seattleeng

So they’ll just die when Google has a seamless AI-first mobile experience where consumers just ask their Pixel to do X and it happens. Disruption comes for everyone, hardware isnt a moat

These numbers are certainly wrong lol

It’s more like conditioning the posterior of a response on “Ok, so…” lets the model enter a better latent space for answering logically vs just spitting out a random token.

I think you're almost right

I suspect its more a structural issue

e.g. because of Big VC (a16z, other firms with dozens or hundreds of investing staff), VCs don't stick around firms long enough for real returns (cash distributed) to matter. If you're at a place and your stuff is marked up 10x after 4 years, you just hop to become a GP and try to ride the next markup wave. Even if these people exit VC after 10 years that is still 10 years of deals that all flopped.

This might not even be the individual VC's fault -- there may just be too many venture dollars chasing too few power law returns, so you get a surge of startups circa 2019-2022 that all disappear

literally how openai attracted talent with deepmind as the boogeyman. its a playbook that works

As is usually the case, check the data! A lot of the dataset used has fairly morbid scenarios, so the model is working as expected. All the data was synthetically created with GPT4

Yes, the climate boogeyman shows up as a common anti-tech trope since most people don’t look into the breakdown of emissions. Basically everything in direct consumer control pales in comparison to transit (car/truck/boat/plane).

Speed and scale has a good breakdown: https://speedandscale.com/tracker/

Plus most new AI datacenters are trying to go pure nuclear wherever possible

Im reviewing linear algebra now and would also love to know that book!

I actually think there are a few, only some have achieved critical mass with smaller subnetworks. BlueSky and Threads both seem to have attracted non-tech networks although Threads downranked political content, so BlueSky seems like the natural next mass appeal product.

Others:

Mastodon: many tech subnetworks (security, data)

Nostr and Warpcast: Crypto

This is exactly how Ive seen things occur. It makes a lot of sense, given some tools are great new additions (coding especially), but others fall flat (IMO, search)

You can learn things top-down or bottoms-up. I can read & understand most reverse engineering posts like this because I have a strong "bottoms-up" foundation with an EE degree and worked with microcontrollers. But when I read posts by hobbyist mechanical engineers about some 3D printed piston that uses ball bearings I have to approach it in a top-down "recreate what they did and go deep any time I'm lost" manner.

I think you’re exactly right, it’s a show for insiders to know what these 4 people think so the next time they directly or indirectly encounter them they are known quantities. Its masked as news & informative media but its principally brand marketing for these 4 people and their funds & companies.

Cut 80% of jobs and 80% of revenue

That is not what shareholders want, hence why no one is copying it

preserving the vector art after transforming is really cool, anyone know the relevant papers? or was this original research done by Adobe?

Why is working locally important?

A new programming model for distributed computing is desperately needed. Something between a full operational system a la Temporal, but without the extreme operational overhead + a sane cooperative runtime like Golang.

I think we're probably too early to build this today. Ray is used at my current job for scaling subroutines in our distributed job system. It's the closest I've seen.

A credit based model with minimum add amounts like that does sound like a great alternative if people prefer usage based pricing. From what Ive seen usage pricing actually drives a lot of anxiety even if the end cost is lower

I get your point but its $3/mo or $18/yr. If a diagram isnt worth 1.80 to you then you probably wouldnt have bought it even with pay as you go

Its cool but this feels really late. I just use Ideogram now. The social features are nice

OpenTracing doesn’t actually specify when a span is sampled. IIRC Lightstep sends all spans to a service that makes sampling decisions based on moving latency percentiles, while Jaeger works as you described, sampling in the client. Its implementation specific.

what is `fd` in this context?

I agree with what you've written, and the phrasing I used was unclear. I was mainly making a note about joining a mid to late stage pre-IPO startup (series B or later) vs joining a large, low growth public company. You can't easily invest in most pre-IPO companies today (though with secondary markets growing in popularity this might change in 5 years). In the case of choosing between two large public companies, sure, take the cash & liquidate your positions as soon as you vest so you can stick them in SPY or vanguard.

I agree that there are tiers of companies when it comes to compensation, and this site tends to skew towards recording datas for higher tiers (but many of the companies this site provides levels for do NOT pay as much as FAANG - as more salary data is added this will become clearer). And this can certainly be frustrating/depressing when comparing with individual compensation. It should be noted that the pay differential between top paying companies and those below comes from two factors:

- Geography. US based companies in the Bay Area will almost always pay more than companies anywhere else in the world. Many companies have different compensation bands for different regions of the world, even within the US.

- Equity. The base salary for an entry level developer position at a top company will pay somewhere in the range of 110-130k. I've seen many entry level dev jobs at startups in the bay area paying in the 80-110k range (I can't speak to hard data that supports this though, because open salary information is hard to come across!). So, the salary differential when comparing upper/lower bands between top companies and median companies exists but isn't outlandish (between 20-40% more). The difference is, entry level devs at Google & FB will also get a 50k/yr equity grant. Based on my experiences, this is an order of magnitude more than the median company (where lottery tickets or 1-5k/yr grants are common).

So my personal advice for optimizing compensation would be:

1) Move to the US (Bay Area/NYC/Seattle) or work for a US (Bay Area/NYC/Seattle) company remotely or at a satellite office in another city. Obviously, everyone has personal restrictions so this may not possible.

2) Work for a company who you believe will have equity growth. There is a wide spectrum here between 5 person startup lottery tickets and established behemoth that have 0.5% YoY growth stocks. A good recent example of this is Square, which gave out equity grants that were something like 50% lower in cash value than the equivalent role's offer from a FAANG company (this is based off of personal anecdata). However, Square's stock exploded over the past year and that equity today outcompetes many of the equivalent FAANG-level offers. Of course, the opposite could have also been the case -- I've heard stories of underwater options being granted pre-IPO by Square. Sure, a few years later they're worth a lot, but at the time, employees weren't happy.

The world is wider than FAANG and tiny startups, and each company has its own set of hiring criteria (e.g. both Twitter and Square started as Rails shops so your Ruby experience would be more valuable to them than Google or FB). You can't predict the market, but if the choice is between 20 year old Company A that gives you 3k/yr in equity or a recently IPO'd Company B that gives you 1k/yr in equity, I would on average take the gamble with Company B (in practice taking into account team strength, product vision/market fit, & company direction after interviewing).

Senior staff is something like <10% of all engineers (probably closer to 5%), the technical equivalent to a managerial “Director” title. Most engineers top out 2 promotions below.

Huge fan of this site, for two reasons:

- It lets “outsiders” to the tech scene understand salary and levels at big companies. If you went to MIT you probably already have frat brothers/sorority sisters or an alumni network that consists of senior engineers at these big companies that know the promotion and compensation schemes well. But if you were an equally smart student who went to a non target school, many of these organizational “open secrets” are hidden and must be earned through work experience, which costs time and career opportunity.

- Even if you do have some access to a network, these schemes evolve over time. Salary especially (levels not so much). Having up to date data is huge for assessing options during a job search or even planning for one.

The site isn’t perfect— a lot of the leveling data is subjective AFAICT and not based on cross company moves, and the comp info seems a bit skewed for some companies and more senior roles. But its a huge step in the right direction towards empowering employees.

This is a great post. Im curious if these techniques are widely used in hedge funds/prop shops today. I was an equity options trader many moons ago, and the Chicago prop shop I traded at generally had every trader using a simple vol arb approach. Apart from real time vol calculations per strike in the trading tools using black scholes and historical vol for the equity, nothing more complex was used to execute trades (eg no stochastic vol or jump diffusion). The concept of smiles was understood, so people would generally trade at the money options exclusively. Positions were delta hedged over night, generally at market close. Deep out of the money options were used essentially for gambles. Vega was the main source of edge, and as you can imagine, most traders exited positions once their options drifted in or out of the money since our vol model didn't really account for it.

Conceptually, I think you're right -- to serve gRPC requests only requires http2 as a transport and more shouldn't be necessary (edit: I stand corrected, seems like not all HTTP2 features are supported by all modern browsers). In practice, I believe the official gRPC web setup it serves as a translation layer and provides some monitoring bells and whistles. E.g. the official gRPC-Web implementation allows for a base-64 encoded wire format in addition to the typical protobuf binary format, so without a proxy layer, you'd need to define middleware server-side that translates the base64 encoding to the protobuf format.

I think the full set of differences can be found here: https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-WEB.md

Did you go to Caltech by any chance?

I think you'd be surprised. I went to Caltech (graduated 2014), which is a fairly well known university for their Electrical Engineering program, and I learned ABEL in my Sophomore/Junior year. My instructor, an admittedly old school hardware engineer, was in love with the language and had it as part of our upper level digital design curriculum for a few labs. FWIW, I think it was super intuitive and a hugely valuable learning tool. I suppose that doesn't mean it isn't "dead" for professional purposes, though.

Venture Deals by Brad Feld is a good one. Mind you, I'm an earlyish employee and not a founder, but the startup I'm at is thankfully transparent and gives us enough financial metrics for me to put a lot of that book to use in contextualizing our previous fundraising rounds.