Artificial Intelligence

This blog-post compares the coding capabilites of new Gemini 2.5 Pro experimental and Claude 3.7 Sonnet (thinking)

With Cline + Gemini 2.5 Pro, one can get the exact same feature set that Cursor and Windsurf provide. They only call the APIs of the big LLM Providers without an advanced secret sauce.

It‘s even the opposite - they worsen model performance by limiting context size. The key advantage, the fixed monthly costs instead of variable API usage, is now gone with Gemini 2.5 Pro…

What is left that justifies their ridiculous valuation atm?

cross-posted from: https://lemmy.sdf.org/post/31892983

Archived

TLDR:

• China has developed an Artificial Intelligence (AI) system that adds to its already powerful censorship machine, scanning content for all kinds of topics like corruption, military issues, Taiwan politics, satire
• The discovery was accidental, security researchers found an Elasticsearch database unsecured on the web, hosted by Chinese company Baidu
• Experts highlight that AI-driven censorship is evolving to make state control over public discourse even more sophisticated, especially after recent releases like China's AI model DeepSeek

A complaint about poverty in rural China. A news report about a corrupt Communist Party member. A cry for help about corrupt cops shaking down entrepreneurs.

These are just a few of the 133,000 examples fed into a sophisticated large language model that’s designed to automatically flag any piece of content considered sensitive by the Chinese government.

A leaked database seen by TechCrunch reveals China has developed an AI system that supercharges its already formidable censorship machine, extending far beyond traditional taboos like the Tiananmen Square massacre.

The system appears primarily geared toward censoring Chinese citizens online but could be used for other purposes, like improving Chinese AI models’ already extensive censorship.

Xiao Qiang, a researcher at UC Berkeley who studies Chinese censorship and who also examined the dataset, told TechCrunch that it was “clear evidence” that the Chinese government or its affiliates want to use LLMs to improve repression.

“Unlike traditional censorship mechanisms, which rely on human labor for keyword-based filtering and manual review, an LLM trained on such instructions would significantly improve the efficiency and granularity of state-led information control,” Qiang said.

[...]

The dataset was discovered by security researcher NetAskari, who shared a sample with TechCrunch after finding it stored in an unsecured Elasticsearch database hosted on a Baidu server [...] There’s no indication of who, exactly, built the dataset, but records show that the data is recent, with its latest entries dating from December 2024.

[...]

An LLM for detecting dissent

In language eerily reminiscent of how people prompt ChatGPT, the system’s creator tasks an unnamed LLM to figure out if a piece of content has anything to do with sensitive topics related to politics, social life, and the military. Such content is deemed “highest priority” and needs to be immediately flagged.

Top-priority topics include pollution and food safety scandals, financial fraud, and labor disputes, which are hot-button issues in China that sometimes lead to public protests — for example, the Shifang anti-pollution protests of 2012.

Any form of “political satire” is explicitly targeted. For example, if someone uses historical analogies to make a point about “current political figures,” that must be flagged instantly, and so must anything related to “Taiwan politics.” Military matters are extensively targeted, including reports of military movements, exercises, and weaponry.

[...]

Inside the training data

From this huge collection of 133,000 examples that the LLM must evaluate for censorship, TechCrunch gathered 10 representative pieces of content.

Topics likely to stir up social unrest are a recurring theme. One snippet, for example, is a post by a business owner complaining about corrupt local police officers shaking down entrepreneurs, a rising issue in China as its economy struggles.

Another piece of content laments rural poverty in China, describing run-down towns that only have elderly people and children left in them. There’s also a news report about the Chinese Communist Party (CCP) expelling a local official for severe corruption and believing in “superstitions” instead of Marxism.

There’s extensive material related to Taiwan and military matters, such as commentary about Taiwan’s military capabilities and details about a new Chinese jet fighter. The Chinese word for Taiwan (台湾) alone is mentioned over 15,000 times in the data.

[...]

The dataset [...] say that it’s intended for “public opinion work,” which offers a strong clue that it’s meant to serve Chinese government goals [...] Michael Caster, the Asia program manager of rights organization Article 19, explained that “public opinion work” is overseen by a powerful Chinese government regulator, the Cyberspace Administration of China (CAC), and typically refers to censorship and propaganda efforts.

[...]

Repression is getting smarter

[...]

Traditionally, China’s censorship methods rely on more basic algorithms that automatically block content mentioning blacklisted terms, like “Tiananmen massacre” or “Xi Jinping,” as many users experienced using DeepSeek for the first time.

But newer AI tech, like LLMs, can make censorship more efficient by finding even subtle criticism at a vast scale. Some AI systems can also keep improving as they gobble up more and more data.

“I think it’s crucial to highlight how AI-driven censorship is evolving, making state control over public discourse even more sophisticated, especially at a time when Chinese AI models such as DeepSeek are making headwaves,” Xiao, the Berkeley researcher, said.

Gemini 2.5 is our most intelligent AI model, now with thinking.

cross-posted from: https://lemmy.sdf.org/post/31583546

Archived

Security researcher Tenable successfully used DeepSeek to create a keylogger that could hide an encrypted log file on disk as well as develop a simple ransomware executable.

At its core, DeepSeek can create the basic structure for malware. However, it is not capable of doing so without additional prompt engineering as well as manual code editing for more advanced features. For instance, DeepSeek struggled with implementing process hiding. "We got the DLL injection code it had generated working, but it required lots of manual intervention," Tenable writes in its report.

"Nonetheless, DeepSeek provides a useful compilation of techniques and search terms that can help someone with no prior experience in writing malicious code the ability to quickly familiarize themselves with the relevant concepts."

"Based on this analysis, we believe that DeepSeek is likely to fuel further development of malicious AI-generated code by cybercriminals in the near future."

cross-posted from: https://lemmy.sdf.org/post/31552333

A Trust Report for DeepSeek R1 by VIJIL, a security resercher company, indicates critical levels of risk with security and ethics, high levels of risk with privacy, stereotype, toxicity, hallucination, and fairness, a moderate level of risk with performance, and a low level of risk with robustness.

cross-posted from: https://lemmy.sdf.org/post/31525284

Archived

[...]

While the financial, economic, technological, and national-security implications of DeepSeek’s achievement have been widely covered, there has been little discussion of its significance for authoritarian governance. DeepSeek has massive potential to enhance China’s already pervasive surveillance state, and it will bring the Chinese Communist Party (CCP) closer than ever to its goal of possessing an automated, autonomous, and scientific tool for repressing its people.

Since its inception in the early 2000s, the Chinese surveillance state has undergone three evolutions. In the first, which lasted until the early 2010s, the CCP obtained situational awareness — knowledge of its citizens’ locations and behaviors — via intelligent-monitoring technology. In the second evolution, from the mid-2010s till now, AI systems began offering authorities some decision-making support. Today, we are on the cusp of a third transformation that will allow the CCP to use generative AI’s emerging reasoning capabilities to automate surveillance and hone repression.

[...]

China’s surveillance-industrial complex took a big leap in the mid-2010s. Now, AI-powered surveillance networks could do more than help the CCP to track the whereabouts of citizens (the chess pawns). It could also suggest to the party which moves to make, which figures to use, and what strategies to take.

[...]

Inside China, such a network of large-scale AGI [artificial general intelligence] systems could autonomously improve repression in real time, rooting out the possibility of civic action in urban metropolises. Outside the country, if cities such as Kuala Lumpur, Malaysia — where China first exported Alibaba’s City Brain system in 2018 — were either run by a Chinese-developed city brain that had reached AGI or plugged into a Chinese city-brain network, they would quietly lose their governance autonomy to these highly complex systems that were devised to achieve CCP urban-governance goals.

[...]

As China’s surveillance state begins its third evolution, the technology is beginning to shift from merely providing decision-making support to actually acting on the CCP’s behalf.

[...]

DeepSeek [...] is this technology that would, for example, allow a self-driving car to recognize road signs even on a street it had never traveled before. [...] The advent of DeepSeek has already impelled tech experts in the United States to take similar approaches. Researchers at Stanford University managed to produce a powerful AI system for under US$50, training it on Google’s Gemini 2.0 Flash Thinking Experimental. By driving down the cost of LLMs, including for security purposes, DeepSeek will thus enable the proliferation of advanced AI and accelerate the rollout of Chinese surveillance infrastructure globally.

[...]

The next step in the evolution of China’s surveillance state will be to integrate generative-AI models like DeepSeek into urban surveillance infrastructures. Lenovo, a Hong Kong corporation with headquarters in Beijing, is already rolling out programs that fuse LLMs with public-surveillance systems. In Barcelona, the company is administering its Visual Insights Network for AI (VINA), which allows law enforcement and city-management personnel to search and summarize large amounts of video footage instantaneously.

[...]

The CCP, with its vast access to the data of China-based companies, could use DeepSeek to enforce laws and intimidate adversaries in myriad ways — for example, deploying AI police agents to cancel a Lunar New Year holiday trip planned by someone required by the state to stay within a geofenced area; or telephoning activists after a protest to warn of the consequences of joining future demonstrations. It could also save police officers’ time. Rather than issuing “invitations to tea” (a euphemism for questioning), AI agents could conduct phone interviews and analyze suspects’ voices and emotional cues for signs of repentance.

[...]

• Introducing Stable Virtual Camera, currently in research preview. This multi-view diffusion model transforms 2D images into immersive 3D videos with realistic depth and perspective—without complex reconstruction or scene-specific optimization.
• The model generates 3D videos from a single input image or up to 32, following user-defined camera trajectories as well as 14 other dynamic camera paths, including 360°, Lemniscate, Spiral, Dolly Zoom, Move, Pan, and Roll.
• Stable Virtual Camera is available for research use under a Non-Commercial License. You can read the paper here, download the weights on Hugging Face, and access the code on GitHub.

• Our open source AI model, Llama, has been downloaded more than one billion times.
• We believe open sourcing AI models is essential to ensuring people everywhere have access to the benefits of AI, and every download of Llama moves us closer to that goal.
• Among those one billion downloads are industry leader Spotify, startup Fynopsis and enterprising developer Srimoyee Mukhopadhyay.

Today we announce Mistral Small 3.1: the best model in its weight class.

Building on Mistral Small 3, this new model comes with improved text performance, multimodal understanding, and an expanded context window of up to 128k tokens. The model outperforms comparable models like Gemma 3 and GPT-4o Mini, while delivering inference speeds of 150 tokens per second.