The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Fellow "AI" fags - how much time do you think we have left?
Poast new message in this thread
Date: January 9th, 2026 9:03 PM Author: comical offensive death wish 3 years at most before all computer-facing jobs = toast?
Opus 4.5 is basically IMO there once you utilize Claude Code/MCP Servers/Skills/etc. |
Date: January 9th, 2026 9:05 PM Author: outnumbered disrespectful turdskin opus 4.5 with stockfish mcp still fucks up chess positions semi regularly. decade + (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577544) |
 |
Date: January 9th, 2026 9:14 PM Author: Green Floppy Spot Neural networks can learn to play chess well though, which makes this interesting. The neural network evaluation function built into Stockfish is quite strong even without explicit search. There’s no reason in principle that the neural network in LLMs couldn’t learn this skill. The stockfish network is trained much harder on chess positions than LLMs (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577579) |
|
Date: January 9th, 2026 11:22 PM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577892) |
|
Date: January 9th, 2026 9:05 PM Author: Orchid school I think the industry will implode because kikes & jeets can't help themselves. What rises from the ashes will destroy humanity, though. (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577546) |
|
Date: January 9th, 2026 9:07 PM Author: brindle ungodly location But who is "we" and who are "they"
No company will be able to exist, regardless of AI taking the employees jobs.
There are no moats. Every new conceivable breakthrough idea will be instantly cannibalized, disseminated, open sourced. New flash in pan social media apps will emerge and fall, getting silod along niche communities like the rest of the internet. Just an automated orgy slop fest of never ending scale, washing out all incumbents and users to recycle them from liquified refuse again. |
|
Date: January 9th, 2026 9:41 PM Author: Indigo Private Investor Selfie this is retarded
also written by ai |
|
Date: January 9th, 2026 11:02 PM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577840) |
|
Date: January 10th, 2026 6:55 AM Author: brindle ungodly location Its not retarded and I wrote it fag (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49578326) |
|
Date: January 10th, 2026 10:04 AM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49578512) |
|
Date: January 9th, 2026 9:08 PM Author: Fuchsia Glittery Theater 5 years (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577557) |
|
Date: January 9th, 2026 9:46 PM Author: marvelous topaz old irish cottage bawdyhouse two weeks (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577662) |
|
Date: January 9th, 2026 11:02 PM Author: comical offensive death wish Link? (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577841) |
|
Date: January 9th, 2026 11:10 PM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49577873) |
|
Date: January 10th, 2026 9:54 AM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49578470) |
|
Date: January 10th, 2026 10:01 AM Author: Cyan Spectacular Hell my specific field of codecucking is mostly safe from being taken over by AI. i do use AI for many coding tasks, but the codebases i work with are so enormous, complicated, disconnected, out-of-date, hard to update, often affected by ridiculous regulations, etc. that AI can't eliminate the codecucking parts of my job (i'm a manager, but i still write code occasionally). this shit is just way too complicated for AI to take over at the moment or in the foreseeable future.
but i could easily see other types of codecucking jobs being hit hard. |
|
Date: January 10th, 2026 10:03 AM Author: comical offensive death wish ???
Context7 MCP server with Claude Desktop/ Claude Code as the client
|
|
Date: January 10th, 2026 10:11 AM Author: Cyan Spectacular Hell don't want to reveal too much about what i work on specifically, but it's way too complicated, spread out, and disconnected to centralize through a simple MCP set up. the main point is that what i work on is.. exceptionally harder to work with than some relatively simple application built by 10 engineers at a startup. it's been built over several decades with tens of thousands of engineers being involved during that time. (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49578519) |
|
Date: January 10th, 2026 10:20 AM Author: comical offensive death wish if you can't use external tools for security/regulatory reasons, that's valid, but ultimately a death sentence.
Enterprise IT is its own prison.
Your "too complicated" cope is what every field says right before AI eats it for lunch. Lawyers said it. Radiologists said it.
"My job requires nuance and institutional knowledge!!!" |
|
Date: January 10th, 2026 10:36 AM Author: Cyan Spectacular Hell i will readily admit that AI could eventually take over my specific field of codecucking. i just don't see it happening soon. and i'm not fighting against it, like you might be suggesting; there have been multiple failed attempts to adopt AI at my organization (all of which i've supported). one limitation is security, as you mentioned; some AI tools we've tried to use have been limited in functionality due to security restrictions. in my particular role, i can't trust AI's suggestions for architectural changes due to several factors. at the moment, our experiments with AI have shown that it's a useful aid, but restricted to the point where it can't take over a good chunk of the work.
and as i said, i readily see other types of codecucking already being impacted. what used to take several experienced engineers to build can now be done by one or maybe two people with very limited engineering experience. |
|
Date: January 10th, 2026 6:12 PM Author: comical offensive death wish (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49579472) |
|
Date: January 10th, 2026 6:26 PM Author: Trip mint internal respiration We still have several years. If GPT 5.2 still acts like an idiot, and lies and has no empathy, and the previous model was better, and they haven’t fixed that shit in almost a year, what could change that? They’re going to got struggle to get it to work the way it should. (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49579514) |
|
Date: January 10th, 2026 6:30 PM Author: comical offensive death wish I was about to unsubsribe from ChatGPT Plus, but 5.2 came out, and I used it as an opportunity to tweak its custom instructions, and now it's easily earned its high-ranking spot on my AI Panel (only Opus 4.5 & Perplexity Pro are better than 5.2). (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49579523) |
|
Date: January 10th, 2026 6:47 PM Author: grizzly step-uncle's house AI compute is doubling every 7 months and there is still a bunch of technology on the semiconductor roadmaps that will permit substantial improvements in performance for a while.
https://epoch.ai/data-insights/ai-chip-production
Models are becoming significantly more capable still for agentic coding tasks and that’s not likely to stop anytime soon.
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
The gigawatt scale training clusters are really only coming online this year, and we are likely to see far more capable autonomous coding agents soon. Training runs will be larger, there is more room for architecture experimentation with more compute and increasingly SWE agents will do the coding work or even experimentation to identify promising research directions. There are a lot of reasons to not expect AI progress to meaningfully stall |
|
Date: January 10th, 2026 6:56 PM Author: umber sex offender theatre This thread belongs in DSM IV (http://www.autoadmit.com/thread.php?thread_id=5820183&forum_id=2\u0026mark_id=4973486#49579587) |