psyklic

Another major issue is that if enough people believe/support a lie, it doesn't get noted. So in practice, you end up having "official" lies that go un-noted, making them appear true.

Not seeing fact checks likely means it's working: "Once third-party fact-checkers have fact-checked a piece of Meta content and found it to be misleading or false, Meta reduces the content’s distribution "so that fewer people see it.""

The issue with Community Notes is that if enough people believe a lie, it will not be noted. This lends further credence to a certain set of "official" lies.

Everyone must learn new things after university. For example, LLMs, blockchain, and new frameworks likely weren't covered. This is why they say in university you learn how to learn.

My personal opinion is that universities should teach things that would be useful to an expert which you won't have time to learn at a job. Like math and gaining deep understandings of things.

That said, I'd assume if you went to university in computer engineering (i.e. embedded systems/digital electronics), you'd have numerous labs and projects designing and using microcontrollers.

In CS, you might have one lab/class on CPU architecture/assembly but I'm guessing this would be a specialized focus area.

Larger clients will often require you to have liability insurance. I was able to get some quotes for this as an independent consultant myself. However, given the extra expenses you'll have to charge more. I'd suggest using this higher cost as leverage toward the liability they want you to accept (for you, ideally as-is). I'd also explain since it's hourly if they dislike your work they can just replace you.

IANAL but my understanding is that for contractual work in the US you can only realistically be sued for at most the amount of the contract (supposing non-negligence). Asking you to accept liability beyond this sounds risky.

I've taught beginners both languages and have not noticed a difference.

The hard parts for beginners are identical -- e.g. for loops, functions, and finding the motivation to consistently practice.

It's very likely that Stripe initially pitched very cutting-edge, ambitious ideas to VCs. So, comparing seemingly "boring" mature company to new startups may be misleading.

The easiest way to get started is look around for others by themselves, then walk over and say hi. Realize they're likely feeling similar to you.

Every time you sit down for a talk, challenge yourself to sit near someone and introduce yourself. At lunch, find an empty seat at a busy table. As you do this, you'll see the people you met earlier and will feel more confident.

Also ask people if they know if anyone's gathering afterward. Nearly everyone will either invite you, tell you what they know, or be interested in doing something after too.

The NYT recently published a great article on how sugar affects the body: https://www.nytimes.com/interactive/2024/10/30/well/eat/suga...

Caltech's Honor Code is simply:

"No member of the Caltech community shall take unfair advantage of any other member of the Caltech community."

- https://deans.caltech.edu/HonorCode

Same reason as programming language flamewars. Coders assume their individual experience holds for every domain. Yet LLMs are good for some things, bad at others.

Facts still must be interpreted. It could be a fact that a credible source claimed X and a non-credible source claimed the opposite. A newspaper's job is to also provide this context, not merely the "facts" of what was stated.

Similarly, an endorsement could in countless ways not simply be partisan. For example, based on what is most predicted to help the country with X/Y/Z.

I believe this wouldn't be meaningful, since any size LLM can be trained on any amount of data.

You could measure how well it memorizes via prediction accuracy on the training set, but this wouldn't indicate whether it generalizes well.

Yes! Also, there's no need for a blog unless you have a passion for writing. You can just post past projects and make it like a more engaging CV.

- Making your website more unconventional will result in more variance of opinion. This can be really good if some people especially like it (ofc, others may especially dislike it!).

- I consider my website as controlling the top Google result for my name. Also, my email uses my domain. So people I email will also likely visit my website, which hopefully leaves a better impression than LinkedIn would.

- If I apply to something I care about, I can see in the logs that someone from there likely looked at it. So from that I can say it likely helped with grad school admissions, and certainly most clients have looked at it before hiring me (in fact many have mentioned it positively).

True, but if they accept credit cards then customers will rarely pay with cash. We assumed here cash is a major transaction method, since otherwise their cash management costs are minimal.

Merchants are able to price things to result in exact numbers, and some do especially if they sell individual items.

However, many customers (1) throw odd change in the tip jar, and (2) an exact amount might discourage add-on purchases.

In this case, an exact amount would certainly result in zero tips. The customer would need appropriate bills and wonder how much would be appropriate (since 15% of $60 is too much).

Additionally with cash, people often just don't buy things. If they're running low of cash on-person, don't want to lose a certain type of bill (e.g. for the bus/tips), or simply don't want to carry coins.

For the LLM itself, length matters. For example, the final logits are computed as the un-normalized dot product, making them a function of both direction and magnitude. This means that if you embed then immediately un-embed (using the same embeddings for both), a different token might be obtained. In models such as GPT2, the embedding vector magnitude is loosely correlated with token frequency.

It is very common to have co-authors solely at other institutions. It does not imply that the first author asked for (or even wanted) co-authors at their own university, muchless that others rejected the opportunity.

> In the real world, a "co-founder" is someone who founds the company.

In the real world too, not all founders are there at the exact moment the company is incorporated. If I incorporate a company, 6 months later have little accomplished, then realize I need skilled people to come all-in for equity -- they're certainly going to be co-founders.

In this case we lack the full story, but this company likely would still be considered very early. They're raising a seed round and asking someone who would be critical for growth to risk everything only for equity (presumably). If he wasn't critical, the entrepreneur would likely wait until after funding and hire a founding engineer.

Normally vesting is over a much longer period, e.g. four years, so he is insanely trusting of you. For example, what if you worked for six months got the equity then just left? Then he might find it difficult to raise funding.

At such an early stage, 25% and no salary would usually be considered a co-founder. Of course, there's no legal definition so in actuality it's whatever you agree to. This situation is very common, where a founder didn't really derisk the company as much as he perceived.

For being a co-founder, the question becomes to what degree can you accelerate the business. Is funding holding him back and your background/involvement will encourage investment? Is the product what's holding him back, and you will 10x its dev speed? You'd make a logical pitch that with you as co-founder, the company will grow way faster than without. Additionally, I'd do some online equity calculators and present a logical case for a given percentage.

I'm also not sure what the strategy is with the seed range. You want to have enough to survive hopefully 2+ years AND propel the business forward. The low end implies maybe you'll pay yourselves minimally. However, you may need another raise very soon since it doesn't help blow up the company -- you'll have to hire/rent office space/marketing/etc.

Setup/config is a necessary evil on any OS. It's part of the game that you must learn to be a productive coder.

If the setup portion is a huge issue, students can use a pre-configured web solution such as replit. There, you're guaranteed to become dependent on an opaque third party platform (all just to save some minor setup hassle).

I've taught coding to hundreds of beginning students, with about 50/50 Windows/Mac (some Linux). From my experience, Windows is not worse for beginners, nor is the setup process more laborious.

For example with Python, the author can standardize by having all students run the Anaconda Python installer alongside Visual Studio Code. This left us with virtually zero install/setup issues. Alternatively, using wsl2 you could set up any language very similarly to Mac/Linux.

In any regard, setup is typically a one-time cost. Once configured, students typically have no additional problems running code. That said, reducing getting-started friction is always welcome, and the author came up with a nice solution for personal computers. (Running `powershell -executionpolicy bypass` may not work on corporate laptops.)

Apparently this is true of nearly all academic labs, e.g. https://twitter.com/Yuchenj_UW/status/1789319531430912042.

Box plots make distributions easier to reason about by oversimplifying them. In a similar way, the mean can be very misleading (but we likely won't forbid its use!).

IMO a good takeaway might be to always use a plot that fairly represents the underlying distribution.

The bigger anything gets, the more haters it gains. This is a guarantee because it will definitionally use more resources, involve more people, and take away attention from other things.

Many of the best engineers have a strong interest in the literature, even if things are not completely understood theoretically. Good engineers provide strong justification for tuning hyperparameters based both on intuition and theory. The same goes with suggesting architectures -- the best people don't guess, but they can debate the theoretical purpose of each element and how tuning it may affect the outcome. If something isn't completely understood, they may code small experiments and/or work with the math to better understand its potential impact.

Beyond this, optimizing models requires a strong understanding of the math behind them. This provides crucial insights, for example that the attention key bias does not affect the attention weights. In industry, engineers might read a paper about a new activation function. They will probably wonder whether there is (theoretical) justification for how it might affect training time or be computationally efficient in the company's architecture. A theorist would be a great fit here.

Some higher-level research could have some commercial application. For example, there are a few papers showing near-LLM performance (for certain things) attained by only searching datasets.

Ultimately, you will be successful if you can find where your skillset overlaps with what the company needs. Sometimes there is a research division which needs exactly what you research. Other times, you may need to work on things that aren't an exact fit but still utilize your skills. But as of right now, it seems that a successful theorist with coding ability would be in demand (e.g. https://www.anthropic.com/research).

You're lucky, my business partner and I put a cap table in writing where he wrote I had the most shares of anyone. Then I found out 9 months later he took nearly all of the shares for himself behind my back.

Arxiv (2023 paper w/ similar title&authors): https://arxiv.org/abs/2212.13881

Summary by co-author (of the above): https://twitter.com/dbeagleholeCS/status/1627819164906975232

My first language was QBASIC. I had fun making a lot of games. But then I ran into things I couldn't do (with low-level hardware), were too slow (with graphics), etc. This motivated me to learn C and assembly. I imagine it will be similar with Scratch!

Building a prototype of this would be very straightforward. The difficulty would be ensuring high-quality results.

1. You'd only have to know basic web dev. Enough to get input from the user, send it to an API (e.g. https://openai.com/blog/openai-api), and display results. There are many tutorials online showing how to do this.

2. You'd then design a prompt ("prompt engineering") that tells the GPT it's read the book, asking it to converse in a friendly manner as an expert literary critic, etc. Place this prompt and the prior conversation before the user's new question. Then send it to the API. (It's likely the GPT at least knows something about most popular books.)

3. You can now try to improve the results. A popular method: If you have the full text for the book, you can do a text search for the user's query. Then just include relevant parts of the book in the prompt.