Hey there, folks! As this is the first edition of The Neuron, I have included some introductions in each section in italics font. Also, I’m still learning and experimenting, so I expect this newsletter’s format to change a lot over time.
In this section, I’ll talk a bit about the latest trends and innovations in the field of AI, or tech in general. Expect some blog article links, news, etc.
Google DeepMind recently published a research paper about their new V2A (video-to-audio) technology. Video-generating models like Sora are on the rise but most of the models generally produce silent videos. The V2A technology takes in a natural language prompt and the video itself and then it generates an amazing audio track for that video.
Their blog mentions the technology along with some very cool examples. Here’s one of them —
Prompt for audio: Cinematic, thriller, horror film, music, tension, ambience, footsteps on concrete
Check our the full blog here: https://deepmind.google/discover/blog/generating-audio-for-video/
Starting 18th of June, Gemini is available as an application in India. The amazing part is that it’s available in 9 languages other than English.
The article mentions a lot about ‘Gemini Advanced’ as well, I’m curious to see how it performs in comparison to GPT-4o. The fact that GPT-4o is available to users who don’t have upgraded yet already gives it brownie points.
I might start using Gemini daily because it’s an alternative to Google Assistant. Yep, I have already switched and let’s see how it works. For other AI-related tasks, for sure I’m sticking to ChatGPT for now.
Checkout the full blog from Google here: https://blog.google/intl/en-in/company-news/technology/gemini-in-india-now-on-mobile-multilingual-and-more-powerful-for-your-everyday-tasks/
This section will include all the stuff I have been learning recently. I will occasionally include GitHub and other links to my projects.
So I recently got inspired by maharshi a lot and decided to build something from scratch. Here’s what I ended up with:
Inclinet: A deep learning library implemented from scratch in Python and Numpy. You can implement multilayer (artificial) neural networks using inclinet. It contains —
Linear layer and Sigmoid, Tanh, (leaky) ReLU activation functions.
Stochastic gradient descent optimizer.
Mean squared error loss.
Heres’ the announcement tweet and it also has a GitHub repository (I’d be grateful if you could drop a star).
Next, I’m thinking of adding the SoftMax activation function and then add an example of the MNIST dataset.
I’m sure a lot of you must have heard about it till now. I’ll be participating in nights & weekends s5 to build something cool. I don’t have an idea yet, but let’s see how it goes.
The first lecture is on 18th June, 11 pm IST. I’m looking for folks to connect with this season and build with them. If you’re one of them, please let me know.
Spectreseek for life.
In this section, I’ll particularly talk about some awesome videos, podcasts, and projects I found over the week.
In this interview cum podcast, Francois, the creator of Keras, talks about AGI and benchmarks. He puts forth the point that “LLMs won’t lead to AGI” and explains how LLMs are just matching patterns and how much more is needed to reach AGI.
The point discussed is how the current benchmarks fail to measure the intelligence of LLMs. Francois himself has made a benchmarking test, called ARC-AGI which rather measures intelligence. Here’s the description of ARC-AGI from the website itself —
Most AI benchmarks measure skill. But skill is not intelligence. General intelligence is the ability to efficiently acquire new skills. Chollet's unbeaten 2019 Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI) is the only formal benchmark of AGI.
It's easy for humans, but hard for AI.
ARC-Prize is $10,00,000 prize money for any LLM that beats the ARC-AGI challenge. I found the website of ARC-Prize very cool and you should check it out: https://arcprize.org/
This was one of the best podcasts I have heard in a while. I highly recommend the podcast to anyone who has even a little bit of interest in AI. You can find it here:
Here’s an amazing thread from Akshay Pachaar that consists of various articles explaining various concepts. I highly recommend you check this out if you’re working/studying in the domain of AI.
Here’s the thread: https://x.com/akshay_pachaar/status/1799056220776771765
That’s it for this week! As this is the very first edition, your feedback is extremely crucial. I would love to hear from you. Also, expect changes in format, writing style or even in the idea of the whole newsletter as I continue to refine it.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.