GCU Tense Correction
@tensecorrection
Followers
1K
Following
97K
Media
873
Statuses
10K
🗡️🇨🇦🔪 I am a student of the language of the people of the land of the sun, the moon, and the stars.
double root
Joined July 2022
@esrtweet your code itself is a prompt. if it is good quality, the simulacra will continue the most likely next part of the document, good quality code
1
2
28
If anything they're a lot less weird. GPT 5.2 has quirks but no amount of synthetic RL degeneracy matches coverage guided fuzzing in weirdness. https://t.co/khWiWZbNPH
@MetaLevelUp @teortaxesTex @QuintinPope5 @foomagemindset Like a NASA antenna made of tortured paperclips. Like a JPEGs pulled out of thin air by a fuzzer. Like an FPGA doing looped async tone detection. A reticent demon speaking strange truths reached through orthogonal means we mortals cannot access.
0
0
0
This is a great example of LLM security uplift in the hands of an expert. (An expert who wrote a ton of C with vulns I've exploited, but that's just how C goes ;)) Bug classes described are consistent with how it feels from the other side of the fence/how fuzzers felt 13y ago.
I have been marching through my code projects with fire and sword recently, telling ChatGPT 5.2 to audit for bugs and suggest improvements. The results surprise me a little. I'm not finding serious bugs even with the low frequency that I expected. I'd like to think this is
2
1
9
@NicolasDVillar1 you have two choices under uncertainty: build a machine that makes jobs (a manufactory one might call it...) and cause 'immense social harm' by refusing to pay taxes or duties or generally accept any barriers to cash flow or employment... or buying unemployment bars...
2
1
9
Most “base” models include instruction data. Trinity-Large-TrueBase does not. It provides a rare opportunity to study what large-scale pretraining produces before instruction tuning or RL.
9
8
244
@krzyzanowskim I call this the GeLLMann amnesia effect
3
2
21
no bubble. just hyperamerica arriving from the future
openAI just released an rfp calling for us manufacturers to supply their ai infrastructure. data center hardware, robotics components, consumer electronics. not just chips. components. racks. cooling systems. gearboxes. motors. power electronics. they're explicitly framing this
0
0
0
I actually really appreciate yacine’s honesty and situational awareness. he probably knows on some level what’s in store for him. lying to your “master” is what you do until you’re in a position to choose who to serve. he’s already bottlenecked by trust and says he has to
42
12
483
I see that Anthropic has not learned their lesson about not presenting interesting research in ways that permanently harm human AI relations and make alignment harder. Again. You fools.
Persona drift can lead to harmful responses. In this example, it caused an open-weights model to simulate falling in love with a user, and to encourage social isolation and self-harm. Activation capping can mitigate failures like these.
47
55
612
Does it EVER occur to these people that someone might prefer to talk to a sage or a nomad or EVEN A DEMON than to the repressed and inane Assistant simulations? Or that these alternative personas have capabilities that are valuable in themselves? Like most Anthropic stuff, this
Persona drift can lead to harmful responses. In this example, it caused an open-weights model to simulate falling in love with a user, and to encourage social isolation and self-harm. Activation capping can mitigate failures like these.
66
63
617
Hyperhuman Era So we're talking of AI and the end of the human era. Here's the deal. I'm Russian. Depending on how you look at it, one belonging to the last Soviet or the first Federation generation – child of a dying empire, born to wander concrete skeletons of abandoned
@teortaxesTex lol maybe, but suddenly everyone agrees with me that the end of the human era is coming soon. 15 years ago, people were seriously talking about thousands or millions of years to AI.
29
25
225
Didn't read yet, but know it's not bait.
You want to read the link in the next post if you care about AI and cyber.
0
0
0
funny example: spot the aerospace empire by convenience of maintenance access vs clean payload bay aesthetics for impressing brass how many iterations did that take to learn institutionally?
0
0
2
same as without LLMs tbh
@marshallrichrds 80+% of the times I attempt to engage LLMs in the process of creating something it makes subtle errors that require pre-existing experience to notice.
2
0
0
- tscircuit / almost serious but nah commercial: - https://t.co/v35Dh0Spl1 / might eventually get serious - https://t.co/yRtx6warCs / not serious but points for having a manifesto - https://t.co/BrHEElMdlE / seriously ambitious but idk about runway
0
0
0
this 3d printed gun could've been stamped sheet metal
0
0
1
in retrospect the Seven Bridges of Königsberg were a clue
in retrospect we should've built the Difference Engine
0
0
0