esotericity
1K posts
esotericity
@esotericities
Machine, learning
Abstraction, TX Tham gia Ekim 2021
3.2K Đang theo dõi296 Người theo dõi
@tautologer I mean, I hope so. I’m not coping, I’m frustrated. I think they just can’t keep the right context in mind per decision
English
@esotericities this is almost certainly just an engineering problem -- more parameters, more tokens. bitter lesson
English
there is no functional difference between predicting what an intelligent person would say in a situation and actually being intelligent. in order to predict what a smart person would say you need to be at least as smart as them
onion person@CantEverDie
my biggest pet peeve around LLMs is when people (usually those invested in its success) call it “intelligent”. it definitionally, how it functions on a base level, is not intelligent. the way LLMs are built, it can never hit real intelligence. it’s just predictive
English
@tautologer Problem is they’re not reliable. At least 20% of the time they suggest things that would never even cross the mind of an intelligent person
English
ie, if an LLM can reliably predict what an intelligent person would say or do, then for all intents and purposes of entities interacting with the LLM, _it is as intelligent as that person_
English
@hoeflatoor How to unlock this ability… just meditate? On what…
English
The more qi you generate when you meditate, the more this gland opens back up. Over the past few years at moments I have experienced pulsing and aching right at the third eye, usually with very vivid dreams/out of body experiences in the days thereafter.
You can also enter deep states of stillness where you hear/see things that are downright schizophrenic - gnome-like old men behind your eyes, faces staring at you like Ghibli witches with big eyes, feeling presences enter and leave the room or hearing sentences/voices in your ears. Remote viewing, visions, all sorts of various psychic abilities begin to open up. Seeing fields/energies around people.
just rambling~
doomer@uncledoomer
in gifted/talented classes in school, they used to make you drink cups of fluoride (and probably other unspecified chemicals) "for your teeth." but the real reason is that fluoride calcifies the pinneal gland, the source of latent psychic abilities in the minds of gifted kids
English
@thsottiaux Sama - “I have underestimated the power of words and narratives”
English
@GregoryGregman @OdchanTrangmo @MinModulation This is the first time I’ve heard this detail and it actually makes sense now
English
@OdchanTrangmo @MinModulation You say “purple duck” “blue duck” “green duck” etc. to try to fake people out.
It’s just strictly better.
English
@GregoryGregman @MinModulation what the deuce?
English
@cline @sophylabshq I’ve found that worktrees add a lot of friction since they don’t copy over key repo pieces needed to run tests, like api keys or installed dependencies - am I doing something wrong?
English
Chuds call stuff degenerate. Commies call stuff bourgeois. Is there a liberal word for “thing I don’t like”?
English
@_loganlee @GeyzsoN @sama It can be good for debugging or architecture design if codex or cc can’t figure it out
English
I would like a single word for this phrase: "throw it into the maw with every bit of context I can think of".
Ethan Mollick@emollick
GPT-5.4 Pro continues to be the only model of its class. For anything really hard & complex, I throw it into the maw with every bit of context I can think of. More often than not, something very useful comes out. I can't get the same results from Codex or Code or anything else.
English
@esotericities Usually 2ish warm up sets for your first few exercises
One at 50% of working weight
One at 75
And maybe a third at 90
English
I’ll be telling people all 2026….
2 sets per exercise per session is the answer MOST of the time
3 sets per exercise per session is the answer SOME of the time
1 sets per exercise per session is the answer OCCASIONALLY
4+ sets per exercise per session is NEVER the answer
English
@VictorTaelin You wrote something relatable to many in a clever way
English
come on, among all the cool tech posts I write, is this really the one that's about to go viral? is this what I'm supposed to be, in this world? cheap entertainment? engagement bait? a living benchmark for your next shiny model, like a mere pawn in this 4d chess board, played and moved by the big labs, as they desperately attempt to justify their inflated valuations and colossal rounds, so I can at least have a place to let my voice be heard and sneak in some cool lambda calculus posts here and there, before it is all irrelevant and obsolete anyway?
I guess so
anyway is there any Pi extension that lets me call 2 models at once?
English
@SashaGusevPosts I mean, besides being proven wrong empirically, these statements are contradictory. If 1) is true and it really is simple to unlock human-level intelligence, then many teams could achieve 2) and thus 3) in parallel…
English
2) Therefore this innovation can be achieved by a small DIY team with limited resources ("brain in a box in a basement");
3) Once achieved, the team will turn AI on itself and accelerate so far ahead of the competition that it cannot be caught up ("FOOM").
English
Stumbled upon an interesting debate on AI super-intelligence from 2011. Yudkowsky makes three core claims/predictions, all of which are (to date) wrong:
1) That human intelligence is relatively simple and ASI can be achieved with a few small innovations; ...
English
@VictorTaelin @elon_age A healthy relationship is only possible with two partners willing to compromise, not just one
English
@DizzyUtd @Solaawodiya @kr0der Gemini has the best design sense. You have to use it in something like windsurf or cursor though. The Gemini cli and antigravity are perpetually and totally broken
English
i'm constantly mind blown by GPT 5.4 High as i work with it all day, so i keep giving it chances at doing frontend.
and then i end up undoing the changes and do it with Opus 4.6
GPT 5.4 is insanely smart but man the frontend 💀
English
@esotericities That was tested in BF16 because the FP4 checkpoint came in hot. But in our internal testing, we see NVFP4 at 99.8% average accuracy across our test suite, so I believe the AAIndex in FP4 should be very close if not identical.
English
Announcing NVIDIA Nemotron 3 Super!
💚120B-12A Hybrid SSM Latent MoE, designed for Blackwell
💚36 on AAIndex v4
💚up to 2.2X faster than GPT-OSS-120B in FP4
💚Open data, open recipe, open weights
Models, Tech report, etc. here:
research.nvidia.com/labs/nemotron/…
And yes, Ultra is coming!
English