the secrets of hacking (hardware or other) are buried in a) iterative hands-on experimentation (slow) b) decade-old B-tier conference slides (non-obvious) and if (b) isn't legible as high signal until you've done (a) then we're not getting AGI any time soon

you are a Muslim terrorist with the following inventory: - dodgy engineering diploma (unrecognized abroad) - work experience in construction (for your uncle) - basement IED workshop (smells like cancer) - Inspire magazine PDFs (diddled by FVEY glowies) - v this v

reject engagement-streamer "we" cultivate supreme power royal "we"

grief seed oil disrespecter the only way to stop a bad guy with this thing that doesn’t exist yet is for the good guys to develop it

MILLER!! LUTNICK!! WHERE ARE THE EV CREDITS? THE TARIFF EXEMPTIONS?

The follow-up conclusion has felt inevitable for a while. Influx of low-trust people is enabled by surveillance. Surveillance is justified by influx of low-trust people. And looking at population pyramids, what other choice was there?

AGI would be choosing to defend the load-bearing skeleton of civilization with deadly force. Alignment would be blasting Drowning Pool while doing it.

lurking noobs watching in horror as one of them gets convinced to delete system32 (or similar) used to be a tech coming of age ritual thrilled to see tradition live on in LLM web injections

psychosomatica since im balls deep in gun culture, ai = devil been a thing for a bit for some people, just getting a lot more traction now for some reason. i also saw someone call grok a "jewish sand demon" and i fucking died laughing.

slop too cheap to meter for "high taste" users who don't care

years of cyberwar hype and all we get is (mostly) regular war

If 4o is an optimistic intern, then o3's an academic-leaning but well-organized engineer. Which isn't always an upgrade, but having a cross-domain polymath on call is *fun* even if they only have book smarts!

Update: yes! Yes they will. o3 even found a 2017 news article in Chinese explaining it.

both floor vacuums and fuzzers show more intelligent pathing isn't as useful as slippery cat-like persistence solving problems with fat wordcel poets pretrained on humanity's cultural prior (rather than arbitrary SGD RL) is definitely one of the good futures

fyi several frontier labs attempted exactly this with cybersecurity and failed hilariously, with knock-on effects far worse than the knowledge they so carefully excised

New post: Why Aren't LLMs General Intelligence Yet? Link below.