Using narrow AI to train narrow AI

Post Reply
User avatar
funkervogt
Posts: 1165
Joined: Mon May 17, 2021 3:03 pm

Using narrow AI to train narrow AI

Post by funkervogt »

The jury seems to still be out on whether more detailed prompts that can be generated with the help of an AI chatbot such as Chat-GPT really make a difference to the rendered results from MidJourney.

I am not decided either way but in the examples below it seems that the additional detail and augmentation that Chat-GPT provided helped MidJourney produce better and more consistent outcomes.

“Scone with a layer of jam and cream, A warm, freshly baked scone with a flaky exterior and a soft, fluffy interior, adorned with a generous layer of sweet, sticky jam and a dollop of whipped, creamy cream, surrounded by a delicate porcelain teacup filled with steaming hot tea, The cozy, intimate setting of an old-fashioned tearoom, with lace tablecloths, polished silverware, and warm, soft lighting filtering in through the large, paned windows, A content, peaceful feeling, with the aroma of tea and freshly baked pastries lingering in the air, Soft, warm, and inviting, with golden light casting a gentle glow over the scene, illuminating the delicate tea-things and creating dappled shadows on the table. --c 15 --s 150 --q 2 --ar 3:2”
http://mostly.photos/blog/2023/2/14/cha ... gmentation

I recommend clicking on the link so you can see the pictures yourself.

My opinion is that ChatGPT and Midjourney are narrow AIs that, even if scaled up, won't result in the creation of AGI. We need a major breakthrough or series of smaller breakthroughs in intelligent algorithms to make AGI.

However, that doesn't mean our narrow AIs can't become enormously useful and even superhuman in capabilities given more compute and training feedback between them. This paradigm has a long way to go before it tops out, and it will probably top out at human expert levels or above. We are only at the beginning.
Post Reply