eachro

Most people don't need sustained energy over a day. For knowledge work at least, the vast majority of work occurs in very specific spurts. So it's more than fine to borrow energy from your future self to get that burst right now when you're in the middle of getting something done.

How can we know for sure that this is the actual initial prompt/instruction? Have engineers who have worked on these systems confirmed that this was the case?

Those trusted tester will likely be their RLHF part of the pipeline.

I believe the author could achieve the same result by working out of one of Meta's office. Plenty of coworkers around to mimic the effect of screen watching.

People have adapted to the subpar quality of results (on all search engines) and figured out how to find things.

As an example: grammarly is a company that makes money off getting users to "correct use of language". They make millions per year and this is just one use case of ChatGPT. Other use cases: copilot , AI personal assistant/scheduler, research summarizer (ex: Ought). Each of these is a multi million dollar business at worst.

Let's not forget that this is iteration 1 of the technology. It is only going to get better and better. The fact that it can already do all these independent tasks without having seen additional data for the task is a complete game changer.

Incredibly short sighted perspective.

Chatgpt-backed tools will become the norm in the same way that every knowledge worker uses Google on a daily basis today. Note that I did not say that people will use it to replace search - search is but a small bit of how it can be used.

No. This is a paradigm changing technology - the kind that happens once every 10 years or so. ChatGPT is going to change the entire landscape of knowledge work over the next few years and I'd rather be up to date with the advances even if it does mean I'll have to click the "More" button at the bottom of the page to see non-ChatGPT HN posts.

If you gave someone with no programming experience access to a search engine during the interview, they would likely be able to find the appropriate LC problem to copy/paste the solution.

If the interview were slightly modified so that the problem isnt googleable, a 2nd year CS major could probably map it to a problem that is LC searchable.

What are the sort of applications that FPGAs enable/accelerate?

can they do this for fighting health insurance bills?

The additional cpu to gpu to cpu round trip time I'm guessing?

- darn tough socks: these retail for 20ish a pair last I checked but they're super high quality

- water boiler: can get a basic one for less than 30 bucks, or go with a higher end one for 100+. either way, having hot water readily available will change your life

- kettlebell (20lb, 35lb): these go for like 30-40 bucks on amazon. there are a lot of compound movements you can do with kettlebells that will highlight various muscle imbalances you have. super easy to incorporate a few kb exercises into your workout routine

For ML tutorials, I feel like I do see quite a lot of tutorials in pure numpy. A fun thing I've been doing with ChatGPT is to ask it to show me some starter code for some topic. Usually it shows me some code that imports sklearn and then I'll followup asking it to give me everything written without these dependencies. It's worked out pretty well so far.

bronze*

It's interesting that so many of these tricks are specifically for testing invariance properties. I wonder if it's because those sort of questions pop up a lot, or maybe those are the type of problems that TT enjoys, or that those problems are particularly amenable to coming up with tricks to solve.

> * Invariance under linear combinations / convex combinations / algebraic operations -> suffices to check basis elements / extreme elements / generators

This is the insight that led me to really start to understand linear algebra while working through Axler's Linear Algebra Done Right.

> * Invariance under tensor products -> suffices to check one-dimensional/irreducible case

I suppose this is case dependent, but I can imagine some groups irreducible representations are way messier to handle than others!

Leetcode's problem is that the free option is already really good. The premium version's offering just isn't compelling. Official solutions while nice are often not as helpful as the community submitted solutions. Getting access to more interview questions also isn't a selling point given how many questions are already on the site for free.

why are they using OpenAI's language model instead of their own?

why was there never a Warcraft 4? Were the World of Warcraft expansions effectively continuations of that franchise?

Confidence usually comes from something. Sometimes it is innate. For those less lucky, it generally comes from preparation and just as important, the belief in the work you put in to be ready for w/e talk you're giving.

What are the engineering and considerations for serving this sort of model to billions of queries a day? Do the economics of a gpt-as-a-search-engine work?

Most companies (or at least the ones doing things properly) will also have a long running retro test to see if impact persists (new test group = don't use the new changes).

I'm not sure how accurate that is: https://www.opensecrets.org/elections-overview/biggest-donor...

What are other good options for intro programming languages if not python?

That's a fair point but from the top 10 law schools POV, they have no incentive to care about this. On the flip side, this is the sort of self-reinforcing feedback loop that allows top ranked schools to collect the best talent (or at least the kind of specific students that have the capacity to grind in the way that is necessary to succeed at Big Law firms).

If the new hires arent up to par, then this is something the the law firms should be incentivized to address. Since hiring practices generally haven't changed over the last 20 years, I'd guess that hiring from top law schools is perfectly fine (and maybe that is an indictment on the nature of work that happens at law firms).

I think instead of overall rankings, it would be more informative for law schools to have rankings for the specific things people actually care about. So things like: prospects for getting a Big Law job, getting a federal law clerkship, doing public interest work, etc.

Sure, people will always care about overall ranking for the prestige (though I expect overall rankings to correlate highly with big law job odds/overall compensation) but at least if it's split up into these very specific sub-criteria, then people can't quite game the rankings in a dishonest way.

Successful consumer product? Probably not. Google's competitive advantage is their technical infrastructure and scaleability, which are helpful to have for but not really what makes/breaks a consumer product.

Just one brand name on your resume can entirely change the trajectory of your career compensation. I see so many people working as SWEs at these small businesses or even fortune 500 companies. Sure they make 6 figures usually but their compensation is not in the same stratosphere as even entry level big tech SWE since so much of big tech pay is in equity.

Bad theory of mind for the person they're communicating with.

what are their recollections like?