Daniel Van Zant

@theresidency Super pumped about this! Reminds me of Toyota open-sourcing their process. It ended up revolutionized the entire manufacturing industry even outside of automotive. Hoping this can do the same for residencies!

the world needs more residencies... today there are not enough spaces for ambition to thrive we’ve had 100+ people across the world reach out to us about starting a residency, so today we are open sourcing our playbook comment playbook and we’ll send it over to you

@theresidency Let's gooooo!

APPLICATIONS ARE OPEN 🪽❕for residencies in sf, nyc, london, & more cities around the world spend summer 2026 locked-in with other ambitious builders learn more below

Really enjoyed writing this article with the Wolfram Institute. Lots of research done in the past four months and more articles like this soon to come!

This is cool. If I'm understanding correctly, it's almost like there are three autoregressive loops. Their is one that might be what we could call the "conscious core" which continually takes in the current "self-state" + personal outputs + environmental input and outputs the next self-state (or maybe jsut appends some tokens on to the end of the internal ticker tape). There is the conventional "active inference loop" that continually takes in the self-state + environmental inputs and creates personal outputs. Finally, the environment itself would be the third autoregressive wheel that continually takes in personal outputs and produces environmental inputs. If I'm understanding what you're saying, as long as their is some predictability and continuity in all three of these then you can have a minimally conscious system.

This question literally kept me up last night, so here’s my current best (still speculative) answer. On my view, a minimal conscious system needs two things: 1) A recursive core whose next state depends on its own previous output. The system carries an internal state forward in time, updating it by folding in part of its own prior output (plus whatever comes from the environment). The state need not store a full history — it can be a compressed, evolving summary — but it must persist and be shaped by its own past. That’s the computational sense in which there is a continuing “self.” 2) Structured sensorimotor coupling in which the system’s outputs shape its next inputs. The environment (physical or virtual) must impose stable regularities so that the system’s actions reliably determine what it encounters next. In other words, the system literally receives the consequences of its own outputs back into its input stream, and those consequences update the recursive state. That creates a continuous trajectory through a structured space of action–consequence contingencies. In short: A minimal subject is a recursive process whose outputs reliably shape the inputs it later receives, through a world that imposes stable structure on those transformations. I initially was thinking that the system might need intrinsic goals/ some kind of teleology, but I’m not convinced. Learning environmental regularities may require optimization but not necessarily goals. However, I do wonder whether true subjectivity requires “caring” about the self. I’ve had quite a few commenters refer to Buddhist teachings the conflict with this premise. Of course, some of them have also rejected the idea that there needs to be any sort of self to have experience, but I am still scratching my head at that one.

In my last post, I argued that LLMs probably aren't conscious because our own linguistic processing is unconscious. We only feel the sensory byproducts (inner speech, visual imagery), not the linguistic computation itself. But what about multimodal models? If an AI processes or generates video and audio, does that make it a good candidate for consciousness? I don't think so. The issue isn't about the format of the data (pixels vs. text). It's about the structure of the computation. Perception isn't just a passive movie screen of "world input." It is intrinsically egocentric and action-oriented. It's inherently subjective information about what you can do in the world. Depth: "How much do I have to move to reach it?" Shape: "How do I have to configure my hand to grab it?" Sound: "Where do I I need to turn and to what?" This is what gives the subjective its subject Your nervous system is an autoregressive loop, but it's not "predicting the next likely token." It's computing: given where I am, what comes next if I do this? That unified loop—a single system continuously computing its own next states—is likely what gives rise to the sense of self. Multimodal LLMs don't change this. They process images and audio, but it's still "what comes next in the sequence?"—not "what do I do from here?" Even a robot probably won't be conscious—and it also probably won't work as well as humans either—if it's just a language model or other generative model separately outputting motor commands. What's needed is a unified autoregressive loop computing over its own sensorimotor trajectory. Consciousness isn't just "processing a sensory stream." It is the experience of being a system calculating its own potential future states.

SF just got augmented ⚡️ Augmentation Lab is now live in SF. shoutout @rhobusiness for fueling our first Founder Dinner. the next wave of human–AI tools is being built here. join us 👇 Discord

Excited about this research proposal with @WolframInst . Will be exploring decentralized science as a way to fix some issues in science funding. Some generous donations so far (especially a recent and large one from @BioProtocol). Hoping to close the gap in the next few days.

@ResearchHub @WolframInst Howdy. Main researcher here! Happy to answer any questions anyone might have on the proposal.

Basic science creates the biggest breakthroughs. Yet it’s the most underfunded type of research. Researchers from the @WolframInst seek funding to explore how DeSci DAOs can realign incentives for better funding and avoid the traps of traditional models. 🧵 1/8

And @JesParent !

Will be presenting and pitching at the @auglab Human Augmentation Summit at the MIT Media Lab in just a few days! Super excited to see some of my favorite people there. @TOEwithCurt @ekkolapto @DuganHammock @BenjaminLy61243 @DuganHammock

@dwinny_21 Thanks Edwin! We appreciate it.

I just funded: A New Kind of Science Organization on ResearchHub! researchhub.com/fund/4300/a-ne…

Have been working hard on this crowdfunded proposal to research a "New Kind of Science Organization" in collaboration with James at the Wolfram Institute. If you are interested even an upvote on the post or a positive comment can help this get funded. #DeSci #DAOs

@JesParent Here's the article: doi.org/10.1016/S0018-… that that refers to (I have the AI give everything punchier titles because it is more fun).

@Daniel_Van_Zant Allostasis in the first title looks interesting

When I find an interesting article/paper/etc I don't read it right away, I save it in a folder on my computer. Once I have enough, I run a small script to get it printed as a glossy magazine with an AI-designed cover (here's the latest) and read through everything at my leisure.

@jposhaughnessy @osventuresllc @TOEwithCurt I am seconding that emotion from @auglab. Would love to meet you in person at the event if you're able. Will have some exciting stuff I am showing off which has been funded by the OSV grant.

Congratulations to our @osventuresllc Fellow @TOEwithCurt 🔥🔥🔥

🤖 What if #AI could accelerate the way science builds its theories? At the @WolframInst, @Daniel_Van_Zant (PhD Student & Center Fellow) presented a system designed to speed theory-building in neuroscience, and maybe reshape research itself. 👉Watch now: youtube.com/live/S9qzaDKs7…

I am presenting online for the @WolframInst at 2pm ET today. Will be talking about an AI system that I am building and how it might be used to validate abstract theories with purely computational "Wolframian-style" experiments. m.youtube.com/watch?v=S9qzaD…

Great conversation with my advisor. Greatly enjoyed it.

I am working on an app to help researchers do a deep literature search on a specific concept. One thing I'm doing, and I know other folks in the space are doing, is mostly limiting things to papers that are available open access (arxiv, biorxiv, pubmedcentral, etc.) as those are much easier to scrape. I think going into the future making your papers open access will be extremely important if you want to get cited and be part of the scientific conversation. Relevant PSA: Researchgate makes their stuff almost impossible to scrape, so if you want your work to be cited by AI research tools, you need to put it up somewhere else.