dtnewman

- Amazon made AWS - Microsoft went from Operating Systems to a cloud company - Even Google has Waymo, which might develop into a large company by itself - Apple was founded in 1976. They really weren't the powerhouse that they are until the iPhone came out. - Netflix went from mailing DVDs to streaming

Of course most pivots don't work and there's selection bias here. BUT, it's not unheard of for business models to change. And it certainly helps that Zuck is the original founder and has controlling shares.

> the fact that their empires are now all they'll ever be

Eh, Mark Zuckerberg is 40. Facebook is planting seeds in some pretty ambitious places (VR/AR + AI). To put that into perspective, Elon Musk is 53 now, but he was ~40 when the Falcon 9 first launched for SpaceX and the Model S was released at Tesla. In June 2012, when the S was released, Tesla was worth about 0.7% of what it is now. Elon Musk was certainly rich, but no where close to the wealthiest folks at the time. Similarly, at 40 years old (21 years ago) Jeff Bezos was worth about 100th of what he is now. Rich, but it wasn't clear that Amazon would ever come close to, say, Walmart, in terms of market cap.

Mark Zuckerberg's empire still has plenty of time to grow.

The article is talking about new grads generally, but I think there's an issue with AI that isn't talked about enough. It's not that it's taking away jobs [1], it's that it is taking away skills.

Even if you are the biggest critic of AI, it's hard to deny that the frontier models are quite good at the sort of stuff that you learn in school. Write a binary tree in C? Check. Implement radix sort in Python? check. An A* implementation? check.

Once upon a time, I had to struggle through these. My code wouldn't run properly because I forgot to release a variable from memory or I was off-by-one on a recursive algorithm. But the struggling is what ultimately helped me actually learn the material [2]. If I could just type out "build a hash table in C" and then shuffle a few things around to make it look like my own, I'd have never really understood the underlying work.

At the same time, LLMs are often useful, but still fail quite frequently in real world work. I'm not trusting cursor to do a database migration in production unless I myself understand and check each line of code that it writes.

Now, as a hiring manager, what am I supposed to do with new grads?

[1] which I think it might be to some extent in some companies, by making existing engineers more productive, but that's a different point

[2] to the inevitable responses that say "well I actually learn things better now because the LLM explains it to me", that's great, but what's relevant here is that a large chunk of people learn by struggling

AWS lambda + API gateway. I use serverless to deploy to these.

If traffic is high, this could get pricy. For example, i have a site up getting about 20,000k visitors a day, hitting it many times each and it's probably costing me $400/month whereas a $50 server would probably do the trick (it's a temporary thing that i'm migrating away soon). But I have a bunch of smaller projects with a few visitors a day and the cost rounds to $0.

The nice thing is that it'll basically scale infinitely when you need it to.

Just fyi, this is now implemented

I’m debating changing it. I do pull in env vars to use as default values (e.g. you already have an API key set). But I might transition way from env variables.

1) When you are selecting a command you get a little description at the bottom telling you what it does.

2) this doesn’t run anything. It goes to your clipboard and you have to run it yourself

3) this a good callout… what do u think? I’m thinking maybe ask the models to return a Boolean is_dangerous plus a small explanation and then I can display dangerous commands in red and show the warning when you select one.

ha, no, just a coincidence. Named after someone i know named Zev. But chose it because it's short and not taken on Pypi

different use case. atuin is for past commands, whereas this uses an LLM to give you options for commands.

Thanks! My main issue is that i'm lazy and although i often know approximately what i want i don't want to type a lot of words to describe it exactly. For example writing `zev 'show disk usage'` is somewhat ambiguous. Am i talking about my current folder or the harddrive? My idea was that rather than typing out what I want explicitly, i want to type the minimum amount and then just select the best of available options.

it should run outside of a virtual env and is intended to be installed locally. That said, it currently has too many dependencies (IMO) and i'm working on cutting them down to avoid conflicts.

btw, feel free to open an issue on github :)

creator here. It pulls env variables from a file in your appstorage directory. I need to change this in a future release to make it cleaner, since I don't think i like it intermingling with env variables.

That said, did you run `zev --setup`?

revision history

I built a chrome extension for teachers that grew to 200k users with no marketing budget.

It solved a pressing problem for teachers. But I built it and showed it to teachers and.... crickets. Absolutely no one found it organically by searching the chrome extension store. I emailed every teacher I know and a few installed it just to humor me, but no one used it. But eventually I got it to grow organically. I'm not saying this process is repeatable, but here's what I learned:

1) Finding the first 10 users was painfully hard. I had to beg people to install it. None of them cared.

2) Finding the right users at the right time is important. I went on to teacher forums on facebook and whenever someone posted something relevant to my app, I privately messaged them. This failed most of the time, but occasionally worked.

3) After lots of messages and begging, I clawed my way to maybe 10-20 users. That's when I started to see the first signs of organic growth. People would tell their friend about it. I'd see 1 or 2 people installing it per day who were hearing about it from friends

4) At 300 users, I got my first "influencer". Someone who had access to a email distro list that goes out to a few thousand teachers. They posted something there and my user count doubled in the next few days.

5) At 30k users, I got a viral video that someone posted about it on tiktok. Got 10k install over the next week.

6) Going from 300 to 200k users was easy... once the ball was rolling, I didn't have hunt down users myself... they would tell each other about it.

Going from 0 to 200k took about 18 months.

Lessons learned:

- Dont worry about your first 1000 users or even 100 users. Worry about your first 5. Do whatever it takes to get them. Finding them is a manual process.

- You will be rejected early on. A lot. It might mean your product sucks, but even if your product is great, you need to find the right person, at the right time. On top of that, most people aren't early adopters.

- Here's what probably won't work for finding those early users unless you really know what you are doing in an existing market: google ads, facebook ads, pretty much anything that isn't manual outreach.

Costco is a bit different. If I don't have a Visa, i can pay with cash, personal check, or debit card. Sure, I'd prefer to pay with credit card, but it's not a huge deal for me to use a debit card there.

With Uber, there simply isn't an alternative. If they don't accept Visa then i can't use a credit card (or most debit cards, since those go through the visa network too) and there isn't a reasonable way for me to pay at all.

> A common question is: “how much are students using AI to cheat?” That’s hard to answer, especially as we don’t know the specific educational context where each of Claude’s responses is being used.

I built a popular product that helps teachers with this problem.

Yes, it's "hard to answer", but let's be honest... it's a very very widespread problem. I've talked to hundreds of teachers about this and it's a ubiquitous issue. For many students, it's literally "let me paste the assignment into ChatGPT and see what it spits out, change a few words and submit that".

I think the issue is that it's so tempting to lean on AI. I remember long nights struggling to implement complex data structures in CS classes. I'd work on something for an hour before I'd have an epiphany and figure out what was wrong. But that struggling was ultimately necessary to really learn the concepts. With AI, I can simply copy/paste my code and say "hey, what's wrong with this code?" and it'll often spot it (nevermind the fact that I can just ask ChatGPT "create a b-tree in C" and it'll do it). That's amazing in a sense, but also hurts the learning process.

I’m not sure I follow. Sure, people lie, and make stuff up all the time. If an LLM goes and parrots that, then I would argue that it isn’t hallucinating. Hallucinating would be where it makes something up that is not in its training site nor logically deducible from it.

Really depends on your use case. For low value tasks this is way too expensive. But for context, let’s say a court opinion is an average of 6000 words. Let’s say i want to analyze 10 court opinions and pull some information out that’s relevant to my case. That will run about $1.80 per document or $18 total. I wouldn’t pay that just to edify myself, but i can think of many use cases where it’s still a negligible cost, even if it only does 5% better than the 30x cheaper model.

I do the same thing as you but with owning the root domain, so no need for a +. In other words, I would use spectrum@example.com for this (i have a catchall rule so it forwards to the same place). I’ve never had an email validation issue and this actually makes more things validate, since some websites require you to enter a “business email” and this passes that (I think they basically filter out Gmail and others).

Why is this an example of users abusing them?

Check out exo labs blog. https://blog.exolabs.net/day-2/

The price is fine. Maybe low. The real question is can you find users? There are probably thousands or tens of thousands of products out there that would improve my life that I have never heard of. There are many I have heard of that I’m too lazy or busy to try.

Your user is the type of person who obsessively checks their stripe account. They log in multiple times per day. Most people aren’t like that, but some are. Of those people, a small subset are early adopters and into trying new things. The real problem is how to find those people. Once you do, I’d suggest charging more. Once I’m getting my credit card out, I’m probably not paying attention to the difference between $19 and $19/yr. It either has to be so cheap (<$4) that the price feels negligible and I’ll buy on a whim or just charge more.

The biggest hurdle is between $0 and $0.01, because that’s the difference between pulling out my credit card or not.

Nice! You should check out a free chrome plugin that I wrote for this called revision history. It’s organically grown to 140k users, so the problem obviously resonates (revisionhistory.com).

From my experience, this is not necessarily true. Some companies might have the budget to pay at the 50th percentile in New York, or the 95th percentile somewhere else, and would prefer to do the latter

To be fair, many many people die in car accidents every day. Yes, self driving car crashes are still newsworthy but you are up against a baseline that’s pretty bad.

On the other hand, virtually no one dies from laundry, that I’m aware of. So the reaction to a single accident might be quite different.

Strictly speaking, yes, I’ll concede that this changed the world if you are part of a military group and your foe is Israel or another highly advanced adversary. But no, I don’t think that this changes the world in the way that the article implies.

9/11 changed the world because it started two major wars in the Middle East, resulting in hundreds of thousands of deaths. It set the stage for world politics for the next few decades.

The reality of this attack is that it really won’t affect people’s lives in any meaningful way. Maybe a few people will be wary of using beepers for the next few months but I don’t see many people thinking of giving up their smartphones over this. Or perhaps I’m wrong and this foreshadows many similar supply chain bomb attacks in the future… I don’t know, but neither does Bruce Schneier and it’s too early to say definitively what the effect of this will be.

I’ve had many times where a restaurant is officially booked online but I call them and say “hey, we’re looking for a table of two, we can come in right now and we promise to eat quickly and have the table turned around in 45 minutes”. Determining whether to allow that is very hard for an algorithm to do.

Israel’s pager attacks did not change the world. They merely alerted the world to the possibility of everyday devices being compromised. What is true today was still true yesterday, it just wasn’t top of mind: we have to be very careful with what we allow our adversaries to sell to us.

This is just a guess, but I think it might be more about the government not wanting to put valuations on complex assets. Let’s say I own a network of dry cleaners in Los Angeles. It’s a private business with no public business to compare it against. Cash flow is X, but is the business worth 10 million, 20m? How is the government supposed to determine what it’s worth? Now, let’s imagine your business is Koch Industries. We know it’s worth many billions but there’s a VERY broad range of what it might be. Without taking the stock public, its basically impossible even for top investors to value (investment bankers who value businesses for a living get it wrong all the time), let alone the government.

Even public businesses are not trivial to value for very large shareholders who don’t have the ability to easily sell all shares at once without moving the market quite a bit. But in any case, removing this loophole would just encourage the ultra wealthy to put their money into opaque businesses and then try to “value” them as low as possible. So it’s not so easy to fix.