Top Tweets for #BioSev

[biosecure] - Utopai Agent - Test #2 Graciosa - Fuel for the Hard Times - Nuno & Astrid Intro The new Utopai film studio AI agent is not that bad with its results - if you ever finally manage to make a video at all, because the process of creating the keyframes is extremely slowww ! Witness the aftermath of the synthetic nano-parasites (SNP/NanoSinp) and BioSev/VaccineGenocide dual crisis from the extremely cumbersome AI film agent perspective.

Since last week, from an economic perspective, AI itself is in the middle of a quiet disruption and almost nobody is pricing it in, because almost no one is mentally aware of the change or thinks it is a sci-fi larp (lol imagine believing this...) The dominant assumption since 2023 or even before has been: more GPUs, bigger models, better outputs. We need one billion GPUs (or TPUs), 100 trillion parameters, use half the world's energy generation output for AI. Imagine believing this nonsense and even investing hundreds of billions of USD/EUR and all the associated human workhours into the economics of it. The assumption here has been that correctness has to be learned from scale. But this is not true for correctness inside structured domains of knowledge. Almost all domains of knowledge relevant to the worldwide human economy are structured knowledge domains of varying degrees of complexity. AEP's Deterministic Adjudication Lattices (DALs) are a governance layer that makes invalid AI outputs architecturally impossible. Once you have that layer, the economics of everything downstream change radically. A small, cheap, specialized model feeding into a high quality lattice will outperform a frontier model running alone. According to our estimations from ongoing research, it will even outperform the frontier model in most cases, when the frontier model is running inside the same lattice. The Deterministic Adjudication Lattice carries the institutional knowledge, while the model carries the proposal throughput. Ninety percent of the time the model barely "thinks", because the lattice has already previously seen the shape of a valid answer. What does this do to compute demand ? Sustained inference demand in lattice-governed domains (UIs, trading, smart contracts, IaC, robotic control, material design, avionics, etc.) collapses towards something that can be fulfilled by 2026 commodity hardware. Compute demand for training does not disappear, but concentrates into rare, targeted capability runs, rather than the continuous scaling ramp. Simulation demand survives for ground-truth calibration and shrinks everywhere else. The trillion-dollar narrative that every AI workload needs another B200 stops applying to the bulk of real work. This is a classic disruption pattern. The incumbents (like NVIDIA) do not go to zero. Their valuation multiples will just stop making sense. The layer that wins is the governance layer. The schemas. The validated knowledge. The lattices. A career advice note for biosecure unvaccinated smart people: This shift is your opening. a) Build in public, under a pseudonym if you need to. The work speaks for itself, your reputation accrues to the body of work, not the LinkedIn profile. b) Pick a Knowledge Domain Schema (KDS) that matters and own it. Medicine, law, materials, biosecurity, energy, sovereign infrastructure. Experts who encode the rules of their domain into validated schemas will be paid in a way consultants never were, because the schema compounds. c) Learn the formal verification tools. Rego, TLA+, a bit of Z3, enough Lean4 to read it. The people who can express invariants in these languages are going to be rare and load-bearing for the future economy (before the implementation of VDLE to replace money obviously). d) Work remote and on your own AI hardware where possible. The cloud is fine today and a liability tomorrow. An NVMe, one GPU, a good UPS and your own keys will outperform a $500k cloud bill for this class of work. e) Stay physically and cognitively unbroken. Long walks, real food, sleep quality, no pharmaceutical compliance you don't personally underwrite. You are going to need a ten-year horizon and a clear head. f) Find the other ones. The network of people who saw the last cycle coming is small, international and quiet. It is also where the next decade of interesting work happens. The compute economy is tilting toward structured governance. The labour economy attached to it will tilt towards people who can think in structures. If that is you, the next years are going to be your economic prime time.