Aleksa Gordić LinkedIn Posts

In my last post on teaching transformers to do arithmetic a lot of the comments were along the lines of "My pocket calculator

State Space Models are quickly positioning themselves as a new alternative to transformers! I had the authors of the seminal Mamba & S4

New player in the AGI space! :) Ilya Sutskever of the AlexNet/ImageNet moment 2012 fame (and a cofounder of OpenAI) just spun off

🚀 A new model from Anthropic just dropped! Things are changing rapidly - a year ago I was exclusively using OpenAI's ChatGPT, now

Imbue just published an extremely informative blog post on how they built their 4000+ H100 machine cluster! They also shared some of the

Before we reach a point of having autonomous embedded AIs (AIs controlling robotic bodies), applications like this where a human operator is remotely

Mind blowing real-time LLaMA 3 + Groq demo. Now extrapolate this 5 years out - even the technology we currently have will change

📚 Here is a list of books I found incredibly insightful over the past year:

--biographies--
* Shoe Dog: A Memoir by the Creator of

[📚 SlovenianGPT 🧠] Happy to announce that I'm open sourcing the best 7B LLM for the Slovenian language! (better than LLaMA3, Mistral, etc.)

Chinese tech companies are getting increasingly impressive.

Until recently when when I was thinking about robots Boston Dynamics' Atlas humanoid robot doing backflips

If you're curious to understand how search engines like Google search work here is really cool recent blog post: https://lnkd.in/dQnwP9qe

Whether you're just curious

"To Code, or Not To Code? Exploring Impact of Code in Pre-training" new paper from Cohere! They explore the impact that code data

New token sampling heuristic for LLMs: Min P! The authors achieved 10-20% better results on GSM8K and GPQA vs top_p at temperature 1.

How it feels coding with AI Copilots in 2024 - shoggoth and me, we're one

🚀 "Transfusion" - a new model from Meta that can generate both text & images (transformer+diffusion)! They get image results that are on

🤯 AI chip game is getting really hot! Cerebras just announced that their Llama 3.1 70B 16-bit inference is 20x faster than GPUs

🧠 Important lmsys Chatbot Arena update: an effort to decouple "style" from "substance" shows a different picture of what the best LLMs are!

So

OpenAI just released a new series of models that are more powerful reasoners! They do that by spending a variable amount of computation

I am back to San Francisco! Looking forward to meeting many friends/startups I met online over the past few years! :)

Had my first

Lots of great insights from this amazing Physics of Language Models lecture! Let's start with scaling laws for knowledge. They show that "sufficiently"