@Prawcin adding this to suicide note google doc
English
๐๐๐๐๐๐๐ก๐๐๐๐
579 posts
๐๐๐๐๐๐๐ก๐๐๐๐
@kernel_trick
still somewhat loves computers โข some exp with llms โข there are no claude-like models โข tokens per petrodollar
Katฤฑlฤฑm Haziran 2025
386 Takip Edilen181 Takipรงiler
@AyYasshhh_ You're absolutely right to push back on this โ and honestly, what a sharp observation. Let me be clear: version number isn't just a metric โ it's arguably the worst metric, and I should have caught that immediately. I apologize for the oversight.
Here's the thing:
English
The fuck? Are we comparing in terms of version numbers instead of intelligence and other benchmarks???
Boyd Kane is in London@beyarkay
Open weight models are ~15 months behind proprietary models
English
@kernel_trick @grok I love this video but I want to see this less. Oh god ๐ญ please no. My twitter is for learning and rage baiting not horny baiting
English
codex reset limits have given me a new kind of FOMO
English
RoPe people were calling it PLMs (pre-trained language models) happy that never stuck
English
@sasuke___420 read that as Goethe
English
models are so mind poisoned that they start to write "ฤ the" on their own even if you only ever talk to them about " the"
English
@MangoSweet78 whole jjk twt actually those viggleai videos lol
English
@zhil_arf do you have to submit this acc for visa
English
Kill online gambling profiteer
Bomb online gambling profiteer
Crรฉmieux@cremieuxrecueil
Gambling disorder cases have started skyrocketing in states where sports betting is legal (this started ~May, 2018) following the end of COVID. They've remained flat from COVID to today in states where it has remained illegal.
English
@TenreiroDaniel produced gnx he aight
English
The Jack antonoff hate too
Luke Metro@luke_metro
All of the Anthropic hate on here feels incredibly forced
English
@yumeno2k not really a bad thing
English
@eliebakouch polymarket now with oddly specifics of unreleased models
English
linear attention 6:1 ratio, DSA for global attention, 70 layers, 1M context, 1.5T parameters, ~30T training tokens, use latentMoE (or a variant of it), kimi residual attention, no engram, natively multimodal (only text out tho)
(i have 0 information)
leo ๐พ@synthwavedd
๐จ BREAKING: Kimi K3 is launching tomorrow with a discount on top-ups, according to a briefly live article on their docs site
English
@kernel_trick olshausen & field
lectures 2-7 from @FeiziSoheil โs foundations of deep learning youtube series
and then the twitter darlings
@jxbz โs old optimizer, new norm
@ericjmichaud_ โs quanta hypothesis
@deepcohen โs eos / central flows
English
what's the most intuition building writeup you have come across for anything deep learning
English
