RE: Embracing Collaboration With AI...
You are viewing a single comment's thread:
I don't know, I just am not that impressed so far. It just kind of feels like it is gathering information that already exists and presenting it in a different way. I think it will get better, but I want to see some real world modeling. I've said for a while now I will be impressed when AI can create an original joke that is actually funny. I know it's not very useful, but it would show that it has evolved to that point. Another good example is crypto prices. It's not like AI is actually predicting crypto prices right now, it is just searching the Internet and finding what other people predicted for specific tokens and then aggregating that information. I think that should be one of the factors AI uses to extrapolate that info in the future, but not the sole source.
Bear in mind, there are still limitations of what it can and can't do - though also a vast spectrum of what opens up with the quality, specificity, and architecture of the prompts.
Some things, it's not totally an optimal fit for doing (yet); others, it may not do upon the first command - but can get alot closer as working with it as a collaborator to build the path there and architecture required. And sometimes, that architecture is not just the external technical stuff, but your inner architecture - the way you're looking at it, thinking about it, approaching it, and bridging your ideas and feelings into well-articulated expression that expands conversations & explorations.
I, too, went through the "underimpressed" at first. But some time later as realizing *the quality of replies is often proportional to the quality of input - that using it not just for data retrieval, but an exercise in self-development of thinking skills to refine the articulation of concepts & questions which would bring back increasingly more impressive responses... indescribable leaps into new territory. And it still my not be 100% amazing at everything all the time; but continuing to test it's limits & challenge it while also your own, you'll learn that what you think are its limits often reflect your own - and can be expanded in parallel. (Even if/when that entails expansion beyond the AI - i.e. discovering that to get what you my be after in regards to crypto prices requires different Crypto API tools, a custom-built system using spreadsheets or Obsidian with GPT plugins to organize & process all that data in specific intended ways, and maybe even some trading platforms with custom scripting capabilities -> ChatGPT not able to do everything all in one package, but help guide you through seeing the overall system architecture, how to setup each of the component, coding your indicators & bots, etc, etc. **But if you simply conclude its underwhelming and can't do what you hoped rather than persisting to asking better questions, you'd have never gotten to that point. 😉)
Most definitely, I've encountered ChatGPT slipping humor into its replies; and have heard Grok can be kinda sarcastic at times. But there's a BIG difference between a single prompt of "tell me a joke" and going through a process of building a foundation whether via custom GPT or in a single thread of first establishing some perimeters it can work from - i.e. "give me a thorough detailed breakdown of the key principles of standup humor; list the various common types of humor and jokes; etc -> preloading that into your context window for a starting basis, then asking it: "tell me a super funny joke about xyz..."
(And I pose this as a challenge for you to try, so you'll experience what a drastic difference it is for yourself.)
Some stuff, if you're just starting with an empty chat thread - remember, it's just a generalist with access to breadth of data, yet can only access so much in a single inquiry... but by specifying exactly what role it is to take, you narrow the window to refine the depth it goes to in a clearly-defined field... and it's a whole different fucking ballgame.
Another to try - go test something along the lines of: "imagine Kevin Hart and Ali Wong had a baby, raised it on George Carlin and Adam Sandler jokes for all its life - nurtured it's comedic potential until 18, when it went out for its first stand up show and told the most hilarious joke of all time about a guy who was skeptical about AI and thought ChatGPT couldn't tell jokes." And when it asks if you want more a couple times, oblige.
*lemme know if/how those experiments sway your outlook at all..." 😼
Sounds like I will have to set quite a bit of time aside to get to where I need to be. I've thought about looking into notebook LM so I can load my trusted sources and I have been impressed with the image generation side of things based on prompts.
funny, on the joke thing...
Looking at this page on on some technical stuff re: ChatGPT and it uses the reference of knock-knock jokes. Lol.
more of a developer-oriented approach to what I was saying, but adds well to it - generic ChatGPT the moment you open it as a blank canvas might have far too broad of a reference field to pull out top-tier jokes consistently, but it'd totally be possible to build a custom GPT on/with it for specialization in any specific type of joke with the right input/prompting...
Interesting! I haven't been able to carve out any time to look into this more yet, but you have definitely raised my interested and changed my mind a bit about what I previously assumed.