Gergely Orosz Profile picture
Apr 1 4 tweets 2 min read Twitter logo Read on Twitter
Twitter open sourced its recommendation algorithm and while my timeline is exploding with analysis of it; all I can think of is: what is the point?

Code without the training data is interesting but useless.

Is it a recruitment tool for devs? Promote Scala? Something else?
Or is it meant to move attention away from the surprising (and likely damaging) decision to remove the “legacy” verified check marks?

I wouldn’t mind seeing Twitter as a business/service thrive, and have no idea what this repo achieves. Outside of devs, it speaks to no one.
I am under no illusion that #TechTwitter is a small niche within Twitter. And yes, this step has interest within this community.

But for “mainstream Twitter”, celebs like LeBron tweeting daily is what brings them back. And Twitter is nudging LeBron to stick to Instagram et all:
On how pointless all this is:

I talked with former Twitter engineers who worked on this code and they said that while the open sourced code is in prod, the *majority* of the recommendation prod code is not published.

It’s also not the actual prod code, just a new repo created.

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Gergely Orosz

Gergely Orosz Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @GergelyOrosz

Apr 2
In the UK, salaries need to be advertised for positions and the below listing is making the headlines. It’s the Head of Cybersecurity *for the British Treasury* offering £57K.

As comparison, I made £65K as a developer with 2 years of experience in London: more than 10 years ago. Head of Cyber Security  Emp...
The nice thing about salary transparency is that job adverts like this - offering a third of the going rate for such an important role, demanding significant & hard to get experience - will get the mocking and media attention it deserves - and it does!

jobs.theguardian.com/job/8512191/he…
Ofc they can leave the advert up and either of the 2 will happen:

1. No candidates are hired, and there is no one responsible for cybersecurity for the Treasury.

2. An unqualified candidate is finally hired.

In either case, assume unsatisfactory cybersecurity practices.
Read 4 tweets
Apr 1
What if someone took this diagram, measured the size of the pixels to track back the amount paid out per month, and extrapolated Substack's revenue (10% of this)?

I confess: I did this. Here's the numbers, using pixel measurements from the image below: An estimation of Substack's revenue per month. It's likely a
To prove that data visualization is an art, you probably look at the above chart thinking "well that's not very fast growth." Month on month: sure.

But let me share the exact same data, visualized annually.

From 2021 to 2022, rev grew 67%. 214% the year before!

Same data:
An interesting observation: revenue jumps in every January! I assume this is because of this being the time for many people to sign up for an annual sub? Or who knows. But it's noticeable.

All data might be incorrect to some extent ofc, as it's based on me measuring pixels.
Read 4 tweets
Mar 31
After I wrote my analysis on Lyft yesterday (that I see several signs for concern), I got a surprising amount of messages assuming I must have a stake in eg Uber.

I'm now glad I had an ethics policy since starting to write @Pragmatic_Eng. This policy is still the same: Things that I do not do:  I do not accept payments in any fo
Of course, I will naturally have biases based on my past experiences (where I worked) or financial interests (how I make money). I do the best to acknowledge these, share what these are, and to counter them.

All detailed here: blog.pragmaticengineer.com/ethics-stateme… I have worked in the past at Uber, Microsoft, Skyscanner and
This is my analysis on Lyft: blog.pragmaticengineer.com/lyft-in-troubl…

All that I know about the company suggests it is a great place to work at, as an engineer (the culture, compensation, impact etc).

But the founders stepping down + financials + earnings comments do not signal much good ahead:
Read 4 tweets
Mar 31
DoorDash has a market cap of ~$24B w $6.5B annual revenue in 2022. #1 food delivery app in the US.

Lyft a market cap of ~$3.4B w $4.1B revenue in 2022. #2 ridesharing app in the US.

With Lyft trading under its market cap, there’s now a question that was not realistic before:
Would it make sense for DoorDash to buy Lyft while it’s trading at a discount, with the goal to eventually become the #1 food delivery and ridesharing app in the US?

There are a lot of questionmarks here, mostly because both DoorDash and Lyft are losing money - for now.
But any company with a market cap below their annual revenue can become an interesting acquisition target: esp with a buyer who expects to be able to operate it with more efficiency (leveraging their existing business).

Don’t forget Uber paid $3.1B for a much smaller Careem!
Read 4 tweets
Mar 30
A CTO at a scaleup told me they are evaluating GitHub Copilot, but they got cold feet, when looking at legal implications.

E.g. can they be sued if code Copilot produces is copyrighted? What code has Copilot been trained on?

Are there answers to these concerns already? twitter.com/i/web/status/1…
I know of a lot of startups using Copilot, and people using just for themselves (but eg not getting "approval") from above.

But for any serious company to roll it out, they need answers to these questions.

If you have any pointers, do share!
There’s the argument on how even till now, devs could have copied StackOverflow code.

BUT! A huge difference is you can ask the developer what they did, why they did it (including with eg a jury if it gets to it).

You cannot ask an LLM model this. Which is a major problem.
Read 5 tweets
Mar 28
Let's assume that LLMs and "AI" *do* make white-collar workers much more productive and reduce the barrier to this work. What happens then?

A potential outcome: professions that cannot be automated with sw increase in value more.

E.g. construction, plumbing, gardening etc.
For decades, people took longer time to study to have better earning opportunities: college jobs paying much better.

But what if the trend changes? LLMs and "AI" looks to be a massive productivity gain with knoweldge-based professions. But makes no dent with physical work.
In the replies: "automation will come to construction work as well."

Well, I see none, at least for eg house renovations. Try replacing a cracked tile in a house, or put in a custom-made smallish cabinet: the time, effort and cost are surprisingly high.

Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

:(