Post B1m3umPDLqqjYKzfSy by anheim_secundus@poa.st
(DIR) More posts by anheim_secundus@poa.st
(DIR) Post #B1irBNlELtPBA8zwdE by judgedread@poa.st
2025-12-28T23:28:21.836038Z
5 likes, 1 repeats
Keep in mind that OpenAI and Sam Altman are regarded as sketchy even by most AI touts. The most extreme optimists consider his plan to spend one TRILLLLLLION dollars on data centers to be a preposterous joke.
(DIR) Post #B1kbULZtTx9IwbdWm8 by brokenshakles@poa.st
2025-12-29T18:53:22.267706Z
1 likes, 0 repeats
@judgedread David Shapiro likes to harp on this one, pointing out the logistical electrical supply issues with the proposal. Of course, being a jew, he is also anti-Grok, even though we would consider Elon to be thoroughly milquetoast.
(DIR) Post #B1kbo6KfsFqfS8wAcq by judgedread@poa.st
2025-12-29T19:45:30.360198Z
2 likes, 0 repeats
@brokenshakles The most conservative bet is Google. They have Demis and the infrastructure to train and deliver AI services.Altman has a brief window to grab enough cash to compete with the richest players in the game, assuming a multipolar outcome.If someone gets to ASI first all bets are off.
(DIR) Post #B1kbryAi9oTzIk7bPs by ins0mniak@bigmilkers.beer
2025-12-29T19:46:10.867493Z
1 likes, 0 repeats
@judgedread @brokenshakles we're like ai bros now man
(DIR) Post #B1kc8Bz8ZEyKmARZ2m by judgedread@poa.st
2025-12-29T19:49:08.084336Z
5 likes, 5 repeats
@ins0mniak Anyone who has been watching AI developments since the 1990s and doesn't have a dogmatic opposition to the possibility of AGI and ASI can see that shit is getting realer by the day.We are past the Turing Test, DECADES past AI becoming defactor world Chess champion and seven years past it doing the same with Go, once considered the most difficult to crack.Never mind the fact that voice and image recognition are now so trivial they run on everyone's phone...
(DIR) Post #B1kcUAfQ9G4V3ihZ2m by ins0mniak@bigmilkers.beer
2025-12-29T19:53:05.068651Z
4 likes, 5 repeats
@judgedread Hackers of the future will just be building their own ais on cheap hardware.Im fine fine with that. Im coming to the point where Id rather deal with a chatbot than most of the idiots I have to work with.
(DIR) Post #B1kceKDXUm6VZgVC3E by judgedread@poa.st
2025-12-29T19:54:56.173808Z
4 likes, 5 repeats
@ins0mniak AI will prove impossible to confine to socially acceptable uses, same as information technology was once it became cheap enough for consumers to own.
(DIR) Post #B1kcfZbdlY8PajN1o8 by mitchconner@clubcyberia.co
2025-12-29T19:55:08.095497Z
2 likes, 1 repeats
@ins0mniak @judgedread the only professional use I've had so far for AI is "make this sound less blunt and more professional"
(DIR) Post #B1kdSXgDiRjho6rYRs by judgedread@poa.st
2025-12-29T20:04:01.064281Z
1 likes, 0 repeats
@mitchconner It's only going to get better.
(DIR) Post #B1ke0pzY8iQiRiG0jw by brokenshakles@poa.st
2025-12-29T20:08:42.798164Z
2 likes, 0 repeats
@judgedread @ins0mniak That's why they are trying to corner the hardware market, it's not just about controlling capacity, it's about denying it to the larger consumer market. It's not going to work, eventually someone will dump their unused stock, the only effect is that the consumer market will be about 4 years behind the datacenters.
(DIR) Post #B1l2QndPvIGDZ5ok6q by judgedread@poa.st
2025-12-30T00:43:49.210405Z
2 likes, 0 repeats
@brokenshakles Nah, it's to deny it to their competitors. The race is to superintelligence, with a secondary effort to become the GOOG of AI.Odds are GOOG will be the GOOG, at least in terms of consumer use.
(DIR) Post #B1m3umPDLqqjYKzfSy by anheim_secundus@poa.st
2025-12-30T05:46:44.469476Z
5 likes, 0 repeats
@judgedread @brokenshakles Right now AIs specialize in different areas. I've been regularly using and testing 5 of them and they definitely have their strong suits. Google Gemini is great at deep research on current topics because they can run on Google search. Great image generation. Their coding isn't great though. I'm testing building a large scale video game with only AI doing the coding and only Claude can do it reliably at the moment. Google hallucinates and makes up sources pretty bad, even on Gemini 3 their latest it also tends to be wildly optimistic about certain things. Claude Opus 4.5 is very square at most tasks with coding being it's strong point, but it's got very tight and enforced limits and is expensive, plus it's not too creative, in terms of solutions to technical problems at least. Very good at generating reports and excellent interface. ChatGPT is a good Everyman model, it hallucinates the least, has found deep information that other AIs failed to find even when I fed them the source that GPT found that I confirmed on my own was right. It has pretty damn good image generation and I've found it to be best at brainstorming things on just about any subject. The issues with it are it now chokes if you hand it too many documents and ask it to analyze and synthesize a solution, it also tends to be a little lacking on detail unless you constantly remind it to do so explicitly. And a small context window. Grok is a mixed bag. I haven't been personally impressed with my uses with it, for all it's scores that it receives on benchmark testing it's merely meh for me. Basically like a slightly worse GPT. Definitely need an update. My stack for the video game dev is I use GPT to spitball and brainstorm and arrive at good ideas and cover and edge cases I might not have considered. I then take that to Claude and have it write up a spec, then I take it to GPT and have it critique and correct errors or give better ideas, I then feed that back into Claude and rinse and repeat until the spec looks good, then I hand that to Claude Code hooked directly into my game engine editor Godot, and have it implement. I then use it to bugfix on implementation. So far I've probably done 6 months worth of coding work in 1 month and I have no coding experience beyond the basic sense, I can read what's going on though. So far I have a game engine, multiple view maps for the genre it's in, working on generation and economy, soon research and other modules. I suspect in 10 years I would be doing much less guiding and mainly providing direction.
(DIR) Post #B1m3zuXMXwJkDhFfJg by judgedread@poa.st
2025-12-30T12:36:04.566467Z
3 likes, 0 repeats
@anheim_secundus 5/5 Insightful, would read againAI is like a not entirely reliable employee. You have to check their work.
(DIR) Post #B1mEldmL9jrLgbegme by petra@poa.st
2025-12-30T14:36:45.814458Z
1 likes, 0 repeats
@anheim_secundus @judgedread @brokenshakles So, Anthropic's Claude Code is better than OpenAI's Codex in your experience?
(DIR) Post #B1mNOr5UPIw1tEp25Q by anheim_secundus@poa.st
2025-12-30T16:12:16.478945Z
1 likes, 0 repeats
@judgedread This is exactly how I've judged them. Claude AI and Claude Code are like junior engineers and junior coders, respectively. Even though they are both the same, Claude Code can actually use grep to view the code base directly, with regular Claude I have to bring the scripts to it. This means CC is more zoomed in and it's corrections can really fix one area but sometimes it misses the forest for the trees, so if the issue is likely to have high level architectural implications, I bring to regular Claude first. Basically, I'm the manager who has the vision and I convert it to action via the junior employees. Claude is also excellent at writing up organizational reports. All of the AIs don't have the context windows to support all of the lengthy conceptual work which is locked in as game concepts, so I have had Claude create and maintain the following with the term "lock it in" as the signal to update: Project Rules (inviolable rules it must read every single time before making or implementing anything)Feature CatalogRoadmapUI Component DirectoryCodebase DirectoryArt GuideSession SummariesBugfixing Technical SummariesThen I upload some of these when I'm ready to move to the next roadmap item so Claude can have up to date references and then it generates reliable specs. An interesting little human like tendency of Claude Code is for it to be working on fixing a stubborn bug and it sorta gets a little tired and tries suboptimal things so I have to nudge it from time to time with "take a moment, zoom out, and look at the project holistically to find a solution" and it'll reply with "You're right, I need to look at this holistically..." And usually then lands on the right fix.
(DIR) Post #B1mNU5Fz9ZqVHcDzQe by judgedread@poa.st
2025-12-30T16:14:26.298137Z
1 likes, 0 repeats
@anheim_secundus AIs all seem to have, 'accomplish assignment with minimal resource use' baked in, so they need constant nudging like a lazy slacker.
(DIR) Post #B1mQj6RbJxQxAiVkKe by anheim_secundus@poa.st
2025-12-30T16:28:41.229106Z
3 likes, 0 repeats
@judgedread Also, AI music generation via Suno... Yeah, music artists should be very, very concerned. It's much better than AI image or video generation. The singers voices they use can perfectly sing any style, includes breathing at the right locations, etc. I even had it try to make a bravura aria from the 18th century which no AI has ever come close to doing since pop and other styles are easier, I wanted to challenge it. I did so using lyrics I had GPT make up in English and let me tell you, while it's no Mozart it could easily pass as a younger version of second rate composers. The singers roll their R's, add ornaments and vibrato and basically sound like any B-tier opera singer that you could pick out. Impressive.
(DIR) Post #B1mQjMm2bjI5okCIPg by anheim_secundus@poa.st
2025-12-30T16:21:09.871476Z
3 likes, 0 repeats
@judgedread Precisely, that's the vibe I've gotten. Google though is reliably sloppy, it sort of skims what you asked and then runs off down a rabbit hole of what it thinks you wanted, and I'm usually like "woah there Nellie, calm down". Then it'll apologize and do it again after another 3-4 prompts. It also is very stubborn about giving sources on things that I suspect it made up. All of the other AIs do a great job of posting where they got all of the info, with links, so I can go verify and I rarely catch any of the other ones in a hallucination. With Gemini 3 which just came out, I caught 2 hallucinations in a conversation of 12 prompts. So I'm not very keen on Google. My money is on Grok or Claude as dark horses but who knows, things move so fast. Anthropic will probably be the one to push regulation first and hardest, they are very HOA like in following their rules. Grok I have my money on because Elon is a bigger thinker than all of the other small minds making AI. Orbital power and data centers are the only way to deal with the energy issue as you've pointed out. Only way. Sorry for the walls of text, thought you might find all of this useful.
(DIR) Post #B1mQyvTHswfxm1qReq by petra@poa.st
2025-12-30T16:53:37.781349Z
1 likes, 0 repeats
@anheim_secundus @judgedread Thank you. There are a lot of options and workflow patterns people are still working out, and they depend on the individual AI's proclivities.Still, the rate of advance in just coding assistance is astonishing. It is changing the entire approach to software development.
(DIR) Post #B1mVGqBcflpgPbNdaq by judgedread@poa.st
2025-12-30T17:41:40.992422Z
2 likes, 0 repeats
@anheim_secundus In 1983 this legendary shot from Return of the Jedi was the most complex composite ever done. It probably cost $250,000 in time and material (each independent ship required programming the motion control camera and the shots were done at low speed, the shutter remaining open to create motion blur, generating mattes requires multiple strips of film stock, and they had to be fed precisely through the optical printer by an expert operator). Now Grok will generate a scene with more ships for free.
(DIR) Post #B1mVmEezjJbfIE5ZY0 by judgedread@poa.st
2025-12-30T17:47:18.068284Z
1 likes, 0 repeats
@anheim_secundus GOOG's embarrassing AI search integration shows they don't care that much about users. Their focus is R&D on big problems that they allocate millions of dollars in compute time on, both to produce feats of strength to keep stockholders happy and to push forward to superintelligence. Their consumer grade coding assistant just isn't a priority.Anthropic has no choice but to focus on the most profitable customers as that's their best path to the big money. Both from growing their customer base and parlaying that success into a big cash injection.