Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
Hugsun
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
15 ms
·
151.
▲
by
Hugsun
2y ago
Here is the overview: Certain steel and aluminum products will increase from 0–7.5% to 25% in 2024. Semiconductors will increase from 25% to 50% by 2025. Because of ramping legacy production node capacity in China. EVs will increase from 25
152.
▲
by
Hugsun
2y ago
I agree that regulation needs to justify itself. I haven't had any issue with any of the DMA rulings so far and I have greatly appreciated many of them. For that reason I trust the commission setting these rules to have imposed fair an
153.
▲
by
Hugsun
2y ago
> What the EU – and everyone else – needs to be doing is to develop, or encourage the development of, plastics which can do the job of existing materials but which will break down in the environment in a matter of weeks or months, like p
154.
▲
by
Hugsun
2y ago
Of course some people have a hard time with the changed bottle caps. There are always people who have difficulties with any type of change. I prefer the new caps, it's more convenient to manage one object than two.
155.
▲
by
Hugsun
2y ago
This is great. Booking has a massive share of the hotel booking market. It will be interesting to see how they will be made to comply. The times I've used it have been fine, although aggressively advertising car rental for a stay in do
156.
▲
by
Hugsun
2y ago
I already love the architecture and I'm excited about brutalist software. Great idea! We need a brutalist software manifesto.
157.
▲
by
Hugsun
2y ago
I'd wager that it's a humorous description, taking certain virtues to an illogical extreme. The limited functionality is nevertheless likely referring to features beside the ones described as bloat.
158.
▲
by
Hugsun
2y ago
I'd bet it can't do it now. I'd be curious to hear what it says in response. Partially because it requires time dependent reasoning about being a participant in a conversation. It shouldn't be too hard to make this work
159.
▲
by
Hugsun
2y ago
I almost always find it too verbose and unnaturally mimicy when bouncing the question back. It doesn't paraphrase my request. It's more like restating it. What I notice most is that almost always repeats verbatim unnaturally long
160.
▲
by
Hugsun
2y ago
That's very true. I believe it's a variation of chain of thought prompting they're doing. ChatGPT seems trained to do this for one.
161.
▲
by
Hugsun
2y ago
That's funny and a very good point. It also tangentially reminds me of an excellent video I re-watched recently called The Tragedy of Droids[1]. I highly recommend it. It raises interesting moral questions about the nature of droids in
162.
▲
by
Hugsun
2y ago
Very interesting and extremely impressive! I tried using the voice chat in their app previously and was disappointed. The big UX problem was that it didn't try to understand when I had finished speaking. English is a second language an
163.
▲
by
Hugsun
2y ago
At first I thought they were playing some semantic game. Something like LLaVA being a language to vision model but I can't steelman the idea so it makes sense. Maybe they're just lying?
164.
▲
by
Hugsun
2y ago
The model type is a good point. It's hard to track all the variables in this very fast paced field. Thank you for sharing your CodeGemma experience. I haven't found a emacs setup I'm satisfied with, using a local llm, but it
165.
▲
by
Hugsun
2y ago
> New Falcon 2 11B Outperforms Meta’s Llama 3 8B, and Performs on par with leading Google Gemma 7B Model I was strongly under the impression that Llama 3 8B outperformed Gemma 7B on almost all metrics.
166.
▲
by
Hugsun
2y ago
This test is very helpful. Thanks for sharing!
167.
▲
by
Hugsun
2y ago
This page is confusing to me. How is it useful to you? I can see some utility but am curious if there's something I'm missing.
168.
▲
by
Hugsun
2y ago
My SO is very grateful for flowers she tells me to buy. I second your statement.
169.
▲
by
Hugsun
2y ago
The first thought that came into my head reading this title was: "Ah, so they turned off the quantum effects for someone and they became dumb". I guess that someone was me.
170.
▲
QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs (SOTA sub-4-bit inference)
(arxiv.org)
1 points
by
Hugsun
2y ago
|
0 comments
171.
▲
by
Hugsun
2y ago
In my eyes, Dwarf fortresses interesting stories emerge from the events that happen within the game. It doesn't require any blank filling, although it doesn't suffer if you do it either. > ai keeps breaking your head canon. We
172.
▲
by
Hugsun
2y ago
This seems very pessimistic to me. For a linear storyline, highly dynamic NPCs are likely not going to add to the experience, that is correct. There are however other ways of creating interesting narratives. A good example is dwarf fortress
173.
▲
by
Hugsun
2y ago
It is still concerning to many that they are at all capable of deception. The ramifications are significant. Even it it occurred in controlled circumstances.
174.
▲
by
Hugsun
2y ago
What happened with magicka?
175.
▲
by
Hugsun
2y ago
I didn't know about these problems until now and this article didn't tell me anything besides that there are issues in the nix community. It's a bit like long form subtweeting.
176.
▲
by
Hugsun
2y ago
This seems like a good effort. SDXL is however trained on a large amount of data with unknown, likely copyright infringing material. Using a fine-tune or LoRA based on a consenting artists works doesn't remove the infringement of the o
177.
▲
by
Hugsun
2y ago
I'm greeted with a login form. I don't proceed. If you're the author, consider a landing page explaining why I should want to use this. I suspect many people will be put off by having to sign up to see what this is. I was.
178.
▲
Take back your dotfiles with Chezmoi (2020)
(fedoramagazine.org)
2 points
by
Hugsun
2y ago
|
0 comments
179.
▲
by
Hugsun
2y ago
I love this term, openwashing. It stumbles off the tongue.
180.
▲
by
Hugsun
2y ago
On that note, to me, the current year has been the best year for music for a long time. Simply because of volume and variety.
More ›