vishal0123

9 ms·

1.
by vishal0123 4y ago
Just because you can't run it doesn't mean it wouldn't be useful. This likely will give insight on the kind of modelling, experiments, goal function, deployment methods, resiliency etc. And you could easily reverse engineer p
2.
by vishal0123 4y ago
> some idiot executive don't put it in full charge of mortgage underwriting There are two scenario that you are mixing: Mortgage writing using GPT make more money for the lender: I don't think it is a tech community responsibil
3.
by vishal0123 4y ago
Exactly my setup. I tried to use `conda install` few times, but every time after just few globally installed packages, conda SAT solver always struggles, and I now live with assumption that if incompatible package combination does not throw
4.
by vishal0123 4y ago
Long form messaging like letters is close to dead due to the effect of telephone and SMS. Internet eliminated the need to go to library to find information. Hell, even books/writing changed the way people acquired and stored informatio
5.
by vishal0123 4y ago
I think the biggest value would be products tracking users, just as it currently is, like Google, Facebook etc. If assuming AI could learn from low quality data like humans could now, these companies have a huge dataset consisting of multip
6.
by vishal0123 4y ago
From the paper > We tested GPT-4 on a diverse set of benchmarks, including simulating exams that were originally designed for humans.3 We did no specific training for these exams. A minority of the problems in the exams were seen by the
7.
by vishal0123 4y ago
They allow others to use it for noncommercial use. Other research groups won't have to use openAI APIs for some of the usecases, hence the model is competing.
8.
by vishal0123 4y ago
I am surprised that they were able to launch this in stanford's domain. They clearly broke TOS of both Facebook and OpenAI, and even admitted doing that. I would be happy if the research decides to ignore openAI and facebook's use
9.
by vishal0123 4y ago
68/2, not 68 2
10.
by vishal0123 4y ago
> Thus, on one hand, I'm glad they're doing this, as it should help prevent wider bank runs Could you expand? My first thought was that the bank who is in verge of crisis could tip over with additional burden.
11.
by vishal0123 4y ago
Based on my limited runs, I think 4 bit quantization is detrimental to the output quality: > /main -m ~/Downloads/llama/7B/ggml-model-q4_0.bin -t 6 -n 256 -p 'The first man on the moon was '
12.
by vishal0123 4y ago
https://medium.com/@gonced8/can-you-recover-a-blurred-image-...
13.
by vishal0123 4y ago
> It would be idiotic to further disrupt their operations for no reason. Tech has the worst P/E ratio, even big tech. It is fine if the value is inflated for disruption and future growth, else the valuation should be less than half
14.
by vishal0123 4y ago
> These failures aren't common These are common when inflation rate grows faster than expected. Since 2016 is a small timeframe to find the average.
15.
by vishal0123 4y ago
Seriously, just a tweet? Not even press release, let alone Q and A for breaking one of the core promises of the company. While I still think the deposits are safe, Coinbase should outline their worst case plan on this.
16.
by vishal0123 4y ago
I tried implementing the same and I am getting 500ms vs 20ms with wrong answer in the first call in taichi but correct in subsequent calls. I guess I found some bug in taichi: https://imgur.com/a/lpK2iVF Could you shar
17.
by vishal0123 4y ago
My assumption was that ISP couldn't get big cost saving by restricting some sites. I think this assumption is valid as by definition the websites that consumes most of the data are popular sites, and I don't think ISP would touch
18.
by vishal0123 4y ago
The code is already sent to github and hosting provider. We just have to trust that multi billion dollar companies have more to loose by using our code. And most of the codes that people work with is not rocket science, so unless openai dir
19.
by vishal0123 4y ago
This article is such a mess. If you are alleging something, then say it clearly. Judging by my participation in lesswrong and others, there are both feminists and sexist people, just like all other online forums. People are more free to be
20.
by vishal0123 4y ago
If there is competition, which ISP would you select? One that works with most of the sites or one that works with all sites? And if enough people selects the latter, and assuming blocking small percentage of site doesn't give big monet
21.
by vishal0123 4y ago
While category of sets technically could not be expressed as a ZFC set, the idea behind the set theory is enough. Also you could add an axiom[0] in ZFC to make category of set a set. [0]: https://en.wikipedia.org/wiki/G
22.
by vishal0123 4y ago
What could go wrong? It is static site and there is no backend, so the key is only in the browser.
23.
by vishal0123 4y ago
I have hard time believing it is not intentional. I think Europe could easily enforce consistent UI which could be automated through ad blocker, but now they know that most companies would rather stop serving Europe than not track users for
24.
by vishal0123 4y ago
I am not the OP, but observed exactly similar behaviour with a senior hire from Amazon: https://news.ycombinator.com/item?id=25213817 . Also the people who joined from Amazon in our organisation keep playing blame games, and
25.
by vishal0123 4y ago
> regulation will be the major inhibitor to how far it goes. This likely won't be true if China also has the technology. Regulation would give serious disadvantage to western countries while China keep improving their AI without giv
26.
by vishal0123 4y ago
From the paper > For contexts and models with d_model > n_ctx/12, the context-dependent computational cost per token is a relatively small fraction of the total compute. For GPT3, n_ctx is 4096 and d_model is 12228 >> 4096&
27.
by vishal0123 4y ago
> will set you back $100k/yr This is 6 month of salary of one average developer's salary there. And BTW they are likely doing inference on 100s or 1000s of GPUs, not just 8.
28.
by vishal0123 4y ago
https://arxiv.org/pdf/2001.08361.pdf . See the C_forward formula approxiamtion.
29.
by vishal0123 4y ago
> I have no idea how OpenAI can make money on this. I did some quick calculation. We know the number of floating point operations per token for inference is approximately twice the number of parameters(175B). Assuming they use 16 bit flo
30.
by vishal0123 4y ago
I know companies in different sectors who have mandatory expensive training with clause that the employee has to give money back to the company if they leave within 2 years.

More ›