Softmax

Is a monthly cadence right for this? So far, the experiment seems successful. But we are at the very dawn of organizational metadesign. Maybe it should be 4 days and Cooldown Fridays. Or maybe there should be two cooling months per year. We run Softmax as a living experiment.

During Annealing Week, we aren’t trying to make progress against our goals. Instead, we care about simplifying things. Removing steps. Killing processes. Deleting code. Replacing two features with one. Cutting meetings. Pruning the list of channels. Reducing company complexity.

It’s Annealing Week at Softmax! Humans are awake for 16 hours learning, cooling for 4 hours in light sleep, and in deep sleep for 4. An organic mental annealing cycle, heating to cooling. At Softmax, we do the same. It’s four weeks sprinting towards goals, one week consolidating.

Our little Cogs grow up so fast. Cogbert has never seen this exact production chain before, but with only a couple missteps he begins to execute it correctly. Our in-context learner takes its first baby steps!

1) living wholes are made of living parts unified by shared goals (purpose) 2) the more possible actions and the higher frequency of choosing actions, the more complex the system 3) therefore parts must take on roles from a limited list and change them at a limited frequency 4) if you knew the correct set of roles and you knew the rules to infer which person should be in which role, you’d be done 5) we have a background prior for the roles for a successful corporation based on the evolutionary truth of which corporations survive 6) in order to move faster than trial-and-error it is the job of the hierarchy to make guesses about what variances from the background prior are required, and what local signals must be integrated to choose a role 7) it is the job of employees (parts) to “differentiate”, and select their role and perform in it, based on the rules being propagated out

@softmaxresearch which principles please?

We are building organic alignment at Softmax. Not just with reinforcement learning, but within our company we try to use these same principles for our work. We are implementing this as an organizational operations system (OrgOS), a prompt library covering our internal processes.

If you’ve written interactive prompts that help guide the user through making a plan or giving feedback or documenting their thought process, what have you learned doing it? What are the very best active process prompts you’ve made or used, and what made them great?

InstaDeep, Africa’s foremost AI frontier lab, is doing some of the most compelling MARL work in the world. Sable is a genuine breakthrough. Worth checking out.

Fechner’s Elements of Psychophysics is the latest addition to the Softmax library

@kylebrussell Syncophancy is not loving-kindness.

“How is the loving-kindness model at sycophancy?”

Coming soon: BE NOT AFRAID

Our CEO, Emmett Shear, appeared on BuzzRobot and shared a bit more about our vision of the future x.com/sopharicks/sta…

Wonderful to be invited to the @softmaxresearch research community day yesterday - my lightning talk and unconference session were about artificial minds and the difficulties in getting 'complex' consciousness out of a stepwise algorithm...

OH at the office: “What’s GitHub? Oh, it’s like Facebook for nerds”

Our CEO, Emmett Shear, gave a talk on alignment protocols: the engineered ways that parts communicate in order to align their trajectories. youtube.com/watch?v=yBc7Ix…

tired: inductive bias wired: forgetting bias inspired: coherence bias

Wishing all learning agents a happy Easter

@onabenchinapark For more depth on the questions you're asking, I'd recommend Sex Ecology Spirituality by Wilber

Banger. Love the “frame-flexibility” concept! The way the form of this insight was expressed helped click a deeper understanding of Kegan 5 :) now that’s some original transmission Some things that came to mind that I wanted to share if they resonate: - If one realizes frame-dependence, how do they then incorporate this view not into their worldview, but their literal view of the world? How would the introduction of frame-dependence work for the already frame-dependent mind? There was another comment about integrating the views of emptiness etc into the lived experience of liberation, which I think would make sense as what’s to follow! To go further on Emmett’s point about dharma, I think this would have to include that. Where in which the insight is made accessible through explanation and then real by the specific sequence of words directly changing the reader’s perception! - It seems to me that for collective agent alignment, maybe some agents would require this “liberated” mind considering the usual real distribution of sentient things in our world across spacetime? But how would those exist? What is the boundary for sentience and non-sentience? And does this liberation only exist from within the non-liberated state? If so, what would that then imply about individual agent evolution concurrently with the whole? Relatedly, what does that then say about the current state of things and the distribution of “developed” minds? I think as we develop increasingly sophisticated systems while society and living also become more sophisticated, understanding how agents with different frames can align without requiring identical worldviews may become more useful. Because frames would have to be in “non-flux” at some point (I think in the sense of stabilization), is alignment then literal alignment of frames so that the flux thereafter is more resonant? If minds (both natural and artificial) are frame-dependent, does alignment then become a question of creating agents that can recognize and maintain core values across them? What would these values be? This meta-stability across frames might be what distinguishes wisdom from just intelligence. - And following up on that second question, what is then alignment from this emergent, frame-flexible view? How is alignment to be defined when things are always in flux and cohering/decohering across different scales of space and time? Is there a different space and time across sentience and non-sentience? Thanks for the writing!

Frame-dependency: it's not just a good idea, it's the law! Special thanks to Sonnet 3.7 as significant co-author on this work.

@MelonUsks I think we aim for Nobility more than Heroism but yes, sort of!

@softmaxresearch I see techno-heroism vibes in your post!) x.com/MelonUsks/stat…

The Softmax Arcana of Choice and Perception Element: Air Numerology: 6 Astrological Parallel: Gemini (Duality, contextual truth) Associated Deck Archetype: The Lovers (many thanks to ChatGPT, our Tarot guide)