2 ms·
Our bodies implement a script, which is our genome. Our genome is not a script --- more like a blueprint. Once construction is finished, the thoughts and acti
by jqpabc123 4y ago
Our bodies implement a script, which is our genome.
Our genome is not a script --- more like a blueprint.
Once construction is finished, the thoughts and actions of your mind/body are not "scripted" --- unless you *believe* in pre-determination.
- ben_w 4y agoThat also describes GPT-3. The hyperperameters are a blueprint. If you don't want to count the data flow in a human brain as "scripted", you can't count the data flow in an artificial neural network as scripted; if you do count the ANN as scripted, then you should also count brains as such. Current AI are, it is worth pointing out, less complex than a human brain: GPT-3 is appropriately the same number of parameters as a rat has synapses, while DALL•E 2 is approximately a honey bee. But the argument you're making has not yet touched scale.
- jqpabc123 4y agoThat also describes GPT-3. The hyperperameters are a blueprint. Can it diagnose my car engine? It seems to be idling rough and it's using too much gas. Does the throttle body need cleaning or maybe one of the coil packs is failing?
- ben_w 4y ago> Can it diagnose my car engine? Probably. My free credit has run out and I've not gotten around to setting up a payment method for more credit, but this kind of random question is the kind of thing it's surprisingly good at.
- jqpabc123 4y agoCan it really *diagnose* it --- or will it just put together a technically correct and properly structured reply that may sound kinda plausible to the uninformed but which lacks any real insight? Does it ever object to answering a question? What does it think about in it's spare time? Is it scared someone will turn it off? The distinction here tends to highlight the difference between consciousness and programming.
- ben_w 4y ago> will it just put together a technically correct and properly structured reply that may sound kinda plausible to the uninformed but which lacks any real insight? Most humans would fail this standard. (Hence the memes "Never read the comments!" and "Eternal September" and widespread beliefs in paranormal and so on). The (surprising to me) difference is that GPT-3 generally does a bit better than, e.g. Hollywood script writers when facing similar prompts. This surprising quality is why GPT-3 led to GitHub Copilot.
- ben_w 4y agoToo late to edit the other content, but I think this is a perfect example of humans just putting "together a technically correct and properly structured reply that may sound kinda plausible to the uninformed but which lacks any real insight" when asked to draw a bicycle from memory: https://www.mensjournal.com/adventure/people-asked-to-draw-bicycles-from-memory-fail-miserably/ https://www.mensjournal.com/adventure/people-asked-to-draw-b...