The Future of AI - Michael Wu - UC Berkeley Lecture

The video starts out shakey, but it gets better. Sorry. I had to shoot hand held.
Why does ChatGPT appear to type it’s response? For the same reason you can’t trust its output. Here’s some tidbits you can toss out at your next mocktail party, so you can sound super smart.
Yesterday I caught a killer presentation on the Future of AI given by Michael Wu, Chief AI Strategist. Here are his most important talking points.
First point, AI responses use up 10x more energy than a regular Google search. Err…that’s an elephant in the room.
Second, know that amazing MidJourney image you created? The courts have ruled that generative AI images are not copyrightable.
The prompt engineer didn’t create the work. The AI did. So anybody can steal it.
What happens if you modify your image? That is up for debate and will be a hot topic moving forward.
Third, and more important, all generative AI is what the industry calls a “hallucination”. This is huge.
It is literally guessing the next word, and making up its response as it goes. How?
Think auto completion on steroids.
The language model looks at the current word and tries to guess what the next one will be based on what it knows by reading the whole Internet.
This is also why it’s response isn’t instant, and instead slowly trickles in, as if it were being typed out. Because it is.
This is both good and bad. The good news is it can make stuff up that never existed before.
The bad news is its making it up, so you can’t trust it to be 100% accuracy.
What’s the fix? Here’s one that Bard is using:
Use data you can trust, like Google search results.
Then have the AI format that information output.
Takeway: Your new mantra. Never trust an AI. Always edit its output and fact check its results.
All magic has a price.