Hacker Newsnew | past | comments | ask | show | jobs | submit | mynameismon's commentslogin

Well, a good way to test this is hooking up wireshark


The point of the blog is that even at "supposed" deterministic generative sampling, non-determinism creeps in. This in turn has disastrous effects in very real experiments.


My point is that greedy sampling is not just not sufficient but also not necessary for deterministic inference.


Interesting that Mastodon was blocked. How exactly was that ban supposed to be enforced, by blocking every single instance in existence?


I'm pretty sure they didn't do their research well. They probably think mastodon's app is the top result that comes up when mastodon is typed into google. They also decided to block MeWe which is weird because nobody I know has ever heard of it. Another interesting choice was Rumble. Twitch was left alone but Rumble was blocked


> MeWe which is weird because nobody I know has ever heard of it. Another interesting choice was Rumble. Twitch was left alone but Rumble was blocked

From experience, this is a symptom of them wanting to censor a specific piece of content which is on all those platforms. Look for it, you may discover something interesting.

I live in Tunisia, which had one of the most censored internet in the world before 2011.


> decided to block MeWe which is weird because nobody I know has ever heard of it

Seems to indicate they're not actually trying to prevent their citizens from doing anything in particular, they're just trying to get these international companies to follow their local laws since they operate there.


One could argue that. There were also a few services that complied a long time ago: TikTok, Viber et al. Twitter(X) is currently discussing with government about this. Also, a big population in Nepal seem to agree with this decision. I could see a lot of people celebrating the decision to block these services.


Does mastodon actually operate in Nepal?


Mastodon gGmbH does not have an operating presence in Nepal - but it is not the whole of Mastodon (or the Fediverse) by any means!

I had a quick look at the maps I could find that indicated the locations of Mastodon server instances, and I was not able to find anything local to their - of course that's not to say there is not one or more. It is important to the network that there should be many Mastodon instances, in many places, so it would be great if there were some!


It's a big country; surely there's at least one instance running.


How dare they!


Probably the usual, where they don't actually know or care about how it works, and just blocked whichever big instance they're referring to.


Your question is answered further in the article:

> To date, none of Barkley’s investments have flamed out or depreciated, largely because he prefers to come in at later stages of a company’s growth.


At the same time, manufacturers do not release operating systems with extremely obvious flaws that have (atleast so far) no reasonable guardrails and pretend that they are the next messiah.


For those who have an installation of LaTeX: It is pretty easy to use LaTeX for this.


> It is pretty easy to use LaTeX for this.

Without looking it up the arguments/syntax, how do I do "join page 2-3 from PDF A with page 7-23 from PDF B"?

If it's more than one CLI invocation, easy to remember/find in the shell history and less than 80 characters long, I'm not sure I'd call it easy :)


pdfjam [1] uses a LaTeX package under the hood, is included with the TeX Live distribution and acts as a wrapper for a LaTeX package. With this, I believe your example would be:

    pdfjam PDF_A.pdf '2-3' PDF_B.pdf '7-23' --outfile joined.pdf
I'll admit that I had to look it up but that only took about 3 minutes (it's an example in the readme).

[1] https://github.com/pdfjam/pdfjam


From my understanding of cryptography, most schemes are created with the assumption that _any_ function that does not have access to the secret key will have a probabilistically small chance of decoding the correct message (O(exp(-key_length)) usually). As LLMs are also a function, it is extremely unlikely for cryptographic protocols to be broken _unless_ LLMs can allow for new types of attacks all together.


And what if you don't list your social media accounts?


Then the applicant has defrauded the US. Lying on an immigration form is a criminal offense.


Yes. People don't take this seriously enough. I personally have known people who have been arrested and tried because of innocent things they did not disclose in their application. The charge is essentially "Lying to officials".

Of course, the US wanted them out for other reasons, but this is the simplest charge to bring against them.


Can you share any links about this?


How easy is it to run on older GPUs (think 1080Tis)? The reason I ask this is because torch.compile refuses to support that, and that alone makes things much slower.


The other issue is Pascal cards don't have tensor cores, so there much slower than those with them. You could try Unsloth for 2x faster llama fine-tuning - someone made P40s and P100s work. Although I would suggest upgrading to at least RTX 20x series.


The project is very much focused on maxing out tensor cores and since older GPUs don’t have them it’s not where the project shines best


> torch.compile

torch.compile is a pt2.0 feature and has nothing to do with handwritten cuda kernels

> How easy is it to run on older GPUs

this is a torch cpp extension

https://github.com/HazyResearch/ThunderKittens/blob/8daffc9c...

so you're going to have the same exact issue (whatever issue you're having)


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: