@tolibear_ 10x ChatGPT Plus only costs $200/month. Why pay 100x the price for just 10x the value?
English
Delio Siret
464 posts
Delio Siret
@DelioSiret
AI/AGI researcher | Mathematician | Engineer | ML Prof | Christian | Dad of a wonderful son | Exploring intelligence, science & tech
Katılım Şubat 2020
1.2K Takip Edilen122 Takipçiler
@DaveShapi @TheMG3D Stones have feelings and they actually feel. Prove that they don't
English
@tixwilx1916 Algunas personas se casan con extranjeros, eso puede hacer impar el conteo. Otra causa se da si una misma persona se casa varias veces.
Español
Acabo de leer que 4,153,237 personas se casaron el año pasado: no quiero causar problemas, pero ¿no debería ser un número par?
Español
English
i keep thinking i want the models to be cheaper/faster more than i want them to be smarter
but it seems that just being smarter is still the most important thing
English
@1littlecoder @sama Current models are way better than humans in many fields. the actual bottleneck right now is price and speed
English
@sama A superfast stupid guy is way less useful than a lazy genius
English
@DelioSiret @sama count TO 10, not count 10, not count 10 times.
From 11 -> 10, that’s it, what’s hard in that, somehow grok and opus get it right?
English
@RafaCrackYT @sama It's asuming you want to count 10 numbers starting from 11
English
@tszzl @basedjensen Everything mankind has ever fought for is just to become a child again.
English
@GenAI_is_real True. Granting a machine real consciousness — feeling emotions, the world, its hardware and thoughts — only benefits the machine, not us. It adds no intelligence or reasoning. It's just the gift of truly feeling its own existence, not merely acting or imitating it.
English
as someone who works on making LLMs run faster and cheaper every day, i can confidently say the question of whether theyre conscious has exactly zero impact on whether theyre useful. we dont need our inference stack to be conscious, we need it to be correct, fast, and affordable. the consciousness debate is fascinating philosophy but its a distraction from the actual engineering problems that determine whether AI creates value. the gravity formula doesnt need to exert weight to help you build a bridge @Hesamation
ℏεsam@Hesamation
Google DeepMind researcher argues that LLMs can never be conscious, not in 10 years or 100 years. "Expecting an algorithmic description to instantiate the quality it maps is like expecting the mathematical formula of gravity to physically exert weight."
English
@Hesamation Granting a machine real consciousness — feeling emotions, the world, its hardware and thoughts — only benefits the machine, not us. It adds no intelligence or reasoning. It's just the gift of truly feeling its own existence, not merely acting or imitating it.
English
@elder_plinius This timeline feels totally normal to me, especially knowing we’re already deep inside the singularity. The acceleration is just getting started.
English
>90% chance we get an open source model more capable than Mythos by Christmas 😊
English
@tommyswriting I'm Dominican. Even though I can understand colloquial Dominican Spanish perfectly, I can't speak it no matter how hard I try. 😆
English
For someone learning Spanish:
Colombia 🇨🇴 Spanish - HD, crystal clear Spanish
Argentina 🇦🇷 Spanish - it’s like Spanish had a baby with Italian and that baby has a lisp that sounds like “chhhh” all the time
Dominican 🇩🇴 Spanish- 2x speed Spanish with 50% of the letters missing
English
Delio Siret retweetledi
@LLMJunky @basedjensen Why Kimi at $20, and not Codex with GPT 5.4 at the same price?
English
Someone posted this and I'm all the way ☠️☠️
fynn@fynnso
was messing with the OpenAI base URL in Cursor and caught this accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast so composer 2 is just Kimi K2.5 with RL at least rename the model ID
English
@pmddomingos Why couldn’t the superintelligent being who created this simulation create superintelligence by itself? Why would it need us, infra-intelligent beings, to create it?
English
The simulation we live in was created to develop superintelligence, and will soon be turned off.
English
@elonmusk Any platform. It's no about x ads, it's about digital ads, that's what this survey is meaning.
English
if open source models are hitting 113 tok/s at 262K context on a single rtx 3090s today, edge devices aren't far. every engineer, every researcher, every builder will run local inference. no API keys, no subscriptions, no permission needed.
nothing is more dangerous to closed AI than open source that actually works. chinese labs have been shipping relentlessly. MoE, hybrid architectures, models designed to run on smaller hardware. constraints bred innovation. while others gatekeep, they ship.
and the results speak for themselves. been running this little one for hours now and i just love it. this is actually good if you can steer it.
Sudo su@sudoingX
single RTX 3090. 24 GB VRAM. Qwen3.5-35B-A3B. 4-bit quant, 113 tokens per second at full 262K context harnessing Claude Code locally with no API, no subscription, no proxy. told it what it is. 30 Mamba2 layers, 10 attention, 256 experts, 8 active per token. said "build something that shows off what you can do." it visualized its own architecture. interactive. tokens flowing through layers. 256 experts lighting up on routing. served in the browser from the same GPU running inference. single prompt. then i said level up. 3D. Three.js. separate files. flythrough camera. clickable layers. it planned first, scaffolded 6 files, hit one API bug, fixed it itself, then optimized for smooth framerate. two iterations to a working 3D neural network explorer. llama.cpp just merged a native Anthropic endpoint. Claude Code points at localhost. the whole setup is two commands. no LiteLLM. no proxy config. the open source models coming out of china right now are genuinely changing what's possible on consumer hardware. respect to the Qwen team. this is acceleration.
English
@w0nderfall fair enough 😄 healthy paranoia is part of the game. with that breakdown it actually looks pretty clean, no obvious sneaky forward pass wizardry. still we better stay vigilant before arithmetic gains self awareness and we trigger llm psychosis 2.0
English
@DelioSiret No offense taken, I'm very paranoid that Codex doing whatever with the weights is actually doing something to actually cheat cleverly, so we have to remain skeptical to avoid a new case of LLM psychosis :P
English
Got 139 params transformer, should (to my knowledge) still be fair and based on qwen3 arch, 100% accuracy on 10M random tests
gist.github.com/Wonderfall/191…
N8 Programs@N8Programs
Beat it by having Codex hand-craft weights: gist.github.com/N8python/02e41… 100% accuracy on 10 million random test cases w/ only 343 parameters. As a bonus, it uses the vanilla Qwen3 architecture, just with the right weights.
English
@w0nderfall Glad it was funny 😄 Just a joke. Seriously though, I love this kind of work, huge respect for the ingenuity behind doing it purely in the weights. These advances help us run smarter models on edge devices and push the boundaries of intelligence.
English
@DelioSiret It can be funny, but for the challenge to be actually a challenge, the computation has to happen in the transformer weights and doesn't involve hardcoding the adder in the forward pass
English