Get live statistics and analysis of Bigi Lui's profile on X / Twitter

Job market (ZIRP changes, etc.) + GPT = perfect storm I started a local coding class for students here in Elk Grove in summer of 2023. It was something I had wanted to do since a couple years before that and I planned for it in early 2023. I had wanted to build something like the bootcamps, for the younger demographic (high school to college), to better prep them for the job market. ChatGPT was first released to public in late 2022. As we entered that summer it was already apparent that AI was going to have huge effects in coding. While our class was ongoing that summer, the elephant in the room was "why are we learning coding" (this was a class of high school students who are beginners and just learning HTML) as ChatGPT was already popular in school (one of the few products in history that popularized in school work before almost anything else, because of its nature, so all the students already knew about it). Tbh to this day I still don't know what the landscape will look like for students going into software eng now, whether they would focus on traditional CS fundamentals or hands on AI coding. I have a hunch that vibe coding would be a good first step in learning software dev for interested students. But Idk. I haven't restarted the coding class effort after summer 2023 because I myself don't have clarity on this yet.

It's pretty hard to be a world expert at one thing. I'm probably top 3-5% in the world as a ship-things software eng (solopreneur/small startup type), but I'm probably not top 0.1-0.5%. (and I'm probably below top 5-10% as a big corp software eng) But I'm probably a top 0.01% of solopreneur software engineer who also played a ton of starcraft and capcom vs snk and who also plays a guitar and sing in a band and who also is into real estate, suburban living, and being a parent and who is also into crypto and AI. If there was a name for that intersection of things as one subject matter, I'd be the world leading expert.

Google Glass, Snapchat Spectacles, Oculus/Meta Quest, Apple Vision Pro, Meta Ray-Ban Over the past 13 years or so Maybe we (as an industry) should just stop making things you put on your head/on your eyes that has AR/VR tech? Maybe it just isn’t a thing people want outside of movies, and even if they want them, after actually experiencing it, it isn't all that? There's something to be thought about why a phone was a successful form factor and eyes-thing isn't, something about having control of when you want to access something vs when something is actively pushed to you

Several possible outcomes of this: 1. Like Uber/Doordash that made rides/delivery cheap for the mass when VC money was flush, prices will be raised so much (that I never order delivery anymore and only ride Uber when absolutely needed), similarly that might mean Cursor/others prices will costs hundreds of $ and make you think twice before paying 2. Cursor/others cease to exist like some of the other X accounts out there like to speculate, because of this 3. Open weights LLM alternatives + open source software alternatives of AI coding IDE become available and have parity of current Cursor - a. Only cloud-hosted open LLMs are powerful enough to be useful - b. Open LLMs that can run locally on a laptop are powerful enough to be useful Unless outcome #3 wins, and specifically 3b, which frankly isn't easy (herculean effort in open source community to make this happen), a corollary is that: While VC money is flush, maybe we’re currently in a “AI coding capability arbitrage” period where Cursor/AI coding is cheaply available and we’re uniquely able to build products super fast as a solopreneur

X is crazy because sometimes buried in a reply is one of the most useful, easy to understand, high level architecture of netcode for game programming Better than any tutorial you could find online with useless jargon

so not many people have talked about API costs re: nano banana. tldr: it has 10x'd. (I know we're past nano banana being the zeitgeist now, that was so last-month. But I now have 2 weeks of data now to speak on) for those who didn't know, in true google fashion of deprecating stuff unreasonably quickly -- the old gemini 2.0 flash image gen model was deprecated in favor of gemini 2.5 image gen (aka nano banana). API access to the old 2.0 model was sunsetted 9/26 -- no way to keep using it at all. I appreciate having nano banana available, because the output quality (for my use case with ProdShot) is *slightly* better, and in some cases it's a big help (texts on products is pretty much solved now) but one big problem: API costs is now nearly 10x what it was vs when I was using 2.0 flash. needless to say for a small saas like ProdShot it has deeply cut into my margins, and I guess this was made particularly worse because I was generously giving away free credits on signup in hopes of driving more adoption so this is actually a good motivator and lights a fire under my ass to increase my pricing and lower free credits as much as possible (down to just being able to create the first set of photos for an account probably), like levelsio always said, get rid of the free plan essentially. the other update that's less drastic is that the rate limits also got lowered with the release of 2.5, which is fine for me at my scale so far, but annoying since 2.0 (which had a higher limit and was fine for my use case) is completely deprecated and no longer available

I now have two boilerplate/template repos to start web projects with that I am extremely happy with Before you say "just vibe code with AI now": You should always let AI start from a codebase, you'll be much better off Anyway, the two templates are basically 1) SSG React frontend, Fastify API backend 2) HTMX frontend, combined server as backend (server also by Fastify) I initially started (1) last year and it's what I built ProdShot off from, it started as a simple React SPA (fully dynamically hydrated client), as I begin building/launching I realized I really do need SSG for better SEO. Not having static html was killing its SEO performance. Took some time to convert it and now I finally ported it back over to the template. This is a versatile template suitable for many kinds of web apps/saas, pretty much anything that has considerable UX interactivity where you want a React frontend. There are certain types of sites/projects where SSG isn't suitable for, which is when you have tons of dynamic urls/slugs (think ecommerce or blog or listings sites with lots of pages, url routes driven by DB data). That's exactly what (2) is for. It's like traditional PHP apps in a nodejs stack. It doesn't have a frontend/backend server separation, it's just one server, it serves HTML output. It could as well even just use jquery for ajax interactivity but I chose to go with HTMX given its good rep, and I'm starting to build with it. Both templates are set up with all the niceties for SEO and marketing, including favicons of all sizes, sitemaps, meta tags, social share tags and preview images, as well as tailwindcss, DB with migrations, etc. NO SSR ON EITHER STACK! I have a strong aversion to SSR. It's slow and bloated and very backwards. It's my belief that very very few types of sites (if any!) actually make sense to use SSR, it's a very small overlap of various characteristics of a site. The popular saying in TPOT is "SSR is a psyop" which I don't disagree.

Maybe cutting my nerd cred but I've never actually had a proper LAN party I play plenty of PC games solo Played plenty games online, both online games, and offline games while chatting on ICQ etc Had friends over or been over at their house to play console games together Even had set up high school computer lab to play GTA2 like this (closest thing) But never actually attended any proper LAN party at someone's house with a row of PCs like this

If you ever made a windfall, assuming that's multimillions (to even be considered a windfall), you may want to take at least $1m (preferably $2-4m if your total is high enough) as "don't touch it" money, only put in the stock market index funds like the S&P to let it grow, and do angel investments and start businesses etc only with the rest. Keep the safety net. On the flip side though, you do get unreal motivation when you're absolutely out of money, so if that's what you want... NFA!

With ColdPlay being in the zeitgeist I figure it's a good time to try my AI Karaoke Maker on one of my favorite songs of theirs to sing actually Had to hand-fix a few words after the AI transcribed the lyrics but results pretty good! Note this is an Asian style KTV video meaning two things: 1. lyrics are printed on original music video at the bottom, instead of a no-video text-only karaoke like American karaokes 2. left/right audio channels: left is instrumental only, right is original voice, so Asian karaoke systems playing this can switch between the two. You can use a left/right switcher on your computer/phone to get that effect.

Google Glass, Snapchat Spectacles, Oculus/Meta Quest, Apple Vision Pro, Meta Ray-Ban Over the past 13 years or so Maybe we (as an industry) should just stop making things you put on your head/on your eyes that has AR/VR tech? Maybe it just isn’t a thing people want outside of movies, and even if they want them, after actually experiencing it, it isn't all that? There's something to be thought about why a phone was a successful form factor and eyes-thing isn't, something about having control of when you want to access something vs when something is actively pushed to you

It's pretty hard to be a world expert at one thing. I'm probably top 3-5% in the world as a ship-things software eng (solopreneur/small startup type), but I'm probably not top 0.1-0.5%. (and I'm probably below top 5-10% as a big corp software eng) But I'm probably a top 0.01% of solopreneur software engineer who also played a ton of starcraft and capcom vs snk and who also plays a guitar and sing in a band and who also is into real estate, suburban living, and being a parent and who is also into crypto and AI. If there was a name for that intersection of things as one subject matter, I'd be the world leading expert.

Job market (ZIRP changes, etc.) + GPT = perfect storm I started a local coding class for students here in Elk Grove in summer of 2023. It was something I had wanted to do since a couple years before that and I planned for it in early 2023. I had wanted to build something like the bootcamps, for the younger demographic (high school to college), to better prep them for the job market. ChatGPT was first released to public in late 2022. As we entered that summer it was already apparent that AI was going to have huge effects in coding. While our class was ongoing that summer, the elephant in the room was "why are we learning coding" (this was a class of high school students who are beginners and just learning HTML) as ChatGPT was already popular in school (one of the few products in history that popularized in school work before almost anything else, because of its nature, so all the students already knew about it). Tbh to this day I still don't know what the landscape will look like for students going into software eng now, whether they would focus on traditional CS fundamentals or hands on AI coding. I have a hunch that vibe coding would be a good first step in learning software dev for interested students. But Idk. I haven't restarted the coding class effort after summer 2023 because I myself don't have clarity on this yet.

X is crazy because sometimes buried in a reply is one of the most useful, easy to understand, high level architecture of netcode for game programming Better than any tutorial you could find online with useless jargon

With ColdPlay being in the zeitgeist I figure it's a good time to try my AI Karaoke Maker on one of my favorite songs of theirs to sing actually Had to hand-fix a few words after the AI transcribed the lyrics but results pretty good! Note this is an Asian style KTV video meaning two things: 1. lyrics are printed on original music video at the bottom, instead of a no-video text-only karaoke like American karaokes 2. left/right audio channels: left is instrumental only, right is original voice, so Asian karaoke systems playing this can switch between the two. You can use a left/right switcher on your computer/phone to get that effect.

Several possible outcomes of this: 1. Like Uber/Doordash that made rides/delivery cheap for the mass when VC money was flush, prices will be raised so much (that I never order delivery anymore and only ride Uber when absolutely needed), similarly that might mean Cursor/others prices will costs hundreds of $ and make you think twice before paying 2. Cursor/others cease to exist like some of the other X accounts out there like to speculate, because of this 3. Open weights LLM alternatives + open source software alternatives of AI coding IDE become available and have parity of current Cursor - a. Only cloud-hosted open LLMs are powerful enough to be useful - b. Open LLMs that can run locally on a laptop are powerful enough to be useful Unless outcome #3 wins, and specifically 3b, which frankly isn't easy (herculean effort in open source community to make this happen), a corollary is that: While VC money is flush, maybe we’re currently in a “AI coding capability arbitrage” period where Cursor/AI coding is cheaply available and we’re uniquely able to build products super fast as a solopreneur

so not many people have talked about API costs re: nano banana. tldr: it has 10x'd. (I know we're past nano banana being the zeitgeist now, that was so last-month. But I now have 2 weeks of data now to speak on) for those who didn't know, in true google fashion of deprecating stuff unreasonably quickly -- the old gemini 2.0 flash image gen model was deprecated in favor of gemini 2.5 image gen (aka nano banana). API access to the old 2.0 model was sunsetted 9/26 -- no way to keep using it at all. I appreciate having nano banana available, because the output quality (for my use case with ProdShot) is *slightly* better, and in some cases it's a big help (texts on products is pretty much solved now) but one big problem: API costs is now nearly 10x what it was vs when I was using 2.0 flash. needless to say for a small saas like ProdShot it has deeply cut into my margins, and I guess this was made particularly worse because I was generously giving away free credits on signup in hopes of driving more adoption so this is actually a good motivator and lights a fire under my ass to increase my pricing and lower free credits as much as possible (down to just being able to create the first set of photos for an account probably), like levelsio always said, get rid of the free plan essentially. the other update that's less drastic is that the rate limits also got lowered with the release of 2.5, which is fine for me at my scale so far, but annoying since 2.0 (which had a higher limit and was fine for my use case) is completely deprecated and no longer available

I now have two boilerplate/template repos to start web projects with that I am extremely happy with Before you say "just vibe code with AI now": You should always let AI start from a codebase, you'll be much better off Anyway, the two templates are basically 1) SSG React frontend, Fastify API backend 2) HTMX frontend, combined server as backend (server also by Fastify) I initially started (1) last year and it's what I built ProdShot off from, it started as a simple React SPA (fully dynamically hydrated client), as I begin building/launching I realized I really do need SSG for better SEO. Not having static html was killing its SEO performance. Took some time to convert it and now I finally ported it back over to the template. This is a versatile template suitable for many kinds of web apps/saas, pretty much anything that has considerable UX interactivity where you want a React frontend. There are certain types of sites/projects where SSG isn't suitable for, which is when you have tons of dynamic urls/slugs (think ecommerce or blog or listings sites with lots of pages, url routes driven by DB data). That's exactly what (2) is for. It's like traditional PHP apps in a nodejs stack. It doesn't have a frontend/backend server separation, it's just one server, it serves HTML output. It could as well even just use jquery for ajax interactivity but I chose to go with HTMX given its good rep, and I'm starting to build with it. Both templates are set up with all the niceties for SEO and marketing, including favicons of all sizes, sitemaps, meta tags, social share tags and preview images, as well as tailwindcss, DB with migrations, etc. NO SSR ON EITHER STACK! I have a strong aversion to SSR. It's slow and bloated and very backwards. It's my belief that very very few types of sites (if any!) actually make sense to use SSR, it's a very small overlap of various characteristics of a site. The popular saying in TPOT is "SSR is a psyop" which I don't disagree.

Maybe cutting my nerd cred but I've never actually had a proper LAN party I play plenty of PC games solo Played plenty games online, both online games, and offline games while chatting on ICQ etc Had friends over or been over at their house to play console games together Even had set up high school computer lab to play GTA2 like this (closest thing) But never actually attended any proper LAN party at someone's house with a row of PCs like this

If you ever made a windfall, assuming that's multimillions (to even be considered a windfall), you may want to take at least $1m (preferably $2-4m if your total is high enough) as "don't touch it" money, only put in the stock market index funds like the S&P to let it grow, and do angel investments and start businesses etc only with the rest. Keep the safety net. On the flip side though, you do get unreal motivation when you're absolutely out of money, so if that's what you want... NFA!