Artificial Superintelligence won't happen any time soon

Soon, we’ll be chatting about a new buzzword: “agentic AI”

DeepSeek’s new AI thinks it’s ChatGPT. Training shortcuts may explain—and compromise—its trustworthiness
https://techcrunch.com/2024/12/27/why-deepseeks-new-ai-model-thinks-its-chatgpt/

DeepSeek’s new AI thinks it’s ChatGPT. Training shortcuts may explain—and compromise—its trustworthiness
https://techcrunch.com/2024/12/27/why-deepseeks-new-ai-model-thinks-its-chatgpt/

Generative AI has dominated 2024, with advancements in LLMs, custom chatbots, and content generation. But a new paradigm is emerging—Agentic AI

Former Google CEO Eric Schmidt warns it's time to reconsider AI development if it starts to self-improve—sounds like we're stepping into sci-fi territory https://www.windowscentral.com/software-apps/former-google-lead-seriously-think-about-pulling-the-plug-on-ai

Meta FAIR's latest research reveals that Large Concept Models (LCM) revolutionize language modeling by separating reasoning from language representation, much like how we form complex thoughts for communication. Discover more in the full paper:
http://go.fb.me/mn2vvw

What's your opinion on Google’s newest and most advanced AI text to video Veo2 model?

Reasoning AI models are the new frontier in AI, with OpenAI's o1 model leading the charge. Unlike regular AI, these models "check their work" mid-process, aiming for smarter decisions. But they’re costly — OpenAI charges $60 for 750K words, 6x more than GPT-4o.

What is the tech behind ChatGPT?
ChatGPT runs on GPT (Generative Pre-trained Transformer) – the magic behind its human-like responses. This thread breaks down what GPT is, how it works, & its evolution from GPT-1 to GPT-4. Let’s dive in! 👇

Microsoft’s Phi-4 14B parameter LLM surpasses GPT-4 on STEM QA. Phi-4 uses multi-agent workflows & self-revision techniques, making it the best open-weight model for tasks like coding (HumanEval) & math. Here's what's new!

Meta AI has introduced the Byte Latent Transformer (BLT) – a game-changer in LLMs. Unlike traditional models that use tokenization, BLT processes raw byte sequences directly, improving scalability, efficiency, and robustness. Here's how it works! 👇

Google has unveiled a new chip which it claims takes five minutes to solve a problem that would currently take the world's fastest super computers ten septillion – or 10,000,000,000,000,000,000,000,000 years – to complete 🤯

How SORA Works (text to video model from openai)

- Trained on video datasets + captions
- Encodes video into "latent space" (concepts, not pixels)
- Denoises step-by-step using diffusion
- Ensures 3D consistency w/ Transformers
- Decodes latent space back to video

What's your opinion on AI influencers?

I was recently suspended from X(no one is answering btw) so I came here to share my journey and thoughts!