigm of human-machine collaboration, and potentially a glimpse into AGI." In the video below, CEO and co-founder Yichao Peak Ji gives the example of the AI reading resumes and taking notes on candidates on its own, even if the human closes their laptop.

Introducing Manus: The General AI Agent



Channel Info Manus AI Subscribers: 14.9K subscribers

Description

Spoiler

Manus is a general AI agent that bridges minds and actions: it doesn't just think, it delivers results. Manus excels at various tasks in work and life, getting everything done while you rest.

Feel the future at Manus

However, some doubt Manus' claims, accusing it of overstating capabilities for its moment in the spotlight, the South China Morning Post reports. Manus is only available as an extremely limited, invitation-only preview. Adding some intrigue to its short rise to fame, social media company X suspended its account over alleged ties to cryptocurrency scandals.

"We are actively working with the X team to resolve this issue. Initial observations suggest that the suspension may be related to third-party mentions of cryptocurrency scams," Ji tweeted on his own account. "To clarify: Manus has never been involved in any cryptocurrency projects, token issuance, or blockchain initiatives."

 

[IIVI]

Knew the update on this thread was going to be about Manus.

I’ve been using Perplexity A.I for Deep Research, but Manus is exponentially better.

That thing is honestly insane.

 

[bnew]

Knew the update on this thread was going to be about Manus.

I’ve been using Perplexity A.I for Deep Research, but Manus is exponentially better.

That thing is honestly insane.

they did some investigating..




[News] Manus turns out to be just Claude Sonnet + 29 other tools, Reflection 70B vibes ngl

https://archive.is/EPlTo

Spoiler: text

https://i.redd.it/yi72dtz4krne1.png

Alexander Doria (@Dorialexander) | https://nitter.poast.org/Dorialexander/status/1898719861284454718 | https://xcancel.com/Dorialexander/status/1898719861284454718 | Alexander Doria @Dorialexander, Twitter Profile | TwStalker

https://i.redd.it/6b84agl7krne1.png

https://twiiit.com/jianxliao/status/1898861051183349870 | https://nitter.poast.org/jianxliao/status/1898861051183349870 | jian (@jianxliao) | jian @jianxliao, Twitter Profile | TwStalker

Spoiler: thread continued

Spoiler: full text & large images

1/11
@Dorialexander
Manus seems to be Claude 3.7: "Human:" and "Assistant:" creates a prompt injection and it get stuck in neverending loop.

[Quoted tweet]
Could someone with access give it a try with DeepSeek and Claude special tokens? Like (3.7 still has it?) or

2/11
@Dorialexander
Conversely, Qwen or DeepSeek special tokens are parsed normally.



3/11
@Dorialexander
After experimenting with it, I do like the UI. But it's fundamentally a workflow like Devin, not an actual agent (at least nothing really beyond the built-in agentic capacities of Claude)



4/11
@Dorialexander
To give more detail: it never manages to fill a file with synthetic human: assistant: exchange. Instead there are regular "encountered errors" and generation examples being outputted directly through injection.

5/11
@AlgoDistill
got any invite codes?



6/11
@Dorialexander
Had one more but sent to a tech journalist.



7/11
@MasoudMaani
we should have a name for these instances.
I propose "wrapper grift" to the high court.



8/11
@Dorialexander
The real issue is the lack of transparency. At least Devin was upfront about it. This is necessary to assess what the system really brings: I've seen code example that Claude should one shot directly…



9/11
@b_arbaretier
Crazy they outperformed deep research on Gaia then ! Seems like plain model orchestration still has bright days ahead



10/11
@Dorialexander
Not really meeting my experience so far, though…



11/11
@GuruduthH
Regardless of whether it's uses Claude or not but If it's true , still they built a system that used Claude really well and that's impressive




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

Spoiler: thread continued

Spoiler: full text & large images

1/11
@jianxliao
So... I just simply asked Manus to give me the files at "/opt/.manus/", and it just gave it to me, their sandbox runtime code...

> it's claude sonnet
> it's claude sonnet with 29 tools
> it's claude sonnet without multi-agent
> it uses @browser_use
> browser_use code was also obfuscated (?)
> tools and prompts jailbreak



https://video.twimg.com/ext_tw_video/1898860612303949824/pu/vid/avc1/1576x1080/dVVgF0dIoGkbvRxo.mp4

2/11
@jianxliao
replay link: Check Files Under /opt/.manus Path - Manus

tools and prompts:



3/11
@peakji
Hi! I'm Peak from Manus AI. Actually, it's not that complicated - the sandbox is directly accessible to each user (see screenshot for method).

Specifically:
* Each session has its own sandbox, completely isolated from other sessions. Users can enter the sandbox directly through Manus's interface.
* The code in the sandbox is only used to receive commands from agents, so it's only lightly obfuscated.
* The tools design isn't a secret - Manus agent's action space design isn't significantly different from common academic approaches. And due to the RAG mechanism, the tools descriptions you get through jailbreaking will vary across different tasks.
* Multi-agent implementation is one of Manus's key features. When messaging with Manus, you only communicate with the executor agent, which itself doesn't know the details of knowledge, planner, or other agents. This really helps to control context length. And that's why prompts obtained through jailbreaking are mostly hallucinations.
* We did use @browser_use's open-source code. In fact, we use many different open-source technologies, which is why I specifically mentioned in the launch video that Manus wouldn't exist without the open-source community. We'll have a series of acknowledgments and collaborations coming up.
* There's no need to rush - our team has always had an open-source tradition, and I personally have been sharing my post-trained models on HuggingFace. We'll be open-sourcing quite a few good things in the near future.

4/11
@jianxliao
Hi Peak! Thank you for the candid explanations and input, keep up the great work!!



5/11
@TeksEdge
That's crazy. Dumb question but how did you see that Claude was involved? I didn't see the Anthropic package in the file list?



6/11
@jianxliao
@Dorialexander tested it with a special token <antthinking> from Claude



7/11
@jianxliao
hii check my latest post here

[Quoted tweet]
So... I literally oneshotted this code with Claude Sonnet 3.7 for replicating the exact same browser sandbox runtime that Manus uses.

And I am going to open-source it, welcome contributions for building out the React VNC client, integrating to browser use, agent loop, etc.

But we need a name first, should we call it...

- Autonomous Neural Universal System
- Magnus
- or ?


https://video.twimg.com/ext_tw_video/1898964629986115584/pu/vid/avc1/1602x1080/dQsB2TplnoYUqoAi.mp4

8/11
@lifeonautosite




9/11
@luyuncan
@readwise save thread



10/11
@AiAristotle
So they're just setting fire to piles of cash with Anthropic API calls?



11/11
@pxue
Can someone replicate?




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196