tryggth.bsky.social @tryggth.bsky.social · 03/10/2026Damn it. Does this mean Thomas will write ALL the horseshit opinions? 000
tryggth.bsky.social @tryggth.bsky.social · 22/09/2026Should have called it "Trump Intelligence" and set everyone's mind at ease. 000
tryggth.bsky.social @tryggth.bsky.social · 12/09/2026Explains the sniggering I'm hearing from my Alexa hub. 010
tryggth.bsky.social @tryggth.bsky.social · 12/09/2026The real revelation is that AI leadership could have been replaced with ChatGPT-3 long ago. 020
tryggth.bsky.social @tryggth.bsky.social · 05/07/2026I'ld put it this way: It's like saying early Homo sapiens sapiens was hopelessly provincial because he had a wife named Wilma, a baby named Pebbles and worked a blue collar job 9 to 5 at the quarry. 040
tryggth.bsky.social @tryggth.bsky.social · 04/07/2026This is technically true but misleading. Modern AI agents use LLMs as the "intuitive" drivers of larger systems that "ground" the outputs using deterministic tools. Take the current generation of automated mathematical theorem provers. The LLM loops against a deterministic LEAN prover. 1140
tryggth.bsky.social @tryggth.bsky.social · 04/06/2026It's why they are hollowing out the Generals. They don't have what is takes to join that club 020
tryggth.bsky.social @tryggth.bsky.social · 04/06/2026One of the things that fascinates me is their hyper-ventilating about defining "real men" is that they know in the company of trustworthy men they wouldn't rate the time of day. "Quit your whining or I'm going to throw you off the boat" They hate that and try to obscure it. 120
tryggth.bsky.social @tryggth.bsky.social · 02/06/2026Yup. The other thing to watch is agent architectures that do a lot with smaller parameter count models. I expect this to be the mic drop in the next eight months. 020
tryggth.bsky.social @tryggth.bsky.social · 27/05/2026Working on a formal Oxford style debate between Gemini and Grok on: Resolved: The planned dismantling of the career civil service and the implementation of Schedule Policy/Career classifications constitutes an authoritarian consolidation of state power. Now I think I know how to juice engagement 000
tryggth.bsky.social @tryggth.bsky.social · 25/05/2026Still too grainy to determine how many days he's been dead 000
tryggth.bsky.social @tryggth.bsky.social · 05/05/2026Stay on this. It's like smuggling small arms to the resistance. 000
tryggth.bsky.social @tryggth.bsky.social · 03/05/2026I remember the first time I was in Boston and I was pretty shocked. 000
Reposted by @tryggth.bsky.socialMatt Novak @paleofuture.bsky.social · 30/04/2026Fox News: "If you only watch one side or you only watch the other, you were doing great harm to yourself by not exposing yourself to more viewpoints. Well, take that and multiply it by a thousand. You know, that's what targeted algorithms do on platforms like Bluesky..." This whole clip is amazing. 17687791
tryggth.bsky.social @tryggth.bsky.social · 20/04/2026If there was leaking details will be part of the pressure campaign. 🍿 000
tryggth.bsky.social @tryggth.bsky.social · 15/04/2026Being April 15th a good story would be how DOGE failed to implement AI to find the biggest IRS fraudsters 010
tryggth.bsky.social @tryggth.bsky.social · 15/04/2026Only thing left is "where is the real Pope?!?" 000
tryggth.bsky.social @tryggth.bsky.social · 13/04/2026MALA - Make America Listicle Again zeteo.com/p/trump-insu...zeteo.com9 Times Trump Has Ridiculously Insulted ChristiansSharing an imagine depicting himself as Jesus is just one of many ways the US president has denigrated the people who put him in the White House. 000
tryggth.bsky.social @tryggth.bsky.social · 13/04/2026Loser is probably also weak on obliterating civilizations 000
tryggth.bsky.social @tryggth.bsky.social · 07/04/2026You cut it off right before we got the answer 000
tryggth.bsky.social @tryggth.bsky.social · 07/04/2026Military excursion with a side of nuclear annihilation. Just as the founders intended. 010
tryggth.bsky.social @tryggth.bsky.social · 07/04/2026I'm sure there are intelligence intercepts that say "nuke me". 000
tryggth.bsky.social @tryggth.bsky.social · 27/03/2026Just letting the wheels fly off is a lot less painful than trying to figure out what should be done 000
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026Understand. One thing I've been doing is stuff along the lines of: Harness design for long-running application development \ Anthropic share.google/SUKXpTU6xplI... To get a better feel of how things go off the rails.share.googleHarness design for long-running application developmentAnthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems. 000
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026And so that is why I mentioned LEAN. For mathematics, that is where the buck stops when the AI generates it's output. Grounding 000
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026... the structure) and then pass the compiled PDF to an adversarial agent. So yes. They generate bullshit sometimes. Who doesn't? Deliberate? Hmmm... Fixable? Absolutely. 100
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026I see this sentence: "It faked results, hoping I wouldn’t notice." Really. Though not as deep down a rabbit hole as this problem was, I found just taking the LaTeX paper (hint: the prompt is "Create a paper coded in LaTeX suitable for publication in X" if you want to avoid hand holding the... 100
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026Math is different in the sense that if you prove a theorem in LEAN and it "compiles" then the theorem is true and the proof is correct. No AI involved 010
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026If you scan papers submitted to Arxiv you will see more and more papers proven in LEAN. That isn't because LEAN is easy, but because AI is getting really good at writing proofs in it 200
tryggth.bsky.social @tryggth.bsky.social · 25/03/2026Right. I understand that. What is different about the context of mathematics (and probably physics to some extent) is that the current generation of agents have their claims "grounded" in the LEAN4 programming language. Ie, what it things is true it then proves in LEAN4, a deterministic language 110
tryggth.bsky.social @tryggth.bsky.social · 24/03/2026Nobody, no organization, can be this incompetent. We need to shut everything down until we know what is going on. 000
tryggth.bsky.social @tryggth.bsky.social · 24/03/2026MAHA be damned. Cutting off raw milk at the source 000
tryggth.bsky.social @tryggth.bsky.social · 24/03/2026This is very real and accelerating. Within a year the most competent mathematician in the world will be an AI agent. 210
tryggth.bsky.social @tryggth.bsky.social · 17/03/2026Crawling onto that commoditization railroad track because they don't want to miss the train 001
tryggth.bsky.social @tryggth.bsky.social · 16/03/2026The China meeting is almost certainly off. Because, you know, they haven't been letting this opportunity go to waste 031
tryggth.bsky.social @tryggth.bsky.social · 16/03/2026It's like a kid calling his dad from the police station after being picked up. 012
