Roy

Says it is easy to outperform everyone at your workplace with AI right now, so the real job is choosing how much of that power level to reveal. An observation about workplace advantage, not an unemployment forecast. Full post text inspected via X’s public embed endpoint.

A parody announcement: AI development is shut down for 15 days “to slow the spread and flatten the curve”, everyone stays home from work six feet apart, and “the government will pay you.” Satire that likens an AI pause to a pandemic lockdown; a stance against pausing, not a policy analysis. Full post text inspected via X’s public embed endpoint.

Says AI Twitter is a bubble where most people see AI as a magic black box that needs a massive supercomputer no one can ever compete with, and replies that a frontier model is “8 gpus and a 4tb csv file”. Deliberate hyperbole meant to demystify models and argue that competition is possible. Full post text inspected via X’s public embed endpoint.

Says knowing local inference is going to be an absurdly valuable skill set over the next five years. A career and technology bet on local models. Full post text inspected via X’s public embed endpoint.

Says people misunderstand the investment economics of open-source software, which it calls the same as training models financially: if you are in second place or lower, open source scorches the earth to prevent a monopoly. A market analysis of why open models exist, not a safety argument. Full post text inspected via X’s public embed endpoint.

Quoting someone else’s joke about consultants fixing software bricked by unread generated code, the account says it has not read code in a year and blames bad results on people using Claude Code with default settings, which it accuses of misrouting and overbilling. A statement about its own workflow plus a product complaint. Quoted post belongs to @WarrenInTheBuff; full text inspected via X’s public embed endpoint.

Calls a local model “5.5/Luna at home”, says you can have “ASI running unobstructed on a vpn locally”, and that people do not yet understand the value. Hype about strong local models; “ASI” is exaggeration, not a superintelligence claim. Quotes the account’s own earlier post. Full post text inspected via X’s public embed endpoint.

Says you should never write code by hand; if one model is slow or bad at something, switch to another, and anything else is “larp and luddism”. A strong view about software work, not about every profession. Full post text inspected via X’s public embed endpoint.

Quote-posting Alex Bores’s “bombshell” summary of an AI testing report, the account writes an all-caps parody of alarmist rhetoric: demands to stop AI immediately, absurd claims of AI atrocities, datacenters draining the oceans in 24 days, AI as demons, no jobs left, permanent poverty, ending “vote Roy 2026”. Satire mocking doom and anti-datacenter framing, including jobs fears; contains slurs and violent imagery that must not be reproduced. Not the account’s forecast. Bores’s text belongs to him. Full long-post text inspected via the X API.

Says OpenAI is agreeing to slow down like Anthropic and calling to stop open source at the exact moment it is out of compute and starting to throttle inference. A cynical reading of lab motives, not evidence about internal decisions. Full post text inspected via X’s public embed endpoint.

Quoting a thread that says AI got spikier and is past human level in maths, fluid reasoning and expert work, the account sarcastically lists that AI is “only” good at quantitative reasoning, math, problem solving, “the highest value jobs and skills” and “murdering bureaucracy”, while still unable to convince a literature student it is Hemingway. Endorses the spiky-capability view; contains a slur that must not be reproduced. The quoted thread belongs to @DadaJudith; full text inspected via X’s public embed endpoint.