iirc one of the main results from Heggarty is that Indo-Iranian and the Indic branch itself are dated to before the IVC. That would mean the IVC inhabitants spoke an early Vedic language, which would explain the humped zebu bulls that the Vedas talk about and give importance to. Is that right?
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset.
72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute.
What a time to be alive…
Descartes' Discourse on the Method of Rightly Conducting One's Reason and of Seeking Truth in the Sciences is deeply enlightening. I started reading it for a personal agentic AI project, but I soon got hooked on the life insights he drops along the way. Reading his work feels like finishing a meditation session.
if the MIT post about Astra capabilities got you interested in weight editing, i have an old post from 2023 about doing exactly this to make a transformer from scratch for a simple sequence prediction task.
https://t.co/v3bB5DfrF0
i've reproduced the results from the MIT blog post today on weight-space programming. but one additional result i have to add: this is a new, step-change emerging capability that's appeared in just the latest set of models.
Note the Twitter thread doesn't include the *failures* of LLMs to improve other LLMs.
So edits help for Qwen 3.5, which *has the preexisting visible flaw* of reaching the token limit in 93% of cases; but edits don't help Gemma 270m, which doesn't have this flaw. --