shiftingleft

fine-tuning a model is very different from training the whole model in terms of resource requirements.

> Sleeping pills can cause dependence, and they often treat the symptom rather than the underlying cause.

I found gwern's take on Melatonin interesting: https://gwern.net/melatonin

A small excerpt:

> One might object that they do not wish to tamper with their natural sleep, even if melatonin is a normally-secreted hormone.

> Sad to say, I would point out to such readers that they are already profoundly tampering with their natural sleep cycle, and indeed, all of Western civilization is tampering with it; most of my readers do not even sleep multiple times during the day, as ‘Nature intends’ and as humans have usually slept through history, but rather in a single 7–9 hour long block.

> [...]

> Finally, there are multiple lines of research suggesting chronic sleep deprivation is prevalent among young adults (including historical comparisons). It is striking that unemployed adults sleep a full hour longer than the employed , and that when normal adults are placed in settings without artificial light like camping or without any time indicators, they sleep longer than before - exactly as if they were sleep deprived.

Is it really the full pipeline running on Huawei hardware? That is training and inference?

The report only talks about validating the "fine-grained EP scheme" on Huawei hardware.

I don't think memory mirring features available today allow you to race two DRAM accesses and use the result that returns earlier?

If there's anywhere I don't want LLM slop it's probably my database system.

Ah I can't read, thanks :-)

Shouldn't your snippet be using lzcnt? I can't see how this would result in the desired lookup.

for Zen5 rustc creates the following:

  utf8_sequence_length_lookup:
    shl edi, 24
    mov ecx, 274945
    not edi
    lzcnt eax, edi
    shl al, 2
    shrx rax, rcx, rax
    and al, 7
    ret

https://rust.godbolt.org/z/hz1eKjnaG

What's the API here for Google Ads? How does their site report to Google Ads whether that was a good/bad user? Is this done through conversion tracking? If so, why would you track anything but a completed purchase in the first place?

As I understand it they are placing ads on other sites and are paying for visits to their site.

  After we implemented advanced bot traffic detection and filtering, their reported traffic plummeted by 71%. [...]
  But then the sales report came in. Their actual sales went up by 34%.
  Their real conversion rate optimization (CRO) efforts had been working all along, but the results were buried under an avalanche of fake clicks. They were not bad at marketing; they were just spending thousands of dollars advertising to robots programmed never to buy anything. Their marketing ROI went from "terrible" to "excellent" overnight.

I don't understand how detecting bot traffic would directly lead to less ad spend.

Can you just tell e.g. Google Ads that you don't want to pay for certain clicks?

Did they modify their targeting to try to avoid bots?

The author admits to it.

https://www.reddit.com/r/rust/comments/1mh7q73/comment/n6uan...

The reply to that comment is also a good explainer of why the post has such a strong LLM smell for many.

The easiest way to parallelize this RNG is to just run it in parallel on multiple states.

Do they help deter people from becoming smokers in the first place?

I doubt that this is a problem in need of a technical solution. In any case, this system can easily be circumvented by emulating the key presses on that website.

Looking at a few files, there's definitely some generated comments in there. Do you have any method to quantify how much of it is (likely) generated?

> Shouldn't be too hard to do even with pen and paper since the 2-adic eval of 52! is large.

Could you elaborate?

Their talk was quite nice, they talk about experiences with other HSMs, their history, what lead them to design their own, the many aspects of their design and then go through potential attacks:

https://youtu.be/zD5EdvGs98U?t=13m23s

> The article mentions the Cochrane Review which rigorously concluded the opposite.

Do you mean this one?

"Many commentators have claimed that a recently-updated Cochrane Review shows that 'masks don't work', which is an inaccurate and misleading interpretation.

It would be accurate to say that the review examined whether interventions to promote mask wearing help to slow the spread of respiratory viruses, and that the results were inconclusive. Given the limitations in the primary evidence, the review is not able to address the question of whether mask-wearing itself reduces people's risk of contracting or spreading respiratory viruses."

https://www.cochrane.org/news/statement-physical-interventio...

Previous discussions:

https://news.ycombinator.com/item?id=24090632

https://news.ycombinator.com/item?id=34202099

There's some quite complex cryptographic machinery called Direct Anonymous Attestation that would make this possible. I don't know if they plan on using this though.

> Chia replaces Nakamoto’s energy hungry proof-of-work consensus with an eco-friendly proof-of-space.

Don't be fooled, proof-of-space will eat up SSDs like no tomorrow and you're still hoarding HDDs for little gain.

Proofs of space prove that you're storing some useless data. Typically you use HDDs to store the useless data, but to prepare such a HDD you need an SSD. The setup process will perform so many writes that the SSD will be garbage very soon.

It looks like this is a popup for a different setting. Did you watch the video outlined in the post?

The author is arguing that such a popup should also exist when locking a vault with a PIN only.

> the current "compiled" spec document is about 5 years behind the actual production ISA

How could I verify this information?

From the article:

> Some months later a person took the stage at FIDO’s Authenticate conference and annouced “a passkey is a resident key”. Because of the scale and size of the platform, this definition has now stuck. This definition has become so invasive that even FIDO now use it as their definition.

> Part of the reason this definition is hyped is because it works with an upcoming browser feature that allows autocomplete of a username and webauthn credential if the key is resident. You don’t have to type your username.

> This now means that we have webauthn libraries pushing for residentkey as a requirement for all registrations, and many people will follow this advice without seeing the problem.

> The problem is that security keys with their finite storage and lack of credential management will fill up rapidly.

Microsoft net income 2022: ~$70B. Why force pay cuts or fire workers when you have roughly $300k of net income per employee to work with?

To elaborate on this: Rust always performs bounds checks on array accesses, so you can't get an out-of-bound read/write.

Thanks for the insights!

In my defense, 1.5X is "not much" when compared with 5 to 100X :-)

Not much if you adjust the ISA for it!

https://arxiv.org/abs/2007.15919

This paper adds (more or less) a really big branch delay slot to fill the cpu with work while it's waiting for the branch to resolve.

Impossibility of Spectre-like attacks is a neat side-effect.

How much faster than the speed of light are we going here

"Mozilla:

    The DID Core spec has not demonstrated any degree of practical interoperability, instead delegating that to a registry of 50+ methods.

    The DID architectural approach appears to encourage divergence rather than convergence & interoperability. The presence of 50+ entries in the registry, without any actual interoperability, seems to imply that there are greater incentives to introduce a new method, than to attempt to interoperate with any one of a number of growing existing methods.

    The lack of restrictions on the registry are allowing methods diametrically opposed to the principles of the group & spec, and methods which are actively globally harmful to sustainability.

    [W]e believe the DID specification may not be fixable (MUST NOT become a Recommendation)."

"The Director concludes that the balance lies in favor of the DID developer community, encouraging it to continue its work and search for consensus on standard DID methods. The objections are overruled."