The engram idea is actually technically clever but imo sees the solution from a bottom-up approach while Louf's real argument is a top-down view. His solution (declarative specs) solves that by centralizing the spec, making it versioned and composable, independent of any actual model.
Engram layers just move the coordination problem earlier and lock it in. Coordination problems between models & providers would still exist, requiring a layer injection in each open source model and another variant produced for each. Users would still need to chose between "Qwen-8b" and "Qwen-8b-engram" x model families and sizes. Is that cleaner?
Its an incorrect assumption, the inference speed and particularly the inference speed of the on-device LLMs with which AVs would need to be using is not compatible with the structural requirements of driving.
"Transportation, like software, is accumulated knowledge. The horse embodied centuries of breeding, training, and hard-won understanding about terrain, endurance, and failure. People learned from the horses they rode. Travel improved through incremental refinement, generation after generation. The automobile didn’t appear in a vacuum.
Building all your transportation yourself—whether by breeding horses or assembling a Model T—cuts you off from that accumulated experience. You lose the benefits of thousands of hours spent by others thinking carefully about the same problems.
I have no objection to Model Ts for personal use, but I wouldn’t be one-tenth the traveler I am without constant exposure to well-bred horses.
Some worry cars make horses obsolete—who needs breeders if anyone can buy an engine? I’m more optimistic. As cars proliferate, people will value good horses more. A Model T gets you the first 90%; it’s the last 90%—judgment, robustness, and adaptability—that differentiates."
Essentially the idea of a context window in modern LLM models, there is implicit domain knowledge to every task in which no matter how capable the model may be, if not in the context, the software will not be functional.
TSMC is a for profit business. Why would they care about the moral virtue purity of the applications running on their chips? Seriously illogical statement
new models like Fable were scoring 30%. wouldnt be suprised if very soon we eclipse 50%