sinenomine

6 ms·

31.
by sinenomine 1y ago
>not a single mention of macroeconomic causes Irredeemably.
32.
by sinenomine 1y ago
Both, but even just the second one could be useful.
33.
by sinenomine 1y ago
Is it healthy for society to spend 20 years (and increasing) of schooling to get maybe 5-10 years of peak scientific output from a person, though? Wouldn't it be interesting if we could witness geniuses honing their unique thoughtcraft
34.
by sinenomine 1y ago
Will they be cited tho…
35.
by sinenomine 1y ago
Aging. Nothing brings more suffering. Resource needs are considerable due to long preclinical & clinical studies necessary, perhaps also due to novel paradigms which will have to be invented along the way.
36.
by sinenomine 1y ago
TLDR: the whole embedded dev is a cost center, very underfunded compared to normal SWE. I say it from position of compassionate interest.
37.
by sinenomine 1y ago
Marcus' points routinely fail to pass scrutiny, nobody in the field takes him seriously. If you seek real scientifically interesting LLM criticism, read François Chollet and his Arc AGI series of evals.
38.
by sinenomine 1y ago
A voice from outside SV (mine): We can extend life and prevent disease in animals and should do the same in human beings. As much as the West has exclusive access to biomedical R&D, it's an ethical imperative for the West to pursue
39.
by sinenomine 1y ago
1. As other commenters point, many cases have intact memory and circuits for what would be considered "lost", which activate from time to time. So it's likely a question of SNR and tissue vitality (e.g. basic capillary functi
40.
by sinenomine 1y ago
It's not entirely binary, but there is an obvious unpleasant tendency in e.g. Canada, to soft-push MAID onto potentially treatable patients who don't even seek medically assisted death. I wouldn't want my government to have a
41.
by sinenomine 1y ago
With all due respect, you didn't even try at anything resembling full effort . There are 2, maybe 3 societies in the world, which have (or had) the capacity to try implementing a serious R&D effort to prevent the diseases of aging
42.
by sinenomine 1y ago
Very likely answer is no: https://slatestarcodex.com/2018/04/04/adult-neurogenesis-a-p... ... but genome and local cell environment is malleable, and it is quite likely that some combination of molecular sign
43.
by sinenomine 1y ago
There is no good answer just yet, but it is either an information theory phenomenon (brain as a prediction machine theory), or this could be linked to activity of dopaminergic, cholinergic systems, or even changes to base neuron firing rate
44.
by sinenomine 1y ago
Reading should be considered a basic ability not hindered much even by average intelligence and being amenable to rote learning. A tangent, but I also wouldn't rule out specific adaptations for reading having been evolved during the la
45.
by sinenomine 1y ago
... Or is it? If it is mostly heritable, pushing children into it will merely bring sorrow. I'm simply against ruining childhood of those who don't have it in them. Environmental interventions are devilish: promising and not deliv
46.
by sinenomine 1y ago
The existence of this article is explained by simple fact of nature: regression to the mean applies, on average, to heritability of various quantitative human traits, such as general intelligence and mathematical aptitude (and to more speci
47.
by sinenomine 1y ago
Just read the code, bro /i
48.
by sinenomine 2y ago
No, you miss the point. My point is very low-level biological in nature: the neocortex is a cognitive system which developed via evolution to process complex sensorimotor loops. Very recently in the history of our species it was imperfectly
49.
by sinenomine 2y ago
Impressive use of reasoner CoT distillation method applied to deepseek R1. MIT license for the weights. Thanks, Deepseek!
50.
by sinenomine 2y ago
That's a lot of consideration from someone who transitioned from hands-on farm work to ML, but I’d propose a simpler take: sensorimotor loop integration is a core brain function. When we spend years extremely prioritizing logic and ver
51.
by sinenomine 2y ago
You could be the first if you were to develop an eval (preferably automated with llm as judge) and compared local deep research with perplexity's, openai's and deepseek's implementations on high-information questions.
52.
by sinenomine 2y ago
It is widely admitted by practitioners that all frontier models with very few exceptions (e.g . Mistral-large, llama 3) are MoE. This includes gpt4-turbo you mentioned.
53.
FrankenVAR: Enabling generative vision modality for text-only LLMs(twitter.com)
2 points by sinenomine 2y ago|1 comments
54.
by sinenomine 2y ago
The idea looks technically solid to me. Would be great if someone committed a few grand to check if it works on Llama-8B.
55.
by sinenomine 3y ago
Remember, this is going to be "it" instead of full-on science-driven "war on Alzheimer's". It is not that we cannot do more; it looks like the US public & private investment arms decided to avoid maximizing the
56.
by sinenomine 3y ago
Reminds me of Houellebecq's "Extension du domaine de la lutte": " It's a fact...that in societies like ours sex truly represents a second system of differentiation, completely independent of money; and as a system o
57.
by sinenomine 3y ago
The same von Neumann that famously argued for (nuclear, apocalyptic) first strike at USSR.
58.
by sinenomine 3y ago
This is a very underrated thought. Untold amounts of suffering could be alleviated, if we made an investment of political & human capital.
59.
by sinenomine 3y ago
The repeated dataset regime is better characterized than people think, see Meta's Galactica paper: https://arxiv.org/abs/2211.09085
60.
by sinenomine 4y ago
If you want quality, use Google's Apache-licensed LLM https://huggingface.co/google/ul2

More ›