Christina Fedor (@christinasatory) header
Christina Fedor (@christinasatory) profile picture

library maximalist | founder @consorvia | sometimes writing @future_remember | sharing nonlinear thoughts on knowledge, culture, + innovation

Los Angeles, CA
Joined November 2011
Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)Photo by Christina Fedor (@christinasatory)
2026 is a year for staying: with complexity, with consequence, and with the invisible, unknowable paths. Here are 26 more insights from this growing garden keeping me company on the way through:
For 2025, I thought I'd add 25 new insights to this greenhouse: 1. A beginning is a transformation made possible 2. the underlying premise of asking about the origins of life is that there is something there that needs an explanation
1
4
1,202
watching political arguments collapse into economic arguments is like watching a fish flapping around on the deck of the boat. one does not reason with the fish to let him know that indeed the ocean is different from the sky and the earth.
This is stunningly insightful from @chamath I'm shook for two reasons 1. The clarity of this articulation 2. It took this long to see the fucking obvious
1
78
It is interesting to watch different cognitive types at work on software concepts that do not travel well into non-software contexts, or perhaps it is the other way around. The clearest case is our use of the word agent. It comes from a verb which originally took on the sense of setting something into motion. Over time, as language tends to do, it began to specify itself and became a setting-in-motion on behalf of another. That notion, of carrying a principal's intent into the world, is useful enough, so it is no wonder our cognitive technologies have adopted it, along with the vocabulary that comes with it. Words travel far more easily than meaning. So we have artificial systems that can be usefully told apart as differences in kind within those systems. That is what words are for. They cut through experience and help us separate things out so that we can make sense of them. But there is a nagging tendency, especially among object-oriented philosophers, to hand the utility of a word's function to similar-enough events in other contexts, without asking whether meaning can and does transmute when it crosses over. The result is a great many theories to explain something like consciousness, which is oddly taken as an object in itself, even by the theories that set out to dissolve it into properties, functions and structural-functional systems. This unquestioned transmutation of agency, in software and in living minds alike, makes for a good deal of confusion now. But the confusion is not necessary. Kinds are local assignments within a semantics. They hold the constraints of the context they emerge from, and blur easily outside those contexts. All boundaries, including the boundary of a kind, hold a difference that is being paid for, moment by moment, at all times. When things blur, we know what to do. We change the lens until we can see clearly again. Which brings us back to the word. When a word seems to travel too easily, it is worth asking which boundary it came from, and who was paying to hold it. Agent crosses into software carrying its grammar and leaving its upkeep behind. What a person spends to hold an intention over time, to mean something and be answerable for it, is not spent by a program. That is the difference the word no longer carries, and at least one reason confusion keeps returning. Before we set out after new and expansive untestable theories of mind in artificial systems, the prudent course would seem to be to give deference to natural design, and to ask whether, at the deepest structural-functional arrangements, the answers we want so badly about the soul, the mind, consciousness, are quite accessible, so long as we look closely enough, and then keep looking, again and again.
1
140
reality itself is now arising before you
🤖 Made with AI
50
Christina Fedor retweeted
🤖 Made with AI
9
6
3
145
9,083
exquisite
Perception is about what comes to presence for us through appearance. Phenomenology, in turn, brings to presence that act—intentionality—which makes things present to us in the first place. What I wanted to do in this piece was to make this process explicit. I am interested here less in the initial sense in which intentionality operates and more in what kind of space we are in when we attend to the process or structure of intentionality as such. I didn’t use this language in the piece, but I want to suggest that this space is well characterized as a theoric or contemplative space (that is, a receptive space), one in which the intentional process still issues forth but where, to speak spatially, our attention is guided upstream in the attempt to witness what it is intentionality is doing as it occurs (much as we find in Husserl’s account of epochē), so that we might understand how to act on—and, ultimately, shape—the resolution at which intentionality occurs. I believe such a practice can disclose to us—again, in a contemplative and receptive way—the manner in which the “transcendental” apparatus is in some sense shapeable through practice. There is much else I wanted to say along these lines but didn’t have time to sort out before presentation time. Husserl on sedimentation. Mary Carruthers on the craft of thought and inventio. Bergson on memory. Sense settles like silt through habit. Practice both loosens and lays down new sediment. Memory is made to transform us; it lays down new pathways for future perceiving. More to come.
1
1
114
same could have been said for filmmaking, but look at the influencer economy. not exactly a collective of filmmakers, but certainly a significant group of people expressing visually driven narrative intent with tools that make it possible to
a few years ago, I started out skeptical that many non-engineers really wanted to make/mod their own personal software -- but I've come around, think I was wrong. range of desires from: - I feel so powerful - no one will make me the workflow/personalization I need - can I just PM the product for a few weeks and fix this thing that is driving me crazy obvious in retrospect, but if the ease:power ratio is good enough, of course people will make their own software
52
Large corporations are not the same as mature corporations, though. The kind of company with more written context than it knows what to do with may be "successful" despite itself, or largely because of the outsized efforts of a few who carry the burdens of the many. These political arrangements are a strategic vulnerability that startups should, in theory, be able to exploit. But in the current climate, access to cheap capital, as we have seen again and again, more often outweighs knowing what to do with what you know. AI shifts that. As building well gets cheaper, capital buys less of an edge, and only then does the market reward maturity over size. xitter.cf/karrisaarinen/status/2…
Who has lot of written context? Large corporations. Who doesn't have lot of written context? New startups. Which ones generally build better?
41
become the author of a world
1
52
Christina Fedor retweeted
Since childhood, I've seen the world as wholes. Systems rather than products. Constellations rather than planets. Soil rather than single trees. My education followed that instinct across a divide, moving from the medical sciences to architecture, between growing and making. Somewhere in between, I came to believe that when everything we make can be grown, design becomes ecology. I named that belief Material Ecology, and for a decade at the MIT Media Lab, my team, the Mediated Matter Group, set out to build it. We almost never designed something without first inventing the technology to make it. We spun pavilions with silkworms, printed glass at furnace heat, and made architectural skins from shells and apple peels, designed to return to the soil at the end of their use. Then, around 2020, the things humans make began to outweigh all living matter. Design was no longer a matter of style, but of consequence. "Mediated Matter: Design and Invention in the New Biological Age" gathers that work, and the conviction behind it: that we don't have to choose between nature and human endeavor. That we can design to add to life. Two decades after coining the term, we're finally building what it described: shoes grown as plants, masterplans grown as ecosystems, and more. The book is out now. With deep gratitude to MIT Press and Thames & Hudson for bringing it into the world with such care. MIT Press: mitpress.mit.edu/97802620491… Thames & Hudson: thamesandhudson.com/products… Amazon: amazon.com/dp/0262049171
25
50
1
279
14,807
it really is a new paradigm. exciting.
Replying to @rahulchhabra07
All that said, I'm generally excited about this new paradigm. It gives us a whole new axis to scale AI models - towards personalized responses. Which is more impactful to large scale adoption than just intelligence. If AI can understand you, it can make things more beautiful for you. We are hoping to help make a more beautiful world for everyone.
83
eventually we’ll just call ADHD “accelerated reasoning”
ADHD is a superpower. "The problem isn't that I can't focus; it's that I can't focus on stuff I don't care about." "On stuff I care about, I can lock in and you’re not going to find me for months." So if you have ADHD ... "The first thing you have to do is reframe the task to find the autonomy."
2
127
range no longer feels like a coordination problem, but like the beginnings of an elegant system
Highly recommended read if you work with coding agents. My general rule of thumb is to get out of the agents' way as much as I can. My top tips from the list: - Run as many evals as you can; always be constantly measuring, especially on agents' use of knowledge. - Run as many automations as you can and allow agents to do more proactive work; it's a good way to constantly test and push your agents further. This can even unlock new use cases along the way. I use complex automation to see what breaks, and it helps me find interesting new applications. My automations constantly evolve. - Scale agents to do more tasks (either in parallel, batch, etc.) and regularly find out what tools/knowledge they're missing. This is how you really optimize workflows with agents and stay ahead. Yes, a few things will break, but that's for the best, as you can always improve it. The benefit is that you constantly find weak areas to improve and unlock new productivity tricks along the way. But don't blindly do it. Always measure and stay skeptical, but trust that the agent can figure out a lot with the right tools and permissions. Your job is to give it the right tools, context, and knowledge to scale. But how can you know what to give it if you don't actually try to scale anything? Overall, just be more ambitious in every task you give your agents and get out of the way as much as you can.
2
79
I would like to add that, in addition to extracting meaning from knowledge for the purpose at hand, we might also inhere meaning to it, for the sake of purpose in the first place
51
The revolution will arrive when the cultural instrument that gives expression to our pain-filled longing can no longer contain it, or, when the longing to undo our confusion of quantity with quality, power with agency, and self with substance exceeds the forms through which we have learned to speak. Until then, the instrument can make even our refusal productive. There is much “progress” in equipping ourselves with another image to circulate, another identity to inhabit, another measure of recognition mistaken for a change in the conditions of existence. It permits us to express what hurts without permitting that expression to place its own authority in question. The rupture comes when expression becomes a demand the instrument cannot satisfy without ceasing to be what it is. When we ask no longer merely to be seen, but to participate in shaping the world in which we appear. When we cease to mistake the accumulation of signs of life for the deepening of life itself. The shift is not in perception alone, but in how we inhabit what we perceive. The aim of it all, of the instrument and our relation to it, ought not to be towards a more powerful means of expression, but towards a kinder, gentler relation between expression and existence.
1
81
have been enjoying reading/about Nikolai Fedorov
44
Christina Fedor retweeted
I built an iOS jhana app this week. Access the full Burbea recordings, a timer + journal feature to track experiments. Enjoy!! → jhanaretreat.com/app (free forever)
just made some updates to the Rob Burbea jhana retreat app, and I'll be going through myself for the next 20 days (or until our baby arrives!) → jhanaretreat.com/
24
10
3
229
10,797
Christina Fedor retweeted
Such a great post! Insights I particularly appreciated: - "Why hasn’t a huge discontinuity in damages happened since the dawn of scaled AI? This non-responsiveness of damages to a rising tide of AI capabilities is the major conceptual crux, and stumbling block, of all AI security policy today." - "cybercriminals are unskilled and get arrested a lot ... Criminal organizations need to base themselves in middle and lower income countries, where they can pay less and hire people who have few formal-economy options. ... I think folks in the AI safety and cyber world sometimes imagine adversaries as their Waluigi twins, implicit MIT graduates; in fact, in cybercrime we’re dealing with a lower-skill demographic." - "Criminals also contend with the risk and reality of arrest, sanction, interdiction, and prosecution, which caps their ability to scale: the larger the organization, the more domestic and international law enforcement attention it attracts." - "It took the attacker ecosystem a very long time to mature the workflows, business models, and industrial divisions of labor that scaled ransomware into a multi-billion dollar industry, and it’ll take a long time to develop new models that centrally leverage AI throughout criminal labor processes... it took 5-8 years for ransomware to become a huge deal" - "As risk professionals we should assume this will all change and that what we have now is a defender-dominant window in which we can prepare for some very bad possible futures" Lots of interesting charts too: joshuasaxe181906.substack.co…
Contrary to the implicit mental models of many on this website, cybercriminals aren't the security-engineer Waluigis of Stanford-educated big tech engineers, or the criminal analogs of a16z-backed AI pentesting startups. They're generally less talented, recruited from middle-income countries with weak legal oversight and fewer legal tech job options, and struggle with arrests, sanctions, retaining talent, and stable access to AI inference. Ransomware took years to become a scaled business under these conditions, and AI-native organized criminal syndicates that fully leverage the power of AI to scale may take awhile too. Nation-states, meanwhile, are quick to adopt AI for vulnerability research and exploit development, but historically slow to deeply retool military operations with new technologies. And potentially devastating agentic cyber swarms may become weapons in their arsenals alongside e.g. ballistic missiles that they possess but don't immediately use. So to those calling php2shell "the worst day in cyber ever" and expecting immediate spikes in cyber damages: there isn't the direct relationship between new capabilities and new damages you're imagining. This is good news, obviously. It means defenders have time to prepare (and we should be shouting from the rooftops that we're in a defense-advantaged interregnum, currently). Huge risks are approaching but at the pace of technology diffusion into less innovative parts of the world economy and state sector, not at the pace of AI capability growth. Full version: joshuasaxe181906.substack.co…
11
22
2
79
13,919
where the most realistic ai safety conversation lives
congrats, you vibe-coded your own software. you’re also the head of security now yesterday, Railway was breached and thousands of vibe-coders (and companies) spent the day rotating every API key / password / token they had stored on there then had to audit their entire tech stack and assess their exposure (i.e. maybe they weren’t exposed directly, but vendor A they use was) same day, PHP (the language behind ~70% of websites) patched two bugs that could let a stranger take over your server without a password if you are building your own software, all of that is on you: > you need to stay up to date on the latest cybersecurity news > you need to routinely audit and update your tech stack > you need to spring into action to reduce exposure on days like yesterday > you’re responsible and liable for your (and your customers’) data or rather than vibe-coding your own storefront, you could pay shopify $39/mo and let 7,600 experts handle payments, fraud, PCI compliance, and all of the above for you rather than vibe-coding your own newsletter platform, you could pay @beehiiv $49/mo and let 130 specialists handle deliverability, security, uptime, and everything else AI made it easier to build your own software… but it also made it easier to break into it. there will be many more p0 exploits like yesterday moving forward SaaS isn’t invulnerable, but at least those companies are staffed with senior engineering teams and security experts who handle issues like this for a living
43
many thanks to these Chinese researchers and to the modern expression of a question humans have asked in different forms for centuries: what remains dependable when conditions change? for me, such a question interrogates the difference between infrastructure and architecture. and I have found that to think in terms of architecture is to enter a dialogue between art and technics, in which we discover how variables such as context, authority, feedback, and human attention may relate well to each other. and yet, as technics scale, art is much more finicky. culturally, there is still a “been there, built that” confidence not particularly tempered by venture capital that disregards the art/technic dialogue and treats ai systems as another platform build, another chapter in man's conquest over nature. we might remind each other more often that the platform era of information technology hangs collective "expertise" on assumptions about cognitive limits. meanwhile, a compassionate regard for human cognitive limits belong to art, and in the domain of design. design-thinking, however, is rare. something here reminds me of the way psychology maps relationships among parts through a family systems model, or the way mythology does through pantheons of deities. these patterns draw our attention to the latent possibility of collective cognition, without any single variable or agent being entirely governed by whichever part is most activated or expressed. we find, across these different modes of understanding ourselves, that the organization of a system matters to the capacities it can express. so, optimizing harness diversity seems a bit like what we're aiming for when we (in psychological terms) transform a perspective into a worldview. the elusive but persistent "self" maps to what we call the model, while the conditions around the model are uncannily like our internal daemons. the capabilities that hold across conditions tells us something about capacity. and ultimately, it’s capacity that bears so much of what we care about most, especially things whose value doesn’t depend on commoditized-time economics. the machine can, we will keep finding, reflect back to us our coherence and our contradictions. ultimately, it is the architecture that will determine which patterns survive the contradictions.
Brilliant work on using many harness configurations to scale agentic RL. Many approaches use mainly one fixed harness, which specializes the model but doesn't generalize well. This work uses many harness configurations as a scaling dimension. But the big question is whether it can generalize to harness configurations never seen in training. They pick a compact training subset with balanced factorial coverage, and vary one part of the harness at a time on the same task. Across benchmarks, model sizes, and RL algorithms, this improves results on harness configurations never seen in training. Models trained this way also transfer to unseen external black-box harnesses better than models trained through one complex black-box harness. If your model will run inside harnesses you did not train it in, treat harness diversity as its own scaling axis. arxiv.org/abs/2610.29458 Chat with Paper: academy.dair.ai/papers/harne…
1
67