Dagmawi Babi

Good nights ✨ "ፈዋሻችሁ እኔ እግዚአብሔር ነኝና።” — ዘፀአት 15:26 #Scripture @Dagmawi_Babi

Enjoy this song: Medhanite by Abraham Halake

And that's my time, ✌️

And here're some interesting projects I've seen other people build on top of these connections. • We taught a fruit fly to read human emotions • Fly Escape Room - the world's first game with NPCs powered by a real fruit fly brain, all 166k neurons. • Animated group of Connectomes

Recently Google researchers mapped wiring diagram of the adult fruit-fly central nervous system — a connectome. From electron-microscope volumes they reconstructed each neuron’s 3D shape and the chemical synapses between them. On the order of 140,000 neurons and tens of millions of synapses. It covers eyes, antennal lobe, mushroom body, central complex, and the rest of the brain — the parts for seeing, smelling, learning, and steering. The data is available online so I vibe-coded this simulation app that uses the full neural connections. Named cell types from FlyWire + MANC drive a NeuroMechFly-style body — walk, turn, jump, track odor, take off. It feels weird seeing the neurons react to being clicked on, smell, light, obstacles and more. Very odd. Anyways it's open-source and live. #MyProjects #VibeCoding @Dagmawi_Babi

Since it's Apple's day, there is one more unrelated thing I want to show you...

Good nights ✨ "ፈዋሻችሁ እኔ እግዚአብሔር ነኝና።” -- ዘፀአት 15:26 #Scripture @Dagmawi_Babi

Enjoy this song: Medhanite by Abraham Halake

Subscribe to ScholarXIV.com

Better-HuggingFace @ScholarXIV

Better-HuggingFace @ScholarXIV :)

I hate the interface of HuggingFace, it looks so vibe-coded even though it's not. Anyways we've got all the resources to build Better-HuggingFace in ScholarXIV and we did and we will launch it in our next Launch Week. You will be able to host datasets, explore from multiple sources, and even explore open AI models. @ScholarXIV

Anyways that should've been a blog I guess lol

Telegram is an amazing goldmine for data. You don't even have to bother yourself with private chats. Public channels and groups are enough. A while back I feared telegram would take the export feature away and I was exporting a few notable channels for public access. If you've got some exports, or would like to share your channel's export, or just want to help out build this dataset just export public channels and send them in the comments and I'll publish it to Kaggle. These exports are extremely useful for fine-tuning and model training. #Datasets @Dagmawi_Babi

And this was the reaction of the person I fine-tuned a model to text like of.

Here are some interesting facts for you, as of 2026 53% of all web traffic comes from bots. 40% of those bots are bad bots used for scamming/spamming and impersonating. 20% of social media users are also bots. Then 30% of all written webpages are AI generated. On YouTube 20% of the videos recommended to new users are AI generated (slops), Meta already has celebrity clone AIs. The fact of the matter is we're stepping into a future of the internet where humans are a minority. I can't even imagine what types of movies, books, and songs are going to be produced. What's even worse is that all of these stats are deemed to increase cause the vast majority of people don't have access to these AI services and isn't even aware of them. So you can image the more people know how much slop will be on the internet. Originality is going to be something interesting.

The interesting part is that the platforms and resources are now available for everyone to be able to do what I just did. Heck, if you and I have talked a lot you can even clone me. So far I was only able to fine-tune a model, host it and text to it through a Telegram bot and it would only respond in text. But this can be taken further. Remember when Addis AI cloned my voice a while back, and now that we have open Amharic datasets from Snapware you can use 15 seconds of someone voice (or their voice message from Telegram) and then get it to sound exactly like them with great support for Amharic. For a long while we could have cloned someone's digital persona if they were exclusively an English speaker, but the resources and platforms have now reached a point where even Amharic speakers can be cloned. You can take it even more further now that your bot can text and sound like the other person. You can attach an image/video generator model, give it a few pics of the person and it would make you extremely accurate looking pictures of them. Essentially cloning what any one is able to do through a chat message. Now finally available in Ethiopia, for everyone to use. Welcome to 2019 EC, very interesting times ahead. So much to think about as well.