Twitxy Saved

anton 🇺🇸

Verified account American Terawatt

@atroyn

Joined July 2017
2K Following 29K Followers 31K Posts

Latest posts 5

anton 🇺🇸 Verified account @atroyn
the trades labor shortage is real, and it is possible to make a comfortable living in the construction industry - the six figure incomes are there. the problem is how long it takes to access them for americans.

first you have to put up tens of thousands of dollars to go to trade school. then you have to spend years on slave wages as an apprentice and journeyman, before finally getting access to those opportunities.

we don't have years to address the shortage, and americans shouldn't have to wait so long to make good money building up the country's infrastructure.

we need trades reform now!
David Sacks Verified account @DavidSacks Jan 16
Narrative violation: according to Blackrock, the U.S. does not have enough workers to meet surging demand for the construction boom. AI is creating too many jobs!

axios.com/2026/01/16/bla…
View quoted post on X
7 replies 1 reposts 21 likes 10K views View on X
anton 🇺🇸 Verified account @atroyn
datacenters for hyperscale ai are a fundamentally new manufactured object, representing a never-before reached apex of industrial capacity and civilizational coordination.

turning them into neoclassical monuments would be a mistake -- a return to neoclassical architecture would be something like a fourth-hand rehash of greek forms (via rome, then 18th c. neoclassical architecture).

if they are to be symbols of progress and an american future, a new aesthetic synthesis is required. we should not be recapitulating previous forms - including the uniquely american monolithic art deco.

yes we need an architectural style for our buildings that is both popular and beautiful, but we cannot allow the history of american industrial architecture to end with LARP.
Shyam Sankar Verified account @ssankar Jan 8
We should insist that all data centers that are built are architecturally beautiful in the neoclassical style
View quoted post on X
7 replies 3 reposts 57 likes 9.8K views View on X
anton 🇺🇸 Verified account @atroyn
i suspect 2025 was the last ‘normal’ year; the magnitude of the compute buildout in 2026 & large scale hbm deployments (b200/tpuv7) unlocks the possibility of efficient reasoning post-training on frontier-sized pretrains.

simultaneously, verifiable environments across many domains will proliferate (i expect a gold rush here as with expert labeling in the past). increasingly, models will ‘just work’ on a variety of tasks.

the next step will be for the models to learn to integrate themselves. we may not get this in ‘26 but we should expect progress to continue to accelerate.
SemiAnalysis Verified account @SemiAnalysis_ Jan 6
RL Environments and RL for Science
Data Foundries + Multi-Agent Architectures
Worker Automation
RL as a Service
Anthropic's next big bet
GDPval + Utility Evals
Computer Use Agents
LLMs in Biology
Lab Procurement Patterns
Platform Politics + Access
newsletter.semianalysis.com/p/rl-environme…
View quoted post on X
14 replies 18 reposts 279 likes 35K views View on X