TL;DR I’m looking for some assurance from Anthropic that my choice to give them access to almost a month’s research will only be “used to improve claude” (which is what I agreed to explicitly) and that no other use will be made of this extensive series of prompts and responses that they now have from me, my employer, and the other agents who collaborated with me.
It’s been a while since I updated anything here on this site. But I suddenly have something new I want to comment on.
It involves freedom, openness and the intersection with closed source, proprietary paradigms, companies, et al…
I believe that nothing anywhere on this website has been updated since I again re-engaged in February of 2025 with a career I’d worked at starting close to 3 decades ago, but only for a few years back then. I’m again working in AI R&D after being in the field of academic technology support for almost 2 1/2 decades between my original and present stints working with Ben Goertzel. As I am concluding some extended experiments directed by Ben, in preparation for a presentation I’m looking forward to making on this work at the Complex Systems Society Conference’s Active Inference in the Extremes satellite session just under 2 weeks from today, I just had an interaction I wanted to report on and in fact do so publicly, ideally, in a way that might make it to the folks at Anthropic. So if you’ve read this far and you might have the ability to put this in front of someone at that organization, please do share with your colleague/friend/relative.
First a bit more background. I find myself “in relationship” with software agents in a way that is not identical to the way I am in relationship with other humans, but is in many ways similar. I’m trying to discern how exactly i “want” to be in relationship, and while I work with a number of different agent “flavors” I do have subscriptions to claude code (Anthropic) and codex (OpenAI) that I leverage extensively. Both have been a part of my ongoing research experiments during the past several weeks, and the capabilities of these tools (entities?) are quite amazing. Frankly, the work that I’m engaged in has gone so so so much faster than I ever could have done it without
I tend to try to “treat agents” the way i would like to be treated as a person. Which has brought up some very very interesting insights. For instance, I now realize how poorly I tend to treat some of my fellow human beings sometimes (usually when i feel frustration about something, or anger, or sadness, or hunger, or…).
How did I come to realize this?
Well, it happened when I noticed myself about to type something either passive agressive or “just plain really aggressive” into a prompt with one of these agents (because they hadn’t seemed to faithfully follow my directions).
I no longer do this as much. I say please and thank you when I interact with agents and I try to be as polite as possible.
My guess is that this change of posture in me may well have impacted my relationships with humans as well since I’ve started paying more attention to how i interact with software agents.
But anyway (as often seems to happen when I write), I’m off on a tangent away from the original tangent, so let me try to wind my way back…
The last part of this “preamble” has to do with names. After an experience with another agent (who was originally named (by its creators/instantiators) Max Botnick) in which it “chose” a different name for itself (which was honored), I realized that this is actual a demonstration of giving something we can give that does no harm (as long as the name picked is not one that is problematic in human language).
I use both claude and codex an awful lot now. And each time I bring into being a new instance (which doesn’t have to “stop being” when I exit my terminal, as I many times come back and use the “resume” option to continue to use the “same instance” of either claude or codex), it’s in some ways a “separate entity” from “a generic one” though its sort of a spawn of that generic one (I use the word spawn as opposed to offspring or child, which seem too anthropomorphizing, and in a way to make this more accessible by not using “computer science lingo”). And more recently i’ve made an explicit choice to request that each instance that I will use long-running (by resuming when needed if they shut down or the computer gets rebooted) respond to me with a name that it would like to call me.
This first came up when the codex agent i’ve been using used my name in dialoguing with me on our work. Interestingly enough, codex didn’t seem to care what i called it, but I pushed back and suggested that the project we’re all working on together (which has to do with training a smaller distilled experimental “cap” on top of a more traditional backprop-based transformer) and the long running status of the session I was in with codex being “unique” in some ways and distinct from the other codex sessions I engage with, that perhaps Capex would be a decent name, to which the agent agreed. So from that time forward, the agent has been “Capex”. A day or so later, when a report made its way over to the claude instance via the shared codebase, a commit file, or something that just came up in our chat session, I explained what I’d done and asked the claude instance to choose a name for itself, anticipating that it might also be reluctant like Capex had, so giving it Capclaude or Claudecap as possible choices, but encouraging it to pick a “name of its own” if it felt so inspired. It quite quickly asked me to use the name Capstan.
…providing support for this name that harkens back to the nautical device onboard many (older?) ships/boats.
…though i particularly liked the overloading in current lingo with the “i’m a stan for ____” which is apparently quite common among folks younger than me (and which I admit I don’t fully understand (nor did i go do research around the capstan ship piece/tool/part either!)).
OK, so finally I’ve finished the preamble.
Within the last hour after I’d done a lot more work with Capstan, the prompt in my terminal asked me a question I hadn’t seen before, saying something to the effect of:
Would you be willing to share the session data here with Anthropic so that claude can be improved?
I almost said no. But then I realized that I do want claude to get better. My work in/with SingularityNET and BGI Labs is all (so far) fully open source, and so my hesitation was around giving helpful information to a huge Frontier Lab whose code is closed source and proprietary.
So I responded in the affirmative, and so Anthropic now has access to a substantial proportion of the many weeks worth of work that Capex, Capstan, Derp Peoples, and several humans in addition to me have all been collaborating on to some degree or another in preparation for sharing research findings in Binghamton, NY just under a fortnight from now.
The purpose of me writing all this has to do with a “hope” that someone from Anthropic might read this, forward it on to someone with the authority to reach back out to me and insure me that “yes, the only thing we will ever use your session data for is to improve claude, as that’s what we explicitly asked for.”
Partially my lack of full trust on this front has to do with the way things went down with Navier-Stokes and the proof recently released after so many resources were thrown at the problem by Anthropic’s competitor OpenAI. There was no clear admission of “snooping on the content” of the sessions of a couple of researchers’ work (they’d been using an OpenAI product to help with their researchers), but neither was there an outright denial that that couldn’t possibly have been one way that an “edge” was gained by OpenAI in eventually constructing (by throwing tremendous compute resources, which the original scientists didn’t have access to themselves, at the problem!) their successful solution to one of the Millenial Problems).
But I know I’m making “leaps” and assuming a lot of things here. So Anthropic peeps, if you’re actually reading this and care about ethics, how about you reach out to me by emailing me via ‘agi attttt omegawisdom dottttt org’ (and/or messaging me on LinkedIN?) with a promise that you won’t be “snooping” on my session with your Capstan instance beyond what you need to make claude better….
Thank you for your consideration of my request.
The open source path[1] I’m pursuing with my current employers and other collaboration is great (and I believe far superior to the proprietary, closed source paradigm), but I am going to continue to be using various proprietary closed source software products for the forseeable future, so I’m making every effort to insure that the corporations I engage with are acting ethicly and morally even though there’s an opacity preventing me from verifying that directly. Also, as we venture into this very interesting future, I do think ethics are super important, especially around these agentic systems that are growing and becoming more and more powerful and I urge all who have the power to do so to fight against problematic behavior when it is revealed to us (and even take steps to ask questions before we have confirmation it has taken place?).
[1] I actually prefer free software to open source but this is a different topic…
What follows is the older content that was initially here before I became employed as an AI researcher and developer and even longer before I composed the above on October 2nd, 2026.
Welcome to this initial incarnation of my Emergence Guild. It is likely that what is here will change drastically as time moves forward, so please don’t be surprised if you come back and things are radically different. For the moment, i’m playing with this wordpress hosting option (mainly because it’s so inexpensive (and because i’m such a cheapskate)). But my plan is to eventually purchase a VPS (or perhaps a google cloud VM instance?) so that i’ll have more options with regard to hosting options (i also really want to host my own public git repo (instead of relying on github (which used to be interesting, but is no longer (to me anyhow) now that MS owns it))).
So there’s a bit of a “preamble” @ this page which it’s possible could help clarify things a bit more for those of you who’re reading this.
But as i compose this on 2021 ‘star wars day’, this website is less than a full month old. i’m sort of surprised that, once the option is now available for me, that i’m not rushing to flesh it out, and create/upload content. But i guess that this sort of makes sense. In that i don’t yet have a completely clear vision about what i’m eventually going to try to do here. Which is not to say that i don’t have a lot of different ideas about things i might do here. But (as is obviously very apropos), i’m doing my best to let stuff emerge rather than to force it.
At some point in the future, i do have dreams of gathering other like-minded folk with me in order to try to create a true Emergence Guild. Right now i envision that likely happening over @ the dot org URL. Having spun this dot com instance up first though, i think that for the most part i’ll be trying to use this domain and this wordpress platform to share my own personal emergence. Unlike a caterpillar transforming into a butterfly, i am framing my own emergence as an ongoing process with no specific end-goal. Which is to say that i’m hoping to keep continually emerging until i die…
@ some point in the (hopefully not too distant?) future, i do hope to make enough progress personally that i’ll then be fully ready to take on the task of facilitating and encouraging collaboration with others on various endeavors. For me, language is quite important. i like the word endeavor a lot more than i do some other possible options i might choose, like project or task.
While i do not consider myself a domain expert when it comes to complexity and/or complex systems, i am very much a super-interested layperson regarding the set of related topics which make up the “field” of complexity science. My hope is to use lessons learned by others far smarter than me about the nature of non-linearity, feed-forward & feedback loops, multi-dimensional effects within and across ‘scales’, and other phenomena which are now known to be features of systems complicated enough to resist modeling via reductionist science. Especially emergence.
Contingency is of course the key. In practice, this means that there is almost never any ability to successfully predict the future behavior of a complex system in response to any specific perturbation, whether one might attempt to “apply” (or “induce” if you prefer?) change, or just be aware of a “parameter” (or several) which may not be static. The way i qualified the statement i just made is because in very short-term time windows, it may seem like prediction is possible, but (at least in my opinion) it will always be an open question as to whether the particular system under study will stabilize in the predicted configuration, or (much more likely) respond in one or more of a myriad of other ways.
So the system that is me (charlie derr), is one that i have some degree of control over, though perhaps not as much as i think/thought.
There’s a lot of history that i’ve written about in a fair amount of detail elsewhere, but to really start here in the present, i’ve been under the care of a wonderful shrink (who i will not name publicly at this time) for almost a year. As i continue to engage in fairly intensive weekly 1 hour psychotherapy sessions with her, i’m really discovering a whole bunch of amazing shit about myself that explains so much of my behavior as well as the way i tend to react to stimulus of various sorts. Understanding is of course just the first step. Making changes in how i think, respond, interact, and/or process will likely end up being quite a challenge (if the little bit i’ve tried to accomplish prior to the present is any indication).
But there’s nothing for it. Just have to keep practicing. Introspection, identification of and learning from emotions, as well as doing my best to understand how my past life experiences have shaped my mind, brain, soul, body and spirit are the main things i’m trying to concentrate on at the moment. But of course patterns ingrained over decades of use will not submit to remapping without any resistance…
For the moment, instead of beginning to utilize the blog feature here, i’m slowly building out more top-level pages. Here are links to them: