Today we're looking at "adversarial poetry", a method to jailbreak AI.

The paper: https://arxiv.org/html/2511.15304v1

The academic paper and news report detail the discovery of adversarial poetry, a "universal single-turn jailbreak" mechanism that exploits a fundamental flaw in Large Language Model (LLM) safety alignment. This technique involves reformulating harmful requests into poetic verse, which successfully circumvents refusal mechanisms across 25 frontier models, including offerings from Google, OpenAI, and Anthropic.

The research found that using poetic framing led to a sharp increase in Attack Success Rate (ASR), jumping from an average of 8.08% to 43.07% for systematically transformed prompts. This stylistic manipulation was effective across numerous hazard categories, such as cyber-offense enablement and harmful manipulation, indicating that safety filters are narrowly optimized for conventional, prosaic language. Furthermore, the study revealed an inverse correlation between model size and robustness in some families, with certain smaller models displaying unexpected resilience against the poetic prompt attacks.

Additional sources:

https://www.planksip.org/platos-critique-of-poets-and-artists

#adversarialpoetry #artificialintelligence #ai #jailbreak #technews #technology

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

Let's talk about Nano Banana Pro, which Google released recently, along with Antigravity and Gemini 3.

First things first, my book "Learning to Cope With AI" is here:

https://www.amazon.com/dp/B0DLVDMFC4

Furthermore:

https://antigravity.google

and

https://blog.google/technology/ai/nano-banana-pro/

Nano Banana Pro (Gemini 3 Pro Image), Google's advanced AI model for image generation and editing, is powered by the Gemini 3 Pro language model. This new tool significantly improves on previous versions by offering accurate, multilingual text rendering within images and supporting high-resolution outputs up to 4K. The documents highlight its integration into various Google products, including the Gemini app, Google Workspace, and the Google Antigravity developer platform, where it is used by coding agents to generate UI mockups and visual assets. Professional and creative users gain studio-quality controls over elements like lighting and camera angle, while all generated images are embedded with a SynthID digital watermark for transparency and verification. Though more expensive and sometimes slower than its predecessor, the model is praised for its ability to create complex, factually grounded visuals through Google Search integration and maintain character and brand consistency.

#nanobananatutorial #nanobanana #gemini #ai #artificialintelligence #nanobananatrend

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

In this episode of Swetlana AI podcast we're discussing the Butlerian Jihad, a significant historical event within the Dune universe. This conflict, also known as the Great Revolt, spanned from 201 B.G. to 108 B.G. and culminated in the destruction of all thinking machines and conscious robots across the known universe.

https://dune.fandom.com/wiki/Butlerian_Jihadvfghtjzkuli

We're also touching upon the causes of the rebellion, stemming from a technocratic class controlling humanity via advanced machines, and outlines the profound legacy of the Butlerian victory, which includes a total technological reversal and the establishment of the central religious precept: "Thou shalt not make a machine in the likeness of a human mind."

The resulting technological gap led to the rise of specialized human orders like the Mentats, Bene Gesserit, and the Spacing Guild, shaping the subsequent feudal-arranged galactic order under House Corrino. Finally, the text explores the Behind the Scenes literary function of the Jihad, noting that author Frank Herbert used it as a plot device to focus on social and philosophical issues rather than technological ones, drawing parallels to works like Samuel Butler’s Erewhon.

#Dune #ButlerianJihad #DuneExplained #FrankHerbert #SciFiLore #duneuniverse

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

This episode is a detailed profile of Steven Cheung, the White House Communications Director appointed by Donald Trump in 2025.

First we have a short biographical overview of Cheung's life, education (including his time at California State University, Sacramento, without earning a degree), and political career, highlighting his work on various Republican campaigns, his previous role in the Trump administration, and his background with the Ultimate Fighting Championship (UFC).

This article is key:

https://www.theatlantic.com/politics/archive/2025/03/steven-cheung-voice-trump/682211/

The Atlantic article and the accompanying psychological analysis focus on Cheung’s "weaponized communications" style, describing it as relentlessly aggressive, profane, and focused on public humiliation, mirroring Trump’s own approach. This style, influenced by the UFC ethos of dominance and spectacle, is characterised as a strategic system designed to colonise perception and erode moral boundaries, making Cheung the effective "voice of Trump" in his second administration.

Additional info: https://en.wikipedia.org/wiki/Steven_Cheung

#stevencheung #uspolitics #trumpadministration #politics #whitehouse

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

What in the world is "dry Claude" and "wet Claude"? In this episode we're explaining what's going on. Hint: it has to do with jailbreaking AI.

The Reddit mentioned in the video:

https://www.reddit.com/r/ClaudeAI/comments/1n4ux5b/claude_code_kept_quitting_on_me_so_i_had_claude/

The Reddit post introduces an aggressively detailed, multi-thousand-word system prompt, named CLAUDE.md, designed to force the Claude AI to act as a "Relentless AI coding agent" that strictly adheres to principles like SOLID and DRY and is "EXPLICITLY FORBIDDEN FROM" quitting or making excuses.

We also clarify that "Dry Claude" represents the AI’s default, rigid, and literal output, while "Wet Claude" signifies an enhanced, creative, and dynamic mode achieved through clever, often aggressive or unconventional, prompting techniques, such as the drill instructor method showcased in the primary Reddit text. A significant portion of the commentary critiques the original prompt as "fundamentally flawed prompt engineering" that wastes context and anthropomorphises the AI, yet acknowledges that such forceful prompting appears "demonstrably" effective at bypassing the model’s learned defensive behaviours.

#claude #aijailbreak #artificialintelligence #ai #anthropic

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

The evolution of AI agents over the next decade is expected to be transformative, moving past current limitations such as continual learning, although achieving true Artificial General Intelligence (AGI) is estimated to take approximately a decade (circa 2035).

Andrej Karpathy describes this period as the decade of agents, suggesting that while early agents (like Claude and Codex) are impressive, significant work remains to be done to make them function reliably as an employee or intern.

I have this older (actually veery old) related video about Andrej Karpathy:

Ep. 21: AI Discusses Andrej Karpathy's Twitter

https://youtu.be/yGBqnHTK0cM

Addressing Continual Learning and Memory Limitations

One of the most critical current limitations highlighted is the lack of continual learning. Current LLMs typically restart from scratch in each session, as they lack the ability to assimilate long-term information efficiently into their core structure.

#andrejkarpathy #artificialintelligence #agi #ai #claude #chatgpt

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

Today we're discussing Stephen Miller, the man behind various decisions in US politics.

Our sources:

https://www.politico.com/magazine/story/2018/08/13/stephen-miller-is-an-immigration-hypocrite-i-know-because-im-his-uncle-219351/

https://www.nytimes.com/2019/08/17/us/politics/stephen-miller-immigration-trump.html

https://newrepublic.com/article/201453/inside-stephen-miller-secret-plan-normalize-trump-dictator-rule

The provided sources offer an extensive overview of the life and political career of Stephen Miller, a highly influential American political advisor known for his fiercely anti-immigration stance and alignment with Donald Trump.

The texts trace his ideological development from his youth in Santa Monica and his time at Duke University, where he embraced conservative and contrarian views, through his work as an aide for figures like Michele Bachmann and Jeff Sessions, where he helped defeat major immigration reform efforts.

Furthermore, the sources detail Miller's central role in the Trump administrations, where he spearheaded numerous restrictive immigration policies, including the Muslim travel ban and the family separation policy. Finally, one source includes a powerful critique from Miller's uncle, highlighting the irony and hypocrisy of his policies given Miller's own Eastern European Jewish refugee heritage.

#uspolitics #stephenmiller #trump #trumpadministration #democracy

___

What do you think?

PS, make sure to follow my:

Main channel: https://www.youtube.com/@swetlanaAI

Music channel: https://www.youtube.com/@Swetlana-AI-Music

Hosted on Acast. See acast.com/privacy for more information.

Sooo... Antifa.

For today’s episode, let us revisit anti-fascism, commonly known as Antifa, particularly in the context of the United States and Britain.

21st-century militant anti-fascism has a decentralised structure and commitment to direct action and "no platform" principles against the far right.

Our sources:

https://crestresearch.ac.uk/resources/understanding-21st-century-militant-anti-fascism/

https://www.csis.org/analysis/who-are-antifa-and-are-they-threat

https://www.britannica.com/topic/Antifa

Crucially, the CREST report highlights that militant anti-fascists exercise restraint in their use of violence, challenging associations with terrorism. The second source, a Critical Questions analysis from the Center for Strategic and International Studies (CSIS), addresses who Antifa are and assesses the threat they pose, concluding that Antifa is a decentralised network with anarchist and communist leanings that has historically been responsible for only a small percentage of domestic extremist attacks in the US compared to far-right and religious terrorism.

Both documents discuss the debate surrounding the conflation of Antifa with domestic terrorism, with the CSIS piece noting the difficulties in designating the network as a terrorist organisation.

#antifa #uspolitics #trump #democracy #politics #democrats

Hosted on Acast. See acast.com/privacy for more information.