Post AzVot8MdbSA3HSdMIa by judgedread@poa.st
(DIR) More posts by judgedread@poa.st
(DIR) Post #AzVT64UT7Q8M2W3orw by judgedread@poa.st
2025-10-23T16:43:01.901895Z
5 likes, 0 repeats
Forget ASI, we haven't solved the government alignment problem.
(DIR) Post #AzVTBrseulnMfV8huq by ACL9000@poa.st
2025-10-23T16:44:04.866284Z
0 likes, 0 repeats
@judgedread What do we think the next abbreviation they change to will be?
(DIR) Post #AzVTL8gpmszXPL1soa by mmmgaming@linuxrocks.online
2025-10-23T16:44:07Z
0 likes, 0 repeats
@judgedread Define "government alignment"
(DIR) Post #AzVTLAOpQnmEi4cX1k by judgedread@poa.st
2025-10-23T16:45:43.966828Z
0 likes, 0 repeats
@mmmgaming People who don't follow me aren't allowed to ask questions.
(DIR) Post #AzVTNskFWijap9w0jw by judgedread@poa.st
2025-10-23T16:46:15.159050Z
2 likes, 0 repeats
@ACL9000 I think ASI arrives soon enough that it's the final euphemism.
(DIR) Post #AzVTboukZ8VWzvUb3o by brokenshakles@poa.st
2025-10-23T16:47:39.889606Z
3 likes, 0 repeats
@judgedread @ACL9000 I'm of the opinion that AGI/ASI isn't necessary, and that a large number of specialized models will serve our needs just fine. LEV doesn't need it, and that's the killer app.
(DIR) Post #AzVTbq06WiNuMp0O2q by judgedread@poa.st
2025-10-23T16:48:45.154009Z
2 likes, 0 repeats
@brokenshakles People might end up making it anyway as they extend the problem solving time horizon.
(DIR) Post #AzVTkaiqhUDtaY6m2K by brokenshakles@poa.st
2025-10-23T16:49:40.188940Z
3 likes, 0 repeats
@judgedread Yea, that's a super-exponential curve if I ever saw one, steady rise in that 3rd-order derivative.
(DIR) Post #AzVTvHYUIssVKPMMQy by judgedread@poa.st
2025-10-23T16:52:17.374613Z
2 likes, 0 repeats
@brokenshakles They want a white collar worker agent, so they need something that can carry out its duties with very little supervision based on a bit of training and a job description even when unforeseen happenings intervene. So a time horizon of months, minimum. Years if you want a manager.
(DIR) Post #AzVaJq6G7OSl8qFUSu by TheSoapMacTavish@poa.st
2025-10-23T17:40:07.112324Z
1 likes, 0 repeats
@judgedread @brokenshakles These AI agents are currently being trained and developed based on employees who hold specialized knowledge within companies. As you’ve mentioned, this talk of us already being in the singularity is very real.
(DIR) Post #AzVaTPeTJ5BE0fogca by judgedread@poa.st
2025-10-23T18:05:41.273243Z
2 likes, 0 repeats
@TheSoapMacTavish Demis Hassabis believes that LLM training is analogous to the 'training' our brains underwent over millions of years of evolution. So the cope that the models don't 'really understand' may have a false premise - that we do! Most people are just muddling through. Machines can do that too.
(DIR) Post #AzVbRiUUOPqLoSu0fI by JoshuaSlocum@poa.st
2025-10-23T18:16:35.249690Z
2 likes, 0 repeats
@judgedread @TheSoapMacTavish yep, i've had this conversation before"does AI think?">of course not"do you think?">of course i do"explain how you think"this is where peeps get kinda touchy, because most people are not "thinking" but running pattern-matching systems, and you can suss that out pretty fastit's like catching a baseballthey're not doing high level differential equations really fast in their headi'm not sold on the end result of AI one way or another, but there is absolutely no value in denying that there is no there there
(DIR) Post #AzVofHCJxmVQciCUSW by MongoMonkey@poa.st
2025-10-23T18:32:20.102110Z
1 likes, 0 repeats
@judgedread @TheSoapMacTavish I have talked about this IRL, many people operate like an LLM now. Many people have no internal monologue and their brain only fires up when prompted with a question or stimulus. They then give the answer they are programed, and then the lights go off and they go back to passive mode.The LLM can sound like a "human" because NPCs are classified as human. And the NPC is just serving up data it has been trained with like the LLM.
(DIR) Post #AzVoiAQhaDMsV9J0E4 by judgedread@poa.st
2025-10-23T20:45:13.481484Z
2 likes, 0 repeats
@MongoMonkey Most employees are NPCs.75% unemployment, here we come.
(DIR) Post #AzVot8MdbSA3HSdMIa by judgedread@poa.st
2025-10-23T20:47:12.047063Z
1 likes, 0 repeats
@JoshuaSlocum AI chatbots don't have any self awareness at the moment... at least not the consumer models. Watch the AlphaGo documentary and you suspect that the models running on $25 million worth of hardware could.
(DIR) Post #AzVou9u9MROwLdnVmS by TheSoapMacTavish@poa.st
2025-10-23T18:12:09.958152Z
1 likes, 0 repeats
@judgedread Learning anything follows an S-curve. Niggas complaining that AI is bust when a model hasn’t hit the initial exponential growth phase yet (or perhaps we already have but cannot perceive it yet…)
(DIR) Post #AzVqDPFEeQAHNfOySO by maritzaamin@poa.st
2025-10-23T20:51:30.174509Z
2 likes, 0 repeats
@judgedread @JoshuaSlocum ChatGPT has a memory of past conversations. This helps the illusion a lot
(DIR) Post #AzVqbUAvyEq3cfs7X6 by judgedread@poa.st
2025-10-23T21:06:21.350405Z
1 likes, 0 repeats
@maritzaamin But we know it's an illusion because they can't emulate billions of cortical neurons per instance with the available cloud servers.
(DIR) Post #AzVrnUdy9gAoVKk5lQ by JoshuaSlocum@poa.st
2025-10-23T21:19:46.790620Z
0 likes, 0 repeats
@maritzaamin most of what i've been interested in wrt the AI stuff has been trying to get a handle on the processpart of the issue is the speed needed so that these things don't take nine months to work the turkey through, which means you're dealing with tiny chunks of working memory, which is more or less what you have available on the GPUsso long-term memory isn't really a thing, exactly, just kind of an "oh, yeah, that"it's really cool what they're doing, and how (although fuck them for making me revisit matrix math), but i fully expect a revolution in the techniques sooner rather than later
(DIR) Post #AzZEY4L6YItD01PdxI by ACL9000@poa.st
2025-10-25T12:18:51.508286Z
1 likes, 0 repeats
@judgedread I'm sure it's a discrediting personal character limitation that I am primarily interested in how all of these things may affect my own life, but anyway I sure hope when that happens it increases what I can expect to be paid for the things of which I'm capable by five times, or more.
(DIR) Post #AzZF1n7L9DxQH29AsC by judgedread@poa.st
2025-10-25T12:24:14.170347Z
3 likes, 0 repeats
@ACL9000 AI amplifies power.My suggestion for young guys, become a robot repair tech. For a few years that will be very hot... until the robots are allowed to fix themselves.