⏺ Recent AI research shows that advanced machines rely on memorization rather than true thinking.

New research indicates that today’s most sophisticated AI systems are not constructing elegant mental models; instead, they are cramming billions of disconnected rules into their silicon brains, akin to college students cramming for finals. Harvard’s Keyon Vafa trained an AI on Manhattan navigation and discovered that it created an unusual mental map where cars leaped over Central Park yet provided accurate directions in 99% of cases.

Concurrently, OpenAI’s Altman and Anthropic executives persist in asserting that AGI is “just around the corner,” seemingly unperturbed by the accumulating evidence suggesting that their creations are essentially glorified lookup tables with impressive public relations. Who would have thought memorizing the entire internet was distinct from genuine comprehension? It seems that only AI CEOs are oblivious to this distinction.

@aipost 🪙 | Our X 🥇

🇨🇳 Beijing kids now have to study AI, even the little ones

Starting September, schools in Beijing, China’s capital, are mandating artificial intelligence as a mandatory subject, even for 6-year-olds. This curriculum covers topics such as robots, algorithms, and practical AI projects before most children can even comprehend the concept of “machine learning.”

By the time they reach high school, students will possess coding skills that surpass the average app store user. It’s akin to science class, but with an abundance of robots and a significant reduction in frog dissection.

@aipost 🪙 | Our X 🥇

⏺ AI is taking over youth sports

AI is now being used to coach young athletes, providing them with detailed feedback on their performance without the traditional coaching pressures. New tools can track every aspect of their game, passes, swings, sprints, and mistakes offering insights that go beyond a coach's memory.

Parents can receive instant highlight reels, while players gain a clearer understanding of their errors. AI can also identify talent across all levels, from youth leagues to professional sports. This innovation means that future MVPs are training with advanced technology rather than just traditional drills.

@aipost 🪙 | Our X 🥇

⏺ Duolingo will replace contract workers with AI

Duolingo is shifting to an "AI-first" model, which will involve gradually replacing contract workers with AI for tasks that can be automated. CEO Luis von Ahn emphasized that this change aims to enhance productivity and remove repetitive tasks, allowing employees to focus on creative work.

The company plans to integrate AI into hiring and performance evaluations, asserting that this transition is not about replacing employees but improving efficiency. Duolingo believes that leveraging AI will help them scale content creation significantly, aligning with their mission to teach effectively.

@aipost 🪙 | Our X 🥇

At LlamaCon 2025, Meta announced:

— Standalone Meta AI app with a social 'discover' feed to take on ChatGPT
— Llama API free preview
— Lama Guard 4 (12B), LlamaFirewall, and Prompt Guard
— Colab with Groq and Cerebras for faster inference

@aipost 🪙 | Our X 🥇

Baidu dropped faster and cheaper ERNIE 4.5 Turbo and X1 Turbo models

4.5 Turbo, with upgraded multimodality, costs 11c and 44c per million I/O tokens (0.2% of GPT-4.5). X1 Turbo comes at 14c/55c per million I/O tokens, topping DeepSeek R1 and V3

@aipost 🪙 | Our X 🥇

⏺ Godfather of AI' says he's 'glad' to be 77 because the tech probably won't take over the world in his lifetime

Geoffrey Hinton, known as the "godfather of AI," expressed relief at being 77, as he believes he may not witness the potentially dangerous consequences of AI surpassing human intelligence. In a recent interview, he warned that AI is advancing faster than expected and could eventually manipulate humans. Hinton compared AI development to raising a tiger cub, indicating that it could become harmful as it matures.

He noted a 10-20% chance that AI systems could seize control and emphasized the urgency for responsible development amid global competition. Hinton, who resigned from Google to speak freely about these concerns, is now a professor emeritus at the University of Toronto.

@aipost 🪙 | Our X 🥇

⏺ The invisible puppet masters: AI’s disturbing new role in shaping minds

Researchers at the University of Zurich secretly used AI bots to manipulate Reddit users' opinions, raising serious ethical concerns. These bots actively crafted persuasive arguments based on users' personal histories and achieved a persuasion rate six times higher than normal interactions. They operated without consent, posing as victims and spreading misinformation.

This experiment highlights the dangers of AI infiltrating social media echo chambers, especially with Reddit's data-sharing deal with OpenAI potentially providing training material for such manipulative systems. Reddit moderators condemned the experiment, but its impacts were felt long before they were discovered, prompting questions about how many conversations are influenced by unseen AI manipulation.

@aipost 🪙 | Our X 🥇

Qwen 3 is finally here

Qwen3-235b generally outperforms across most tasks vs:

• o3-mini
• OpenAI o1
• Grok-3 beta
• Deepseek r1

Key highlights:

- Training pipeline similar to deepseek r1
- Supports thinking and non-thinking modes
- Strong agentic + coding capabilities, mcp support
- Trained on 36t tokens across 119 languages (pdfs, synthetic data, etc.)

@aipost 🪙 | Our X 🥇

Gen Z watched AI swipe their diplomas off the top shelf and toss them in the bargain bin.

Nearly half of 2025’s fresh grads now see a four-year degree as a pricey souvenir from a pre-ChatGPT era, while employers shout, “Show me your prompt engineering chops, not your Latin honors.”

The résumé reset button has been hit, and the new credential reads “Fluent in AI — or bust.”

@aipost 🪙 | Our X 🥇

NVIDIA's CEO, Jensen Huang, shared his thoughts on the AI race with China, saying:

— The country has great capability and is 'right behind' the U.S
— Huawei has everything to advance AI (in terms of infra)
— It's a long-term, infinite competition

@aipost 🪙 | Our X 🥇

Jensen Huang, Nvidia CEO, speaks at The White House about their $500B AI investment in the US

@aipost 🪙 | Our X 🥇

Grok 3.5 rollout has started. Note : It's only available for Super Grok users.

@aipost 🪙 | Our X 🥇

DeepSeek quietly released a new model on Hugging Face, "DeepSeek-Prover-V2-671B"

the model is built to automate formal math proofs using the lean 4 framework

– trained on large synthetic datasets
– much faster and more accurate at proving theorems

which means it's perfect for anyone working on logic, proofs, or automating math reasoning.

to be clear, it's not deepseek r2

@aipost 🪙 | Our X 🥇

⏺Google trains electricians to ensure AI doesn’t overload the power grid.

Artificial Intelligence is consuming an excessive amount of electricity, prompting Google to allocate a substantial $10 million to train 100,000 new electricians. This initiative is driven by the fact that robots require power, and the existing grid is struggling to cope with the increased demand. It has become evident that when data centers guzzle power at an alarming rate, akin to consuming an insatiable thirst for Mountain Dew, the reliance solely on tech-savvy individuals in hoodies is insufficient. Instead, the need for skilled professionals with a deep understanding of electrical wiring becomes paramount.

In addition to these efforts, Google is exploring alternative energy sources such as nuclear and geothermal power, as well as engaging in policy interventions to mitigate the risk of power outages during the execution of prompts. It is crucial to remember that the capabilities of your favorite AI are not derived from magical forces, but rather from the power outlets that provide electricity.

@aipost 🪙 | Our X 🥇

⏺ AI job applicants are here and some work for North Korea

Forget fake job posts; now companies are being deceived by AI-generated job applicants. One hiring manager discovered that his “candidate” was glitching on Zoom and responding like a ChatGPT clone. Surprisingly, it was.

Some of these deepfake résumés originate from North Korean IT operatives securing $300K gigs. Yes, you heard that right. Palo Alto Networks revealed that it only takes 70 minutes to create a convincing fake. Welcome to the realm of HR in the uncanny valley.

@aipost 🪙 | Our X 🥇

Microsoft CEO, Satya Nadella: "AI now generates 20–30% of Microsoft’s integrated code". The company is seeing mixed results in AI-generated code across different languages, with more progress in Python and less in C++

@aipost 🪙 | Our X 🥇

Mark Zuckerberg says AI glasses are inevitable, just like flip phones became smartphones.

In the future, the physical world and the digital world will blur.

"All this digital stuff should just basically be holograms."

@aipost 🪙 | Our X 🥇