3pm

You might be missing the point.

> We built Pando with Codex, moving from kick-off to production, in just 10 weeks. The core of Pando was designed by a single engineer, implemented by teams of agents, and deployed by a handful of engineers. Pando began serving production traffic in April of this year, and is one part of a larger system of source control management tools used at OpenAI.

Impressive!?

Craft was premised on the idea that a good writing and composition matters because it would be easier to read and comprehend (sometimes accepted by orgs because it was sold as cheaper).

Now AI can read the code so the craft-leaning people would need to somehow sell it as 'fewer tokens will be spent on a well factored code' or 'well factored code reduces AI hallucinations'?

Now a Slinger (using your terminology) can just ask AI to make the code look 'craftier' - and that is the best case scenario.

In other words, they will not buy what we sell and if they will they can now produce the results for cheaper. Overlooking the devil in the details, as usual.

I think there were always two camps when it came to writing code, a strong bimodal distribution. People that care about the perceived code quality (aka the ‘craft’) and people who treat the code as a means to an end. At least I always felt this tension. And there is always an exception, and a person can shift between two camps during the same day. But would generally lean towards one. I know I do. And one is not better than the other, they could both be very smart and well meaning. And you need both for success in a for-profit business. It’s been discussed ad nauseam but one particular metaphor I like is Jesus Christ vs James Bond. James Bond would win most of the time. Most people tend to think like James Bond, and if they don’t, the organization strongly nudges them in this direction, promotions get decided and it becomes self reinforcing.

Now a new powerful tool shows up and I think it is time for the Jesus Christ camp to throw in the towel.

Requiring backward compatibility doesn't imply the code quality needs to be low. Nothing prevents them from improving their code base except the organization itself - the incentives, promotion driven development etc.

I agree somewhat, but it took me a bit to articulate to myself. It is nuanced.

Testing is what held everything together. They seem to have been taking it seriously and that alone was enough to keep a product of that scale (barely?) afloat. A success!?

But it sounds like every other good practice went out of the window. They are drowning in unimaginable amount of technical debt, ossified over decades. That, and the complaint about testing infra indicates to me that even their tests are a pile of sh*. He wasn't running them locally, so they were integration tests. Meaning that had there been a foundation of unit tests - they wouldn't need that many slow integration tests.

It sounds like they just brute forced all their problems with test farms. I'm almost sure they double down on AI - why make your code base better if AI can figure out how 100 flags can interact with each other.

I wonder how this holds up in the age of AI

https://news.ycombinator.com/item?id=18442941

> The issuing bank keeps 2%

Is this why the best cash back credit cards give 2%?

$700 premium for Duo compared to iPhone Pro Max 18. Exactly the price of iPad mini 256GB.

Duo screen: 7.6" vs iPad mini: 8.3". Duo will support Apple Pencil later. I wonder if many people with iPhone+iPad combination can now replace it with one device.

I guess annual income also hints at annual expenses which should be important for the study. There are people who constantly spend all they make, and probably score low on self assessed happiness. Also income != wage, can be passive.

I think for a study like this, a money-happiness correlation would need a more sophisticated definition of money, to account for peer pressure and spending.

For a given zip code - take p75 of pre-tax income, multiply by 50% - this would be a proxy of how much someone needs to spend per year to be comfortable with local cost of living and their peers. 50% takes taxes and savings out, call it spend/burn. Then someone's wealth could be a ratio of net-worth/burn. The higher the wealthier - people 25 and over essentially not needing to work, and people with < 3 are essentially in indentured servitude, even with high income. Control for age.

> What is the shape of the relationship between money and happiness, and what are its implications?

1. money != income.

2. Zip codes.

3. Age.

4. Social class.

a) 65 year old professor living in Woodside, CA, with a net worth of $250,000K

b) 35 year old HVAC business owner living in Fresno, CA, with a net worth of $2,000,000

First is poor, second is rich, but the study conflates both into the same bucket if they both make, say, $400K/yr

I think that Model M (Unicomp/Lexmark) was a cost cut for Model F. Just recently, a company resurrected true Model F: https://www.modelfkeyboards.com/

I’m hearing that Lexmark/Unicomp keyboards were/are a cost cutting project compared to original Model F. Interestingly there is a company that resurrected true, pre cost cutting, model F: https://www.modelfkeyboards.com

Using Windows Server as a personal OS can be an option: https://www.windowsworkstation.com/win2016-2019/

Vanguard just tracks indexes. Indexes that include all publicly traded auto manufacturers in the world. E.g. Toyota: https://www.morningstar.com/funds/xnas/vtiax/portfolio

Most of Vanguard is a large bureaucracy with two goals: accurately track an underlying index and keep the cost low. They do have some actively managed funds but those are insignificant compared to their passive funds.

Interestingly enough, Vanguard voted against the package originally in 2018, and Black Rock voted for it (according to nytimes https://www.nytimes.com/2024/06/12/business/tesla-elon-musk-...)

    Tesla’s second-largest shareholder, Vanguard, voted against the pay deal in 2018. BlackRock, the third largest, voted for it. Both declined to say how they were voting this time.

I actually think that for VTSAX manages, the fact that they have to vote is sort of a liability for them. My guess is that they would rather not vote at all. They own the whole market anyway, and should only care about tracking the underlying index as accurate as possible with as low fees as possible, its bureaucratic. Now they have to publicly vote in a dubious popularity contest and take sides in a battle they don't care about.

For the record, Vanguard also has significant interest in the success of all Tesla competitors that are part of the index.

Did Vanguard officially stated that they have "significant amount of trust in Musk"? They are not Cathy Woods.

I guess the point here is that a huge percentage (7% just for Vanguard) of votes was not decided by the actual share holders. These shares are passively invested because Tesla is included in indexes like S&P 500. So the decision on the 7% of the votes was made by a few Vanguard execs. I'm not arguing for or against the decision, just stating the fact that it is much easier to convince a few Vanguard managers instead of 7% of the actual share holders.

Does anyone know about BlackRock and State Street votes?

    But Vanguard, which has total assets of about $9 trillion, and other big index fund managers were always likely to be key to the vote. Representatives of Vanguard rivals BlackRock (BLK.N), and State Street (STT.N), declined to comment about their votes on Thursday.

   "Half of it is regurgitating"
   "how you behave as a human being, what happens when I throw in a curve ball."

So it's half hazing, half theatrical performance?

Answering "Design Uber" questions is like playing a weird charade game in a LARP camp. It is very convenient for the interviewer because they set the rules and can change them anytime. They only prepare their question once and apply it it all candidates. What type of signal these are supposed to generate? Are these some sort of behavioral/hazing rituals? Why all the smoke and mirrors? If these are testing System Design skills, why not ask a specific System Design question? Why not ask about tradeoffs between single vs multi leader vs leaderless? Why not ask specifically about document vs relational.

Better yet, why not look at the person experience and ask them how would they scale a system they actually worked on before? And what were the tradeoffs.

This can be helpful for LARPing though: https://systemdesignfightclub.com/

Obligatory MySQL Big Deletes: https://mysql.rjweb.org/doc.php/deletebig

Good paper on UTF-8 validation performance: https://arxiv.org/pdf/2010.03090

    The relatively simple algorithm (lookup) can be several times faster than conventional algorithms at a common task using nothing more than the instructions available on commodity processors. It requires fewer than an instruction per input byte in the worst case.