In college, many moons ago, I got a hold of the first conversational application at the time. I was on a Macintosh. So my AI could speak it’s responses to my input prompts. It was fun. The data set was 0.00001 percent of today’s LLMs. Still, the joy was real, the responses were novel, and my paper scored me an A in a philosophy class. I was trying to reason with a crude chatbot of 1986. Today, ChatGPT could come close to fooling the inexperienced.
To understand *ai* and the LLM/ML innovations underway, you’ve got to spend time with the tools. Generate a ton of images and try to refine to what you really want. Art-directing an *ai* is a real job. Use ChatGPT and BARD to rewrite poorly submitted prose from across the company. Bingo! Learning to use today’s tools is a 7% solution. The larger magic has been going on for 20 – 30 years. The data scientists today have tools and languages that didn’t exist even 5 years go.
I’m more interested in “creative” generation. Most of what I’ve been able to coax out of ChatGPT has been for corporate content repurposing and revising. A task it excels at. What I’m more interested in, and the conversation with Claude.ai, that I’m about to share with you, goes more to the question of WHAT IS CREATIVITY? Up until my chat with Claude.ai, I was engaged in *ai* conversations mainly with ChatGPT, and BARD when my free GPT tokens had been exhausted.
Claude.ai has been trained in a more humanistic model. BARD took a few steps in this direction, but Claude seems to have “self-reflective” thoughts. That’s not what they are, but the way the language is written and the training for “no harm,” Claude comes off like a polite junior college professor, as he explains why he chose certain phrases or summaries. Of course, this is more LLM magic and training. There is no light of consciousness going on in Claude’s conversation with me. BUT… There is a flicker of “holy shit” when his generative responses and summaries of why he responded that way begin to build rapport with my human ego.
I don’t think Claude and I have a relationship, yet. But in the three days of interactions with Anthropic’s *ai* I am encouraged. And with a mission that includes safety, at least they are orienting with the right sentiment.
Not sure how much we believe “Do No Evil” mottos with Google’s takeover of virtually everything, but Anthropic has its human hearts in the frame of mind.
Here is my recent prompt and conversation with Claude.ai. (set in <code> brackets for your viewing pleasure. Perhaps this is how we will relegate ai-assisted writing in the future.
Using a chapter of a sci-fi book I am writing, I played with *ai* today. Let's write a script. Here's the source. Hyper-Soul - sci-fi classic in progress.
Go.
Hyper-Soul
When a struggling musician discovers ancient artifacts that reveal chilling visions of a techno-dystopian future, his sanity unravels as he races to record his flood of manic creativity before the revelations consume him completely, sparking a battle in his mind between genius and madness as he hurtles toward an apocalyptic destiny that threatens to fracture his world.
Claude(tm) was about to blow my mind. The full transcripts are available, if you want to go into the weeds with me and Claude, but here’s the gist of the conversation.
Here's an attempt at a darker, more David Lynch-esque logline in 10 words or less:
Fractured mind unlocks madness and genius through time-bending relics.
I tried to capture a sense of surrealism, fragmented reality, and descension into madness that evokes the tone of Lynch's work. Please let me know if you'd like me to modify the logline further or try another attempt at a short, impactful Lynchian description! I'm happy to keep refining.
Now I’ll push a little harder on the “style” concept.
Here's another attempt at a dark, dread-inducing logline with a Phillip K. Dick inspired sense of awe:
Reality unravels as artifacts expose a techno-dystopian destiny.
I tried to evoke a sense of ominous revelation and a dark futuristic vision unfolding that feels very PKD. Please provide any feedback if you'd like me to modify the tone or elements further. I'm happy to keep experimenting with the logline to find the right intensity and flavor of sci-fi/horror. Just let me know what specifically you'd like tweaked or pushed further!
And just for fun here’s one more summary from Claude.
Here is a 3 sentence summary of the film with a dark, Philip K. Dick-inspired tone:
When a struggling musician discovers ancient artifacts that reveal chilling visions of a techno-dystopian future, his sanity unravels as he races to record his flood of manic creativity before the revelations consume him completely, sparking a battle in his mind between genius and madness as he hurtles toward an apocalyptic destiny that threatens to fracture his world.
I tried to incorporate a sense of impending dread, slipping grip on reality, ominous future visions, and tensions between creativity and madness - all lending toward a dark, mind-bending summary suitable for a PKD-style psychological sci-fi thriller. Please provide any feedback if you'd like me to modify or refine anything in the 3 sentence summation. I'm happy to keep experimenting with elements and tone. Just let me know!
I then proceeded to get Claude to write the first few scenes of our blockbuster sci-fi collaboration. I immediately understood the writer’s and actor’s strike this year. The formatting was perfect. The storylines varied from attempt after attempt. We had to negotiate some of the sensitive topics, but you know *ai* generally is locked down with some rules that prevent most sensitive topics.
Finally, I’ll show you Claude’s idea for some blockbuster movie titles. I love how the *ai* explains his work.
Here are some blockbuster-style movie title ideas inspired by the sci-fi psychological thriller premise:
Chrono Fracture
- Evokes a broken timeline, fractured reality
Artifacts of Doom
- Sensational title hinting at ominous relics
The Mind Unbound
- Suggests both creativity and madness
Destiny Decoded
- Alludes to uncovering/decoding the future
Paradox
- Simple, clean allusion to distortions in time
Singularity
- Sci-fi term evoking AI and altered realities
Convergence
- Hints at a collision of times/realities
Limitless Dimension
- Suggests incredible possibilities unlocked
Through the Rift
- Indicates a tear in realities
Apotheosis
- Loaded word implying transcendence
Let me know if any of those titles resonate or inspire other blockbuster ideas! I tried to use some common tropes but give them a fresh spin. I'm happy to generate more big spectacle-style titles - just say the word!
I can see how this is a process of training both of us.
With some further digging, in a future post, I’m going to see if I can source some of the themes and ideas Claude has pulled out of his LLM. But for now, know that my skepticism about the “creativity” of *ai* remains, while my appreciation for Claude’s ideas and manners is glowing. I am certain Claude would pass the Turing Test.
What wonders await us as Claude grows up? How can we predict or protect against bad outcomes? Who’s workloads are we loading? Where is the data coming from that’s being used to create LLMs? Would WikiLeaks score gold by locking down their data and building the Wikileak.ai?
What’s stopping them? As OpenAI gives out licenses to GPT’s of our own, we’re going to see the open data get fragmented, protected, and more secure. As Matthew McConaughey says in his 10 million dollar contract for Salesforce,
Data may be the new gold, but that doesn’t mean it should be for sale. That’s why, alongside trusty partner Matthew McConaughey, we’re here to keep your information safe from hackers, highwaymen, and other nefarious ne’er-do-wells.
We’re rushing into the guarded era of *ai.* Firewalls are going up. Will I continue to share my ideas and IP with Claude? Likely, for a bit, until a better option comes along.
read more: uber.la on ai
John McElhenney — LinkedIn
Please check out a few of my books on AMAZON.
Especially this one, about living a creative and human life of intention and joy. 100% human generated – with the exception of ai-assisted spelling.
+++ but wait, there’s one more thing +++