rohit
@knowrohit07
Followers
3K
Following
10K
Media
2K
Statuses
9K
the talisman of deep learning.
Joined March 2022
mother of all demos. pt 2. bodega OS is beautiful, unlimited and all ai models running locally on your hardware. from porsche, westworld, to learning different planes of power, the giests and vishnu. playboi carti & personal notes at the end too. it's crippling to use any
7
5
50
some people think intelligence/iq is just raw mental horsepower, like having a bigger cpu. eff humans aren't just faster processors, they're better searchers. they build efficient search trees over their knowledge space, prune dead ends faster, and evolve better heuristics for
0
1
9
absolutely. and yes you, all of you have hype to ship ratio of 100:1. fake lore, un-realistic signatures, false idols, and the most funny are the company affiliates. grift shops. yes, you know who you are. there’s still time. do some actual work.
I'm very weary of any product where the hype-to-ship ratio is high. In AI, in robotics, in microprocessors, really any category. Whenever I see a company which does hype related activities really well (great at social media, makes awesome launch videos, has an absurdly well
0
0
9
i keep coming back to this. the simpler your explanation is, the better someone understands it. if a person uses more than 2 buzzwords in a sentence to explain something to you, it turns out they don’t know what they’re talking about themselves. they are fools.
@keshavchan no effort high throughput is the moat. write what you think, never spend time making it coherent. https://t.co/eeHI8LN8TE
0
0
4
this is what happens when you fumble the basics. this is how washed up people sound like on twitter when they enforce their abstractions to someone who is clear as a day on what he’s talking about. @RichardSSutton not only was right but he’s also humble and more closer to the
1
0
16
only if dwarkesh would’ve spent more time learning about continual learning instead of arguing with @RichardSSutton
.@RichardSSutton, father of reinforcement learning, doesn’t think LLMs are bitter-lesson-pilled. My steel man of Richard’s position: we need some new architecture to enable continual (on-the-job) learning. And if we have continual learning, we don't need a special training
0
0
5
do you see. can you see. this is what happens when you break their belief. now they are all branded by me.
1
0
6
it’s super cute to see people trying super hard to validate what they or their shop is doing.
anyways, softmax sampling from a conditional probability distribution is (very obviously and unambiguously) exploration this tweet is what happens when you've read the literature from before 2020 and assumed nothing happened in the field during the years that followed
0
0
3
people say i use a lot of “we” in my sentences or experiences or work. umm yeah it’s we. me and him surfing the tsunami together.
0
1
10
if you’re a founder raising money for your shop, come to me. i’ll explain how cheap engineering solutions are— by teaching fundamentals of computer engineering.
0
0
5
on the other hand, if you are obsessive about your lore. and that’s the only thing you talk about day in day out— you are blessed. you don’t need no money, no people brown nosing your posts, no hot DMs. no fake lores, no memetic warfares, nothing
0
0
1
consider this. if founders are practising any kind of “warfare” apart from actually strategising for the war— they are a “no one”. doesn’t matter if it’s trump or famous shit posters of X. if you are posting for more footfall & fake digital signatures, bro you are a corpse.
@knowrohit07 Consider the possibility that Trump won because of his time hosting The Apprentice Don’t forget to practice mimetic warfare my friend
1
0
2
keep dissing zuck. all you guys can do is diss him. all participants standing on the sideline. make fun of him. losers make fun. that neuromuscular wristband is actually generational. i can actually feel him. there have been so many times i went through the same thing. but his
1
0
10
open source is not viable strategy for producing RL environments at scales. the future of AGI lies in none of the shit all of you are doing. close to zero things will be used from what we have rn. rl env for llms is the last thing it would use for anything.
Open source is not a viable strategy for producing RL environments at scale. The future of AGI lies in closed source development.
0
0
8
when i finally get the memory allocator running at 50.3ns and my malloc calls are faster than your thoughts.
0
0
7