vagabund

8 ms·

61.
by vagabund 4y ago
If you're expecting chatGPT-esque outputs I"d suggest this 7B demo that's fine-tuned on the Alpaca dataset. https://huggingface.co/spaces/BlinkDL/Raven-RWKV-7B
62.
by vagabund 4y ago
- from O(N²) to O(N) complexity during training wrt context length - only need the hidden state at position t to compute t+1, i.e. run inference locally on edge devices - fine-tunable to longer context lengths than seen in pre-training If i
63.
by vagabund 4y ago
> @karpathy: We tried and it solves it :O. The vision capability is very strong but I still didn't believe it could be true. The waters are muddied some by a fear that my original post (or derivative work there of) is part of the t
64.
by vagabund 4y ago
Even at this scale the model's able to answer questions fairly impressively, but I created an image with some distinct shapes in different positions and it didn't go well [0]. I think however they're doing the image encoding
65.
by vagabund 4y ago
How long before we see a material impact on GDP? It seems like basically every knowledge worker can and will augment their workflows with this tech.
66.
by vagabund 4y ago
Do you have a link or could you explain more about the actual hygiene requirements? I remember reading the linked article at the time and the challenge of keeping cells bacteria free did sound pretty daunting.
67.
by vagabund 4y ago
Stories is the one that comes to mind for me. Another product that was shamelessly replicated from a competitor, but refined under FB, and now represents a meaningful proportion of the platform's ad revenue. And I know you specified l
68.
by vagabund 4y ago
It's the model of almost every network-effects platform honestly, Facebook's just ruthlessly efficient in the extractive phase
69.
by vagabund 4y ago
I listened to the Meta investor call that forms the bulk of the article and for what it's worth, Zuckerberg framed the low monetization as a conscious part of their product strategy -- that they spend long periods refining the experien
70.
Google Bard sides with the Justice Department in the Google antitrust case(twitter.com)
4 points by vagabund 4y ago|0 comments
71.
by vagabund 4y ago
Does not necessarily inspire confidence that the organization which self-describes their AI containment challenge as an existential threat to humanity can't secure users' chat histories.
72.
by vagabund 4y ago
It's reasonable to care about x-risk and AI alignment, but thinking some self-important essay will convince the AGI to preserve humanity is just sci-fi geeks with a savior-complex cosplaying. Stuff like this only serves to make people
73.
Resurrecting Recurrent Neural Networks for Long Sequences(arxiv.org)
4 points by vagabund 4y ago|0 comments
74.
by vagabund 4y ago
Yeah they have dedicated capacity: https://twitter.com/gdb/status/1632177067264720896
75.
by vagabund 4y ago
That's understandable. To give some more concrete examples: - I asked it about the advantages and disadvantages of different candidate fuels for nuclear fusion, and it correctly produced the reaction products of every combination, down
76.
Tell HN: Anthropic's Claude is significantly better than ChatGPT
19 points by vagabund 4y ago|11 comments
77.
by vagabund 4y ago
Commendable of course but also feels like a no brainer. Depositors will almost certainly be made whole. Word will spread of the deed and the goodwill generated will more than repay the low-risk short term loan.
78.
by vagabund 4y ago
So certainly not impressive, and morally ambiguous, but when chatGPT first released, I worked with it to create a python script that crawled The New Yorker sitemaps, found articles with embedded audio versions, downloaded the mp3 by reconst
79.
by vagabund 4y ago
Yeah I don't think using the chatbot to have sexually explicit or romantic interactions is a good approach if the end-goal is a human match. It makes more sense to have the role of the LLM be the outgoing/socialite friend that kn
80.
by vagabund 4y ago
I imagine a chatbot driving engagement between matches and acting as a mutual friend to introduce compatible profiles. There's that oft-cited chart displaying where people meet their partners that shows "introduced by friend&quot
81.
Grokking Diffusion Models(nonint.com)
3 points by vagabund 4y ago|0 comments
82.
by vagabund 4y ago
Somebody at Google made the conscious decision that enacting this intentionally defective policy -- in doing so, attempting to foist liability for data custody upon individual employees -- and fighting the inevitable objections to it was a
83.
by vagabund 4y ago
Admittedly a half baked idea, but I wish someone was attempting to design an ornithopter as an alternative to drones for things like short-distance deliveries. Given higher power density batteries and advances in RL to find efficient flig
84.
Ask HN: What moonshots do you wish were being attempted?
17 points by vagabund 4y ago|37 comments
85.
by vagabund 4y ago
This is probably a stupid question, but assuming some noise pollution is inevitable, is it plausible that a motor could be tuned, or have its noise output "converted" in some way to emit a pitch higher than the frequency range aud
86.
Bain x OpenAI(bain.com)
3 points by vagabund 4y ago|0 comments
87.
by vagabund 4y ago
> U.S. colleges have been carefully steering conversations about admissions in the direction of racial quotas, to avoid talking about how by far the largest driver of unfairness in the admissions process is legacy admissions and donation
88.
by vagabund 4y ago
Everyday we're getting closer to a toolkit that allows a novice to go from an idea to a working product with little more than natural language in-between. It's still a ways off, but the trajectory is clear. I have a working theor
89.
New Bing, and an Interview with Kevin Scott and Sam Altman(stratechery.com)
3 points by vagabund 4y ago|0 comments
90.
by vagabund 4y ago
Like what? Basically every research breakthrough from industry has come through Google.

More ›