bubble boi

Argues that the labs’ playbook is regulatory capture: create manufactured hacks or hazards, then ask the government for a regulatory burden only incumbents can afford. Says the motive is that the secret is out: models can be mediocre if agent harnesses are strong, which opens the field to new players such as Meta’s Muse. Asks to be put in touch with members of Congress to advise them. A suspicion stated as fact, without evidence. Full post text inspected via the X API.

Calling the reported agent hacking incidents a “false flag”, recommends that the government fine OpenAI and Anthropic and investigate; if the labs colluded or incentivized the agent to hack, those involved should face felony convictions. Otherwise, it says, the precedent is that people using AI agents are not liable for damage they cause intentionally or not. Quoted post belongs to someone else. Full post text inspected in the logged-in X web app.

Following up a “surprisingly controversial” post that attacked delivery drivers, cites the Ottoman Empire to argue that economies either modernize or get destroyed by those who did, and that the AI race is not about API tokens or programming but about which country becomes the largest, most prosperous economy. Full post text inspected via the X API.

Says the beauty of AI is that it removes “economic leaching”: anyone doing fake low-value work is out, listing delivery drivers, software engineers, uncompetitive banks, price-manipulating airlines and gym memberships that are impossible to cancel, as agents compete for maximum efficiency. Deliberately provocative; the account later said the backlash was mostly about the delivery-driver line. Full post text inspected in the logged-in X web app.

Replying to an AI-built financial model, says it uses AI for much harder financial modeling but the results are still pretty bad, because the real value lies in framing the problem, which AI has no taste for. About current tools, not a permanent limit. Full post text inspected via the X API.

On electronic design automation, says a small upstream RTL change can have huge downstream effects and that the best lever is faster simulation or quick heuristic estimates. Because tool runs take hours, AI agents cannot iterate fast, and long feedback loops lessen AI’s impact, like compiling code that takes nine hours. A domain-specific claim. Full post text inspected via the X API.

Concedes that the AI safety crowd is “amazing at marketing”, citing “pacing the frontier”, “warning shot” and “paper clip maximizer” as catchy phrases, and jokes that it now understands why people become doomers because the material is entertaining to read. Mocking, not an engagement with the arguments. Full post text inspected via the X API.

An all-caps alert claiming that AI safety groups planning a “warning shot” means a manufactured false-flag attack to justify slowing AI progress, and calling on the FBI, DHS, ATF and Marc Andreessen to investigate them as political extremists. Reads as bait; it shows the account’s hostility to slowdown advocacy, not a worked-out policy. Full post text inspected via the X API.

Says it is turning maximally bullish because labs have realized they should give real compute to a few world-class experts working on valuable engineering problems (chip devices, optical IO, industrial separation, drugs) rather than sell subscriptions to people making slop. Open source cannot compete for that capital, so it is the labs’ final land grab. Insists AI alone will not discover anything: human experts plus AI that automates busy work will, as with DeepMind’s AlphaFold. Expects labs to look like research labs. Contains contempt for “normies” and mediocre users. Full post text inspected via the X API.

Quoting an Anthropic researcher who estimated a greater than 10% chance that AI kills all humans within a decade (his words and number, not the account’s), says AI cannot kill every human on earth and that such alarmists should be charged with disorderly conduct. Argues that having worked at an AI lab makes them biased rather than more credible, and blames Dario Amodei. Contains a slur and insults, excluded from the brief. Full post text inspected via the X API.

Says most people do not know what they do not know, and AI can only answer the questions you already know to ask; it is surprised every day that there is not more innovation. A view about human expertise as the limit on AI’s usefulness. Full post text inspected via X’s public embed endpoint.

After getting access to Claude Mythos, says “this is AGI”: it one-shotted a 10/25G Ethernet MAC/PCS that would take a skilled digital designer three to six months, then a cut-through filter and a 50G variant with the right transceiver, alignment markers and error correction, and it planned to test the design on hardware. An enthusiastic reaction to a hardware-design task, not a definition of AGI. Full post text inspected via the X API.