jacobsimon

I’m hugely supportive of the new factories and investments, but I’m curious why they decided on Phoenix which seems to be affected dramatically by climate change and incidentally has a very different climate and culture from Taiwan.

But is this lookup mechanism available from one token prediction to the next? I’ve heard conflicting things, with others saying that transformers are stateless and therefore don’t share this information across prediction steps. I might be misunderstanding something fundamental.

Oh huh. Why not make it stateful, like re-use and compute just the “diff” when you add a new token? Assuming it’s not that easy because each token can affect attention globally.

I think I’ve read something about this but I wonder if you could abstract attention to sentence/page levels and then only recalculate the parts that are relevant.

Thanks for highlighting the KV cache, I’ve been wondering the same thing and hadn’t come across that or didn’t remember.

I like your interpretation, but how would they refer back to a plan if it isn’t stored in the input/output? Wouldn’t this be lost/recalculated with each token?

I asked it directly and it confirmed that it is based on GPT-4:

> Can you confirm or deny if you are chatgpt 4?

> Yes, I am based on the GPT-4 architecture. If you have any more questions or need further assistance, feel free to ask!

> Can you confirm or deny if you are chatgpt 5?

> I am based on the GPT-4 architecture, not GPT-5. If you have any questions or need assistance with something, feel free to ask!

It also seems to think its date cutoff is November 2023. I'm inclined to think this is just GPT-4 under a different name.

You can't say what it's about without seeing exactly where the traffic and rankings decreased and looking at the Google search console. I'm happy to take a look out of curiosity if the OP wants to share.

LOL

Hah ok point taken. Here’s the paper I found about traveling salesman - looks like a version of Grover’s algorithm so I think that means it provides a speed-up but not polynomial time? Paper is paywalled: https://link.springer.com/article/10.1134/S1063776123080095#....

Huh yeah I guess I need to learn more. My layman’s assumption was that it would help with a lot of NP problems that involved recursion or backtracking algorithm would benefit from it. From some quick googling it seems like they have already designed QC algorithms for traveling salesman etc. Isn’t that sort of meaningful or am I missing something?

I could totally see the argument that they are physically impractical and therefore not likely to be actually used vs parallelizing conventional computers.

I also find it mystifying and I’m not a SEO expert, but one way I use it is to compare ranking positions over time for specific pages and queries. That will help you identify where exactly the traffic came from and where you lost it. A lot of times this has to do with gaining or losing the #1 spot to a competitor.

If you don’t see a ranking change, it could also indicate seasonality in your visitors or external events. If your website is brand new, this could be hard to detect otherwise. For example, recently I saw a traffic spike for a random blog article, and search console help me see that people were searching for that topic because Elon Musk tweeted about it the day before.

Another helpful feature is to inspect particular URLs and make sure they are indexed—sometimes if you have multiple similar pages and set the canonical URL incorrectly, Google will try to de-dupe the results.

Hope that helps!

Not saying that’s what happened here, but if people are searching for “images of X” and you remove your image of X, how is that Google’s fault?

To the parent commenter - have you set up google search console and reviewed what pages and keywords were affected? There could be other reasons why your pages aren't being indexed properly. If it's a small site, it could have been something as simple as changing an image or page title.

I thought one of the main advantages of QC was that it could (theoretically) solve existing problems that have exponential time complexity with more efficiency. Isn’t the idea that it could make everything faster? Or did I fall for the marketing.

Feedback received! We do have free lessons deeper in the course and lots on our YouTube channel, but will pass this on to the team.

Hey I just want to add, in addition to your advice here, there are other important aspects to interviewing beyond performance on Leetcode questions. In other words, it's sometimes necessary, but not sufficient. Most senior jobs require system design interviews, conversations about leadership, and more.

I think "grinding" on Leetcode also entails learning and practicing a lot of irrelevant topics for many jobs, but there are newer resources out there that help you focus on what you need to know for specific roles. I don't like to plug my own company, but Exponent[1] is designed to help with this more 'well-rounded' interview prep for different types of engineering paths, and we also have a mock interview feature that helps you iron out your communication skills with other candidates.

1. https://www.tryexponent.com

That's definitely one of the main original use cases, but it would have been a bad decision to base the entire language around the limitations of HTML, and I doubt JavaScript today would be the most popular programming language today if they had. You can read the original ECMAScript 1 standard here, which is titled: "A general purpose, cross-platform programming language"[1].

https://ecma-international.org/wp-content/uploads/ECMA-262_1...

I think the article missed an important “why” here that is confusing people here: JavaScript objects and the DOM are fundamentally different, and JavaScript is providing an API via the DOM for reading and writing HTML, which is essentially an XML document. We shouldn’t expect the objects in JavaScript to perfectly translate to their HTML counterparts or vice versa.

If anything, it’s odd that they added these conveniences like reflection that make it more magical than it should be.

(Edited for clarity)

Cool!

I’ve tried similar experiments before by asking the LLM to generate “internal” and “external” dialog, which I think is sort of the same idea at a higher level—-and might be preferable because it would allow for easy introspection vs a new set of tokens? I’m not enough of an expert to understand whether this proposal is intended more for training or inference.

Thanks this is really close to what I was imagining, but looks like it’s been inactive for a long time.

A radical idea!

Yeah that’s what I meant by accessible, agree it’s a problem today

Agreed. I’ll take a more controversial stance that I’ve said here before: CSS/HTML are part of the problem and the community should have invested in proper UI/rendering frameworks a long time ago that weren’t designed around the quirks of HTML and CSS. Building an accessible canvas-based UI with a React-like API would make a lot more sense for SPAs, and maybe other types of websites too.

If Facebook’s website can be characterized as a SPA then wouldn’t most websites fit the definition of a SPA? Except the most static of websites like blogs.

Another option would be to build driverless road lanes, repurpose the carpool and truck lanes, or designate certain streets. Seems like a no-brainer for improving safety.

I think it also would have been easier/faster to build networked vehicles that can communicate with one another over a standard protocol, but for some reason the industry didn’t opt to standardize things in that way, and now we have a bunch of rogue driverless cars that make independent decisions instead of taking into account each other’s plans.

I don’t think this is true - it says in the article that debts of less than $500 will not show up on credit reports. It sounds like it would only show up if it went to collections though as most hospitals don’t report directly.

That's a great point. Imagine you had an AI-based browser that could adapt any website into a consistent UI or reading experience.

Ah thx

Not sure why the article is calling it a podcast, I think they’re referring to his TV show on Apple TV.