< Back to the archive

Like what you see? Subscribe here and get it every week in your inbox!

Issue #214 - April 16, 2023

Here are the top threads of the week, happy reading!

Top comment by darthcloud

I've build BlueRetro [1] an universal Bluetooth controller adapter for nearly all pre-USB gaming console.

I made a gross income of around 3K a month last year out of Royalties on the soft for each device sold. It's Apache 2.0 software so people can do whatever they want.

I started making money when I decided to list on the GitHub README the list of manufacturers/makers that where sponsoring the project. (Only one person at that time) Soon after the others offered to give royalty as well.

I even got a Chinese company, notorious for selling "clone" of OSHW projects, to support the SW development as well via GitHub sponsor.

I've been working on it for the last four years. I entertained the idea to make and sell the hardware myself. But in the end I learned that's it's not something I'm interested to get into. What I really like is working on the software.

It naturally pivoted into a more community driven project where multiple makers are selling various variations of the HW.

I wrote a retrospective last years [2].

[1] https://github.com/darthcloud/BlueRetro

[2] https://github.com/darthcloud/BlueRetro/discussions/289

Top comment by photochemsyn

Herb Gross's ultra-classic old-school chalkboard delivery of "Calculus of Complex Variables, Differential Equations, and Linear Algebra" should not be missed:

https://youtu.be/BOx8LRyr8mU

It turns out he also produced a complete series on the precursor material, "Single Variable Calculus" as well, which I only just now discovered:

https://ocw.mit.edu/courses/res-18-006-calculus-revisited-si...

This professor has a great delivery and a ton of enthusiasm for the subject material, (but you can't just watch it, to absorb it you have to take notes, maybe recreate the examples in Python or something).

Top comment by moosedev

Not just you. I would guess, optimistically, I’m averaging a 5% conversion rate of applications to recruiter chats/“stage 0 interviews”. That is, ~95% of applications yield silence or a generic email rejection without any human contact.

~15 YOE, FAANG experience, usually only applying to roles I feel I’m at least a halfway good fit for (i.e. not a complete scattergun approach).

I’m (financially) fine for now, which is very fortunate. I wasn’t even laid off - I quit voluntarily and took a sabbatical while the good times were rollin’. But since I started looking seriously again, it’s been hard to shake the sense of time disappearing with nothing to show for it. I’m better at Leetcode (ugh) than I’ve ever been, but so is everyone else, and with the slow drip of actual interviews, I only get to demonstrate it once or twice a month :)

ETA: A few of the recruiters I have talked with have mentioned that they’re getting hundreds of applications within hours of a posting going live. So there is likely a “lost in volume” effect as another commenter mentioned. In fact, for some of the roles where I thought I was a great fit but got a generic rejection without a recruiter call, I’ve had some eventual success simply reapplying for the same role, at least when the recruiting platform allows it (some don’t). For reasons of culture and upbringing, it took me a while to get comfortable not taking that initial, faceless “no” for an answer, but it has worked at least twice so far.

Top comment by nindalf

The answer is obvious - most people don’t want to do this. They want to text or call a friend, look up the weather or sports, stream some entertainment or read the news. They’re experts in their own domains and don’t have the time or inclination to learn programming.

There’s some heavy condescension in this thread, claiming that not programming “keeps people stupid”. As if creating can only happen in an IDE and not in ProCreate or iMovie. It would help if programmers like us eased up on everyone else. Not everyone needs to be exactly like us.

Also, you have to be living under a literal rock if you’ve missed the potential impact of LLMs, specifically chatGPT plugins. Once that’s generally available, everyone can tell the LLM what they want in their native language and have it done. Everyone can build custom recipes that combine multiple apps and APIs in novel ways to get stuff done. That’s a revolution waiting to happen - where everyone will be limited only by their imagination, not by their knowledge of programming. And it’ll happen without having to show every smartphone user a BASIC terminal.

Top comment by Garcia98

I've seen this question asked repeatedly in many LLaMa threads, currently the best models that are truly open are the released models from the Flan family by Google, which includes Flan-T5[0] and Flan-UL2[1]. According to its paper, Flan-UL2 performs slightly better than Flan-T5-XXL.

These models perform slightly better than GPT-3 under some tasks[2], but they're still far from achieving the results from GPT-3.5 and GPT-4. This becomes evident when you try to use them in the real world; they're not "good enough" for general use cases, unlike ChatGPT models. However, if you can restrict your use case to one particular domain, you can achieve pretty good results by further fine-tuning these models.

[0]: https://huggingface.co/google/flan-t5-xxl

[1]: https://huggingface.co/google/flan-ul2

[2]: https://paperswithcode.com/sota/multi-task-language-understa...

Top comment by spiffytech

There's an old game from the 90s I loved : The Incredible Machine.

It's all about devising elaborate Rube Goldberg machines to solve puzzles. The cat knocks a ball into the goldfish bowl, which tips over and fills a bucket that operates a pulley that flips a switch that opens a door. That kind of thing.

I remember it being very freeform and exploratory, teaching cause and effect and physics and thinking through complex interaction chains.

I was probably around 7 when I played it. Very age appropriate.

It's available on a web-based DOS emulator!

https://www.myabandonware.com/game/the-incredible-machine-1m...

Probably downloadable somewhere too.

Top comment by emehex

I found myself interviewing with a hedge fund last year because they liked some of my open source stuff (and approach to API design)...

I had about six conversations with a bunch of the team and a conversation with the CEO to sell me on the position. The last step was a seventh “conversation” with the Chief Data Officer to figure out if this was something I really wanted.

The CDO had different ideas… he joined the Zoom, started sharing his screen (with no introductions), opened a Google Doc and told me we were going “write some code together”.

I was pretty confused and flustered but tried to roll with it. He first asked me to “reverse a string”. OK. My lil Python snippet was fine but he didn’t like that it wasn’t “the most memory efficient way to do it”. He then asked me to write a function to approximate “e”. I just ended the interview right there.

Top comment by claudiulodro

My side project is mainly just complaining about working for "the man" while not doing much about it. It's a crowded space, but it definitely makes less than $2k/mo!

Top comment by negative_zero

To me as a user, what's in decline is UI, UX and trust. Lots of software these days is just outright hostile in very "in your factface ways" and subtle ways.

Even software I pay for seems to try to screw me at every turn. "You paid for me? Have some ads! I'll scan everything you do and sell it off to the highest bidder! No you can't copy and send this image to a friend, but here's a link to try and lure them into the walled garden so I can data rape them too. Stay in my eco system it's great. You love it here! Click click click keeping clicking. Moar data to sell! Give moar! Data breach! I'm soz (but no compensation for you."

Oh and when you live off grid with little power you learn to hate electron apps. They are such power wasters. Three things cause my inverter to spin up its cooling fans: games, windows updates and electron apps. (And of course if you have "AC" plugged in most apps seem to think its a damn free for all for power. So its more power efficient to charge powered down and only run using battery if possible).

Top comment by lmeyerov

End users don't care who in your dependency stack is enabling NLP, but that your tool can do things the non-NLP tools cannot

Likewise, that is key signaling for all stakeholders on what is to come. If using AI is part of the core identity & value prop, great. If not, great too, hopefully there is some other wow. Whether users, staff, investors, same deal.

Ex: We do some custom graph neural network & custom auto feature engineering to help ppl look at & analyze their event data more easily & intelligently. OTOH, our LLM work currently uses OpenAI/Azure, and we are deferring self-hosting and fine-tuning for when more useful, as we rather focus more now on bits like vector databases and UIs. In both cases, our users care more that they can get better results and a journey committed to doing more, vs precisely who is doing what & when.