A pure combinational logic byte level RNN will exist by end of 2024
Plus
5
Ṁ135Jan 1
44%
chance
1D
1W
1M
ALL
Similer to https://manifold.markets/Amaryllis/a-pure-binary-llm-will-exist-by-end however must also be:
- Byte level: Uses 8 bit utf8 bytes as tokens.
- RNN: Each step takes in an array of S bits of state and 8 bits input, and outputs a distribution for the next byte and S bits of state.
As before, bits per byte must be at least as good as GPT-2 (1.225 BPB) on The Pile. It does not matter how the combinational logic is created. It is permitted to unroll integer or floating point math. However the resulting logic graph must exist and be evaluable on a commodity computer with at most 1TB of RAM.
This question is managed and resolved by Manifold.
Get
1,000
and3.00
Related questions
Related questions
A pure binary LLM will exist by end of 2024
9% chance
Will we see most new language models shifting to addition-only architectures like BitNet/BitNet 1.58b in 2024?
43% chance
Breakthrough in symbolic regression by the end of 2025?
39% chance
Will a reliable useful neural device that transmits information about users thoughts be released before the end of 2028?
50% chance
an LLM as capable as GPT-4 runs on one 4090 by March 2025
31% chance
At least one of the most powerful neural nets at end of 2026 will be trained using 10^27 FLOPs
81% chance
Will major tech company announce functional brain-computer interface able to read complex human thoughts by end of 2030?
25% chance
Will LLM inference for the largest models run on analogue circuitry as the primary element of computuation by end 2028?
19% chance
At least one of the most powerful neural nets at end of 2026 will be trained using 10^26 FLOPs
86% chance