Gerald McAlister

Focused on #AI #VR #AR #MR #XR #SpatialComputing | he/him | SWE @Meta | Previously @RealityLabs @amazon @alexa99 | Side projects @RGBSchemes | Opinions are mine

[Gerald McAlister @GEMISIS]

If you’re curious to find me on other platforms, my username varies between GEMISIS and gemisisDev! Currently on all others, include various new ones 😉

[Gerald McAlister @GEMISIS]

Glad to see someone else mention spiking neural networks. Been playing with these for a few months and am quite fascinated with them! Hoping to find more time in the future.

[kache @yacineMTB]

there's an entire subculture of people out there that are hard baking spiking neural networks into FPGAs & ASICs

[Gerald McAlister @GEMISIS]

The only folks who think software engineering is dead are too far off the ground to know

The only folks who think coders are still needed aren’t working on complex problems

Both groups are not shipping real products

The industry is changing, plain and simple

[Gerald McAlister @GEMISIS]

@zekramu Hey now, it wasn’t always slop

Sometimes the code just never shipped

Or didn’t do anything of value

But in those cases at least it was beautiful?

[Gerald McAlister @GEMISIS]

@pvncher @kevinkern Yeah, I’m working on a side thing for this specifically. It’s insane to me how primitively context windows are treated across the board today!

[Gerald McAlister @GEMISIS]

Honestly, probably fine going with less cash (most folks Staff+ level aren’t too dependent on the cash by that point) in favor of more equity.

The big things are:
- it needs to be meaningful - paper worth of a minimum 8 digit payout after a few years of work
- anti dilution clauses - if founders maintain their percentage then so should those early folks
- founder buyout matching - if a founder gets bought out, then everyone else needs to be bought out too

You can easily sweeten the deal with remote work policies, fewer work hours (part time folks?), etc. but there’ll be tradeoffs to those of course.

[Gerald McAlister @GEMISIS]

This is some of the most exciting news of the year!

[s&box @s8box]

Box3D is now public & open-source. 🎉

We've used it as our 3D physics engine for the past year, for a Sandbox game where players can build whatever they imagine, or developers can build any game - Box3D handles anything thrown at it.

Open development like this makes games better for everyone. Congrats to @erin_catto on the release.

https://t.co/nfFPYHPL8M

[Gerald McAlister @GEMISIS]

Reading through the founders original post again at https://t.co/sYYh0J5prB it’s looking like this was some early AI “psychosis” that managed to find some funding. The entire thing now reads to me like someone who just asked Claude/GPT to help them structure their arguments, not someone building a product.

Have you heard from him about that bet yall made? 😂

[Daniel Park @danifesto]

https://t.co/D0k4BM4z3n

[Gerald McAlister @GEMISIS]

My 2 cents: the startup market is shifting to favor actual engineers again, not Wannabe Rich founders, due to AI.

Traditionally, founders had to spend a chunk of their time split 3 ways:
- building
- fundraising
- hiring

This is because each part of this feeds into the next:
- Building is how you prove your product and get funding
- Funding is how you get money early on and begin hiring more folks
- Hiring more folks is how you build more and faster

AI fundamentally broke this: a good engineer can do more on their own than with a team, meaning they need less money and fewer people for early success.

[Nick Davidov @Nick_Davidov]

I say “it’s not me it’s the market”, meaning it’s founders we work with and talk to. An increasing number of them. Although if you open my site, our portcos post a ton of jobs that are not SF based. My team is fully remote btw. Yet retards go attack me personally. Looks like it hurts that they can’t get decent jobs and can’t be founders, so they take it out on ppl on X.

[Gerald McAlister @GEMISIS]

@Nick_Davidov I mean, if they aren’t finding these folks, then the problem is them, not the market. It’s quite simple. Those founders will lose out in the end.

[Gerald McAlister @GEMISIS]

Yeah, these days if I don’t think a company can ipo on its own, I avoid it. The era of exits outside of ipos is basically over due to political stupidity, meaning unless you’re a founder you’ll be screwed in these companies. The good news is ai means small founder only company buyouts are significantly more achievable now.

[Gerald McAlister @GEMISIS]

I think this is a bit of a misnomer: you don’t need irrational optimism, you just need to actually pay folks and make the risk worth the reward.

As an example: I had an offer from a (now defunct) VR company that wanted to underpay me (and grossly underlevel me) with the reason being i can “use the experience to break into the games industry”

As if I hadn’t been doing VR and making games for over a decade already.

The reality is, the world runs on money, not optimism and not “experience”. If you want the best folks, then you have to pay.

[Gerald McAlister @GEMISIS]

I’m a fan of what I call “dumb tech”: smart enough for me to get sensor data/control, too dumb for anything else. Just tell me if the fridge door is open, what its temp is, and maybe a camera stream to what’s in it. I don’t need a screen on it, or for Siri to connect to it 😂

It’s why I’m focusing my smart home setup towards a 0 app setup.

[Gerald McAlister @GEMISIS]

@sdamico FirstAlert are simple but:
- Z-wave compatible
- Smoke and CO combined
- give alarm status + battery status as sensor data
- No required 3rd party app

[Gerald McAlister @GEMISIS]

The actual chip software being C/C++ is what im referring to, not the tooling above that. Thats where rust may be a harder selling point since you’ll have 2 languages to manage instead of one.

Rewrites don’t much of a cost now due to agents. I don’t buy that rewriting has a higher cost than the memory leaks.

[Gerald McAlister @GEMISIS]

Hot take: the industry being stuck on python probably does more to hurt performance and increase chip costs than the current improvements in models.

I think the only thing that might be better than C/C++ is Rust, and that’s solely because you can more easily have agents self-improve while getting compile time protections to save time on issues like memory leaks and such.

Of course you’ll still need C/C++ for the chips themselves so maybe Rust is a wash for that. Either way, this change will probably yield great results for X!

[Elon Musk @elonmusk]

@aaronburnett Truly massive gains will come in ~3 months when the entire training and inference stack is written in C/C++ and massively simplified (most software layers will be deleted completely) and we exact-map Grok to work incredibly well on a GB300

[Paul Graham @paulg]

Not using LLMs to write for you won't be like not using Google Maps in a new city. It will be like choosing to run and lift weights, even though we now have machines that can transport us and lift weights for us.

[Gerald McAlister @GEMISIS]

@juliey4 I wonder how much of this is caused by the high stresses of today’s world vs a lack of taking care of one’s body (both physically and mentally via gym + therapy). I suspect it’s more of the latter from what I see day-to-day, but pure anecdotal on that note.