techbruv

7 ms·

1.
by techbruv 10mo ago
At a previous job, I recall updating a dependency via poetry would take on the order of ~5-30m. God forbid after 30 minutes something didn’t resolve and you had to wait another 30 minutes to see if the change you made fixed the problem. Was
2.
by techbruv 1y ago
I don’t understand the argument “AI is just XYZ mechanism, therefore it cannot be intelligent”. Does the mechanism really disqualify it from intelligence if behaviorally, you cannot distinguish it from “real” intelligence? I’m not saying th
3.
by techbruv 2y ago
> So it will not get worse in performance but only faster A bit confused by this statement. Speculative decoding does not decrease the performance of the model in terms of "accuracy" or "quality" of output. Mathematic
4.
by techbruv 3y ago
Some other good resources: [0]: The original paper: https://arxiv.org/abs/1706.03762 [1]: Full walkthrough for building a GPT from Scratch: https://www.youtube.com/watch?v=kCc8FmEb1nY [2]: A simple inf
5.
by techbruv 3y ago
The idea that GPT-4 is 1 trillion parameters has been refuted by Sam Altman himself on the Lex Fridman podcast (THIS IS WRONG, SEE CORRECTION BELOW). These days, the largest models that have been trained optimally (in terms of model size w.
6.
by techbruv 3y ago
> "We then train several models from 400M to 15B on the same pre-training mixture for up to 1 × 1022 FLOPs." Seems that for the last year or so these models are getting smaller. I would be surprised if GPT-4 had > the number
7.
by techbruv 4y ago
ChatGPT and other LLMs for that matter are most definitely not using beam search or greedy sampling. Greedy sampling is prone to repetition and just in general gives pretty subpar results that make no sense. While beam search is better than