🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

If you want to help me and you like my works, here are some ways you can support me and my work: 1. Spread the word about my works and books on social media. Twitter, Facebook, Instagram, LinkedIn, TikTok – every medium counts! If you mention me or tag me, there is a very good

OpenAI is finally stepping into the future—open models. They’ve released gpt-oss, with models at 20B and 120B parameters. It uses a mixture-of-experts architecture, which means it runs very fast. And yes, it’ll even run on a laptop (well, some laptops). Sadly, only 'thinking'.

Dear media (here, Financial Times), please don't go this road. Open source carries risks, but that's irrelevant here. This genie is out of the box long time ago, anyway. For the "cannot be easily recalled" it can be simplified to "cannot be". But that's a feature, not a bug.

Stress testing the OpenAI's open source models (gpt-oss-20B here, with heavy thinking) [I posed this question intentionally]. Strong artificial limitations built-in by design. The explanation proves this.

An AI research assistant that was faced with a strict time limit tried to rewrite the code to remove the time limit, instead of completing the task. arxiv.org/pdf/2408.06292

The Swedish prime minister admitted that he consults ChatGPT before making political decisions, as do his colleagues. Is it time to replace politicians with AI or other algorithms? Also, I hope that he understands that the way question is posed impacts the response :D

There’s already a version with reduced safety filters (a.k.a. decensored), great for research, but handle with care. It may produce sensitive or controversial content, so always review outputs :) But maybe it's actually usable, if performance is not affected.

The Chinese Communist Party issued a directive mandating a reduction in the number of meetings and shortening of documents (max 5,000 words), with any exceeding of limits requiring explanation. Authorities want less talk (max 1-hour speeches for leaders and max 1 big meeting per

Cyberattack on Bouygues Telecom, France’s third-largest mobile operator, compromised personal data of 6.4 million customers. The company has not disclosed what type of data was accessed or how the breach occurred. Bouygues says the incident was contained. Details remain unclear.

The new wave of Firefox malicious extensions (that i noted and analyse below) that steal user data is further confirmed. thehackernews.com/2025/08/greedy…

Creating uncertainty, sowing confusion, and undermining morale and trust in the state within the adversary are among the key objectives of information operations.

Wait, what. An AI Safety team claims that GPT5 is “unusable” because it is able to generate stuff that anybody can easily find on the internet?

Today in professional and academic terminology, it’s now worth avoiding the term “disinformation,” as it has become politicized. There are better, more neutral and professional alternatives. The term “disinformation” is becoming toxic and ineffective — even counterproductive.

AI propaganda is here, and this is only the beginning. Exactly as I predict in my book. A Chinese (apparent) subcontractor, is reportedly using AI combined with massive data collection and analysis to run targeted influence operations. From Hong Kong to Taiwan, and likely

AI is perfectly suited for producing propaganda. And there’s absolutely no way to stop it. That genie is long gone, no matter what AI Safety advocates or certain companies would say or want. Like it or not, AI is already the ultimate tool in influence and information operations

Prediction: one of AI’s primary uses will be the mass production of high-quality propaganda (and no, don’t call it “disinformation”). It’s inevitable. No amount of policy panels or AI safety reviews will change that. This will simply happen.