“Having antifragility is good. Aiming for it is bad.” vs “The way out of mimetic theory is to stop thinking about it so dang much” vs “The single most important thing is not to look for the single most important thing”#
- Two recent Tyler Cowen guests (Luke Burgis⌝ and Jared Diamond⌝) speaking about the appreciative mindset, plus Antifragile Bad source⌝
- Some things are understood or achieved by attending through their particulars to the whole they compose. When we turn those particulars into separate objects of scrutiny, we can disrupt the integration that gives them meaning. Language can guide us into that understanding, but cannot substitute for it.
- There’s something here about pointing at a thing which makes it completely lose the very essence of what the pointed at thing is. Such is the weakness of language.
- “Plans are best fused as a weak resource for ad hoc activity” — Lucy Suchman, Plans and Situated Actions source⌝#
The Cognitive Capitalist: “The returns on attention rise, and with them the incentive to pay attention.” source⌝#
- “…the desire to participate in an expanding universe of consumption opportunities.”
- “The crucial economic question then becomes not how much intelligence people can produce, but how much intelligence they can find uses for.”
- “The cognitive capitalist sees the possibility of an almost limitless expansion of demand, and a world full of minds whose potential has barely begun to be explored.”
“The future of the computer is the generative interface, not apps.” (2022) source⌝#
- I still find it very hard to get agents to understand and implement design systems, but the feedback systems keep getting better (eg. Scrawl⌝) so I just need to keep recording these chat sessions…
- System 1 models (Jev) with json-render point the way
- “We are discovering machine intelligence more than we are inventing it” source⌝#
- Every time you reduce latency by an order of magnitude, you have built a new product. source⌝#
“cyber poverty line” source⌝#
- “I think there are some good reasons to think a number of those interventions will shift economics in favor of defenders, even against model capabilities in 12-18mo.”
- “But there’s a diffusion lag time and tons of frictions. Quite concerned about the “cyber poverty line” and the long tail of e.g. critical infrastructure.”
scaling laws will drive mythos-class capabilities into small, cheap models that stop security bugs permanently source⌝#
- “the scaling laws will have driven cyber mythos-class capabilities into small, cheap models and all software dev everywhere will be using models for fractions of pennies on the dollar to stop shipping security bugs permanently.”
- tips for setting up agents source⌝#
- “Because we both shared this animal work ethic.” source⌝#
- Frontier tech startups will become indistinguishable from doing philosophy source⌝#
- institutional knowhow is API-incomplete source⌝#
- “Algorithmic thinking is really, really, really hard for the vast majority of people who have jobs” source⌝#
- “restate in your own words what you think my goal is” — I collected some prompt spells here⌝. source⌝#
- Using decision models for generative UI source⌝#
“Force failure fast” source⌝#
- Casey’s point is a criticism of university education but applies generally: failure itself isn’t the valuable part. Learning to respond to reality is. Failure with no chance to recover may simply drive people out.
- ie., get your ideas into contact with real consequences early, and learn to adapt
“most of the internet will be completely obsolete very soon … agents will be the de facto mediators of human desire” source⌝#
Give your agent a way to verify it’s work source⌝#
- This is the primary takeaway from last six months of agent work: “e2e tests, feature maps (as @poteto has been advocating), property tests and multi-agent adversarial reviews”
“Imagine a world where every person had their own personal data store” source⌝#
- “They defined their own shape of the world by how they expressed their data. … How would they map meaning from the outside world to their personal view of the world?”
- Pi Durable: “building durable agents that save what they’re doing, can restart, open on any screen” source⌝#
The brain can richly represent what you cannot consciously see. source⌝#
- Tsao’s team decoded both seen and suppressed faces from the same neurons, whose representations alternated roughly every 60 ms during binocular rivalry.
- That doesn’t establish a frame rate for consciousness. It sharpens the distinction between representation, conscious access and experience: what makes represented content become experienced?
- This means the brain responding to something doesn’t, by itself, explain you experiencing it.
“they’re trying to cut open the bird to find the song” — on tech’s “taste” discourse. source⌝#
- “Taste” is a token in the status-desire economy.
- I think this is true though less an explicit attempt than a simple conflation of understanding, primed by the usual desire-for-status mechanisms.
- Rao⌝ says this is a philistine movement, which is quite Girardian.
- “Taste” is the outside perspective where some person sees signs of knowhow in someone else’s work outputs or habits. But it is not a discussion of the knowhow itself.
- A true discussion of a person’s “taste” would point to the ontology of knowhow, where as the Discourse only talks about a sociology of taste.
- I am, of course, being naive shouting about this. Any discourse usually works by the primal human urges, actual knowledge seeking is always marginal. source⌝
- Rao: “to actually teach AIs taste, we must first introduce them to aesthetic risk, both social and material. To the costs of choices someone does not want them to make. What kinds of risk? And what sorts of costs? And imposed by whom?”
A smart pen prototype “off-the-shelf then used Codex to decompile + integrate it” source⌝#
- “If the moat around software is waning… I think there are implications to software-platforms-for-hardware too (eg Oura Ring) Possibly a world where any hardware can be pulled into any software ecosystem”
Handwired: “turn your regular JavaScript code into a neural net!” source⌝#
- Looks awesome. I’m not smart enough to understand the repercussions but I guess it goes toward the Elon quote “neural nets to pixels”, no code intermediary.
“We are all going to hyper-personalized feeds that are actually acting in our individual interests” source⌝#
- “…which suggests that groups of people who consume exactly the same content might be surprisingly powerful.”
- “…installed in our reader’s agent harness of choice, so that they get directly notified when we do something new”
Ontology for a product: “a description at the level of what the product does that is higher level than the code (therefore not a duplication) and that is formal enough (aka well-typed) to unambiguously detect drift” source⌝#
- … “Another approach is to model the organization that produces the product the same way we establish culture and process in organizations that produce products”
“Rather, the most valuable engineering work is going to be having good ideas.” source⌝#
- The important ideas are still coming from outside the software factory.
- …the purpose of this whole machine is to minimize the amount of things we have to think about that aren’t high-upside.
- Killer features are still company-making. Cursor Tab, Descript’s transcript editing, OpenRouter abstracting inference vendors behind a common billing interface … good ideas require problem intimacy and domain expertise, and engineers will need to contribute to that process.
- Killer architecture is still high-leverage. The right simplification still saves an enormous amount of complexity later on, which manifests in fewer bugs and easier product iteration. A little bit of forethought makes high-value features low-hanging later on. There’s still a lot of value in working out a more correct data model or adding a queue in the right place or building the scaffolding for a better way to test the product.