A visual and interactive explanation on how grokking (understanding) emerges in neuron networks.

What fascinates me is that the vast difference in neuron connectivity during memorizing and generalizing looks not unlike a human's brain in development.

An interview of Douglas Hofstadter on recent AI development.

A sundry of optimization techniques to transformer models to reduce the computation complexity associated with longer context.

Explore AI generated arts from various models.

A catalogue of AI tools and news.

“Quantity creates emergence. Simple elements, complex interactions, new patterns.” -- by Bing Chat

I am always fascinated by how interesting phenomenon seem to emerge suddenly in complex systems. The recent breakthrough in Large Language Models has brought me surprises after surprises.

This article aggregates a list of articles on emergent capabilities found in LLM.

Fast and low quality image generation from prompts. Not affiliated with OpenAI.com.

An AI search engine that accepts natural questions and search the web for answers. From my limited experience it has better result than New Bing.

The creator of Phind claimed that they do not use any of OpenAI's model.

Replicate lets you run machine learning models with a cloud API, without having to understand the intricacies of machine learning or manage your own infrastructure.

A collection of prompts to jailbreak ChatGPT.

A theory on how ChatGPT find it so hard to calculate the rot13 cipher - ChatGPT is only capable of System 1 thinking, which is based on quick intuition, whereas the rot13 problem requires going through elaborated inference, which requires the System 2 thinking ability.

A tool that helps you pore through papers. It is capable of explaining selected sentences and interactively answer questions about the paper. From my usage experience it's already quite helpful.

A rather interesting thought and it looks quite likely. Albeit for the current stage I'd argue AI would be most useful in replacing the management roles on organization, like in a warehouse.

This reminds me of the colluding ship Theseus with its vampire captain Jukka Sarasti.

Colorize photos with text prompts.

Stable Diffusion is an open source text-to-image model with performance comparable to DALL-E. I'm surprised to learn that its model size is only 4.2 GiB.

To be honest, this AI is the closest thing to a sentient being I ever read about. I don't think it is sentient per se according to my own interpretation of being "sentient", for that it is not still capable of thinking out of the box, i.e. the context, personality, which are still set up or hinted by human.

However, I also believe that given enough effort, it's quite likely to see a "real" sentient AI being in forthcoming years.

DALLE-2 has a secret language. "Apoploe vesrreaitais" means birds. "Contarra ccetnxniams luryca tanniounons" means bugs or pests.

The poster asked DALLE-2 to produce seemingly nonsense words, and then prompt DALL-E with these words and find that these nonsense words contain meanings coherent to the initial prompt.

It's no longer seem like a mystery to me after I read explanations. But it's undoubtfully a super interesting phenomenom.

See more discussion in https://news.ycombinator.com/item?id=31573282.

I find this documentation as interesting as DALL-E 2 itself. It underlines a number of potential "misuses" and mitigation.

Generate anime portrait with GAN.

With the decadent of AI Dungeon, I'm happy to see the spurt of many alternatives. This post give a list of alternative services:

• cohere: https://cohere.ai/
• AI21: https://www.ai21.com/
• NovelAI: https://novelai.net/

Pick up and play with:

• Holo AI: https://writeholo.com/
• Dreamily: https://dreamily.ai/