OpenAI’s API feature GPT3 is a wonderfully expressive technology. The API is a text in/text out API converts the input text by way of a extremely large language “transformer network” into the output text. The API is so expressive that it is easy to get lost in conversation/interaction with it and miss forensic details about its behavior. The more sophisticated your desired use the more the nuances of GPT3 become important to spot and understand.
As CTO and the person who has done math/stats and machine learning the longest at Maslo the most common question I get is, “why doesn’t Maslo seem to do AI?” and the variant, “When will Maslo obviously get smarter?”
My answer is simple: life emerges as a phenomena of the interaction between complex systems. More broadly: when employees, investors, partners and the science become committed to broad implications, consequences, limitations, and investment.
Yes, I have a “model zero” approach. I assume the null hypothesis. I assume technology fails. I assume no single method or tool is great. …
This is the second part of a multi-part essay series on GPT3 and the OpenAI API. In the first part we covered basic structural concepts of language that are useful to understanding computational language models.
Just as the first essay was meant as a behavioral and forensics experience so to is this part 2 devoid of deep mathematical theory and technical architecture detail. Let us proceed to understand the higher level concepts involved in the semantic and meaning nature of GPT3.
As in part 1 any grayed in blocks or screenshots will indicate a human provided input by BOLD. We will move between screenshots of the OpenAI Playground to provide readers a view of how the fuller environment frames things and sometimes use gray highlighted plain text to make it easy for readers to cut and paste. …
Here’s the unvarnished story of how I met one of my closest collaborators.
I will link you my article that I explained briefly what I will focus on the thesis.
The questions are really meaningful and also challenging. I can say that maybe I can try to build a rich emphatic relationship between intelligent and evolving machine and the users. Since design is still at the beginning of this exploration, I can try to focus and find a meaningful interactions as a first step. So, I can start to work on this question “how do you make device surfaces dynamic enough to be biological?” …
This summer I have spent many hours a day conversing with computer networks. Yes. NLP models, various “AIs”, chatbots, generative systems and more. Sometimes they are purely written language experiences other times multi-modal. I have tried not to “cheat” and program / perturb the interaction but instead simply communicate within the frame, find shared meaning and shape gestural signaling.
My hypothesis regarding computer intelligence (complex computational interaction) is that mechanical programming and “force” is never going to work. There is no deterministic nor ethical laws for human-computer interaction — there is simply empathy and flow. We must become computationally fluent.
Today I spent an hour with a computer network (with OpenAI’s API) exploring how slight tweaks to language signs can change the behavioral flow of the conversation. This is an hour long co-learning experience. …
There continues to be a growing confusion about possibilities for artificial intelligence and computing. Our vocabulary fails and our research paths shaped by business, mythology, fables, history and academic traditions are discordant with our lived experience and wider reality.
There’s no easy fix/path to get AI research anywhere in particular — no one really knows what they mean when they say AI. …
my original art form is the theater and remains the world as stage.
to become the other. to inhabit worldviews. to collaborate with writers, directors, set designers, stage managers, other performers, the audience, the staged reality. to become my fullest self in the abandonment to others. to see my self, my art flowing through and between others.
in the last year and half through a strange brew of events I have found one of the most productive collaborations of my life. my dearest friend, greg, generously engages my ideas and his own. takes them seriously. stages them with love. …
Here’s a remarkable little thing bubbling up…
Various sources of human knowledge in a kerfuffle about who is allowed to charge what for access to human knowledge and when.
Of course, it’s not necessarily wrong, morally or ethically or business-wise, to defend the turf on some of this. Then again, I’m not sure anyone knows what turf they are defending.
If you read between the lines no one is really doing more than selling or unselling access.
The internet has this funny way of cratering any sort of long term gatekeeping. in fact, more than the internet, printing and mass media itself (printing press on) tends to make information free/free information. …
Tiana Cornelius provided this wonderful summary of our efforts to crack the hard NLP problem of text summary. https://medium.com/maslo/using-nlp-to-summarize-human-thoughts-feelings-b64079030104.
So now, can we better summarize the summarizer!?!?
Keeping a log of what I did and what I’m doing. The why will come later, I hope.
During uncertainty, or obvious uncertainty, I just work. I work non stop. As if work/action/activity can magically whisk away the chaos of the universe. Makes me think of all the times of trauma in life or periods of transition, and, hilariously, when I was doing lots of math in my formal education. Something about staring down my own ignorance that makes me want to fill chalkboards of beautiful mind theories to sort it all out.
Knowing this about myself I just embrace it. And in my middle age I get to call it all “art” or “being Russ.” …