nvrmnd

I agree, the nomenclature is impenetrable, it's like reading software that is not well commented. Perhaps LLMs are very good at "challenging" mathematics because what we perceive as challenging is primarily the language component and not the conceptualization.

Yes, we will certainly go that way, probably code already added to gcc has been developed through collaborative AI tools. Agree we don't call that "produced by AI".

I think compilers though are a rare case where large scale automated verification is possible. My guess is that starting from gcc, and all existing documentation on compilers, etc. and putting ridiculous amounts of compute into this problem will yield a compiler that significantly improves benchmarks.

It's not fair to compare them like this!

As someone who leans pro in this debate, I don't think I would make that statement. I would say the results are exactly as we expect.

Also, a highly verifiable task like this is well suited to LLMs, and I expect within the next ~2 years AI tools will produce a better compiler than gcc.

From CNN:

"Commerce Secretary Howard Lutnick told reporters on a call Friday evening that the administration came to the fee of $100,000 per year, plus vetting costs, after talking with companies.

He noted that the payment structure is still under discussion with the Department of Homeland Security, in terms of “whether we’re going to charge the $300,000 up front or $100,000 a year for the three years.”

before there was no $100k/year cost to H1Bs, see post title.

I found this map a few years ago and had it printed online on canvas, to hang on my wall near my bike area, I recommend doing this with other old maps as well.

What is 'UQ', I assume some measure of uncertainty over your model outputs?

One thing to keep in mind with BI software is that the users are often very different than, well, those individuals that prefer to use mutt as an email client.

Many, or most, users for a BI tool will be operations, product managers, and business management who simply will not find the interface to be intuitive, responsive, or well designed. At least that's my experience.

I agree, learning admissible heuristics will retain worst case performance, which has always been the measuring stick for these algorithms. It's not at all uncommon to find faster solutions for the average or even p99 cases that cannot provide guarantees on the worst case.

If there was legislation that required Honda to install certain safety features and they failed to do so, then yes they should be liable.

This and other proposed legislation is attempting to hit the ball out of the park on the first pitch. I feel it would be a lot more sensible and effective to legislate clear and present harms, such as holding developing firms liable for deep-fake technology if used for identity theft for the purpose of fraud.

Some people say that it doesn't matter if someone dies at age 89 -- after they have lived a full life and contributed all they had to give -- it's still just as sad and shocking.

Personally, I don't agree, to me it's just not as sad or shocking. People don't live forever and Wirth's life was as successful and complete as possible. It's not a "black day" where society truly lost someone before they fulfilled their potential.

Is this an advertisement?

So the model is limited to using words and characters that already exist. I agree with you but I don't see why is a limitation worth pointing out.

I don't think that's accurate, it generates novel outputs that were not observed in the training data.

While it's easy to agree with you, I find that my opinion here has shifted after leaving a large tech company for a seed stage startup. Competing against these giant companies is really challenging, you have to me more than 2x better to get a customer to look at you a second time.

I donate to signal, and use it frequently. But I would much prefer for the app to simply charge users rather than beg for donations. Even better would be to charge users in a way that reflects the costs.

For instance, maybe verifying a new number over SMS should cost $0.10 if that's going to make up 14% of the operating costs.

Begging for donations to subsidize excessive use by other users just doesn't seem sustainable.

Free market principles do not apply well for apartments because of the limits on supplies and services and other externalities so I (like most here) am pretty supportive of legislation managing this. However, the legislative ideas and overall process for coming up with good, fair, and efficient laws in SF is absolutely abysmal. The lawmakers listen to the voters and the voters insist on laws that are just, well, bad. This affects every aspect of life in SF from housing, homelessness, schools, small business, etc.

Later in this interview they get into public schools, and the implications of the lottery system. While this is well-intentioned this is really a terrible policy, I personally moved out of SF largely because of this issue (I have 3 kiddos so private school is not an option).

The example he gives is that despite living a stone's throw away from a high school his child was allocated to a school 1 hour bus ride away. This is really how it works for anyone living near one of the better schools. In addition to it effectively destroying the community aspect of schools where your kids go to school with the neighbors which facilitates neighboring families meeting each other and becoming friends, walking to school together, and so on.

The policy is well intentioned, but ultimately does more harm than good and everyone who is closely involved and knowledgable about the system is completely powerless. The voters just don't care what any experts or those affected say, they just vote on ideology. Quite frankly it feels a lot like the way in which Texas voters could not, evidently, care less about the plight of women with complications in pregnancy.

My understanding is that "living in the same county as your last housing" can mean that you moved to California with enough money to rent a room for a month before running out. Though, I have seen many other surveys that do indicate that most unhoused individuals have lived in the area for an extended period before factors forced them out (https://www.nytimes.com/2019/11/06/us/homeless-population.ht...).

In any event, I feel that metric used by UCSF is really not a good one to the point where it's almost dishonest. It's not hard to come up with questions that do a better job at trying to answer whether CA is burdened by individuals moving from out-of-state to make use of more generous social programs and/or lax drug policy.

yes but have you tried a desktop computer on your lap

A laptop can be easily transported for use in a variety of locations.

yes, I can confirm this.

Even if Neeva was free it still was compelling enough to unseat such a dominant incumbent. Especially since the search result quality is not obviously different to the consumer.

I have heard from former Twitter employees that this was very much the policy after the Musk takeover there. Nothing is ever put in writing if it can possibly be avoided.

My oldest got a phone this year, the real issue is that we need to communicate with her when she goes to a friend's place, soccer practice, and so on. Primarily for safety reasons.

I imagine for most people their children will reach an age where they are more autonomous and require a phone for safety well before the kiddos are writing iOS applications.

I haven't taken a ride but I expect that Waymo/Cruise just set their current price/ride to 80% of whatever Uber etc. charge. Since that cost is just a made up number anyway as they have accepted operating at a loss.

When you take a hard look at exactly how to get the costs down to the point where that price is profitable, that must be a monumental challenge. The cars must cost maybe $100-150k more? Maintenance on the sensitive custom-fit equipment, software and operations costs, yikes.

I would like a print of Altman's t-shirt.

Presumably some people click-through to the article. In fact, I would guess that the majority of site traffic is generated through social media postings.

These organic posts/shares likely have significantly better CTR better than a similarly formatted ad, as they would have a "Sally likes this..." header.

I think you could easily argue that social media links to the news websites provides more value to the news website than vice-versa. If, for example, only the headline and sub-headline were quoted in the posts without the link -- which the Heritage Minister indicated would not constitute a fee -- any discussion could still take place, and certainly there would be even less benefit to the news organization.

Any news website can already prevent Facebook from generating link previews through meta tags, as I understand. But very few, if any, websites do this indicating that they understand it's not in their interests.

Perhaps people are not reading the articles because the _discussion_ is more valuable to them than the content of the article? Don't we all occasionally jump to the comment section on HN/reddit instead of reading the article? Seems to me that many news organizations have not yet figured out how to provide their customers with a product they are willing to "pay" for.

I mean, any brief visit to a small news website without an ad-blocker or, worse, a scroll through the comment section, should make this fairly evident.

should I be more worried that there's something invisibly going wrong there?

I think what anyone should be cautious about is thinking that these more sophisticated methods can allow MCMC methods to significantly scale up to larger problems. But they are probably 2-10x at best, if it's well suited to your distribution (however you characterize that). Often it's not enough to keep up with the dimensionality of the models people want to use. I remember struggling to get this to work on a Bayesian neural network that was maybe 100 parameters, tiny by todays' standards.

As for approximating hyper-parameters, again if it's an easy problem, sure, probably something will work. But I do have to say I find this really fascinating when you start to get into these harder problems with MCMC algorithms. The first thing that comes up is how do you know that a sampler is drawing samples from the correct distribution? how exactly do you quantify that? Usually each sampler takes the same amount of time to produce a sample, so what you're really comparing is which one is giving "better" samples. But.. they are all supposed to be drawing from the same distribution, so how can one sample be better than another. If they are not drawing from the same distribution, then how can you possibly know which one is the right one!

So I'm simplifying the problem here, and not to disparage any of the fine work on autocorrelation and mixing rates and all that, but you have to admire just how impossible this is generally.