Why didn't your VC friend drop some seed on you back then if the stealth startup was doing 25MM ARR? They probably could've had a better deal with you!
Oh they had $25MM in funding, $0 ARR - lol - the two reasons I decided to leave it as a weekend project: "Thanks!!! I decided not to build it, that space is already too busy, there is a startup with $25MM in stealth, who else is in stealth? On top of that, this method will get stale very very quickly, foundation model businesses are just too hard to work around right now, it's a silly way to do business. My magic is I've build a startup from scratch to over 400 people and watched what they do, it won't be long till that isn't worth much." and "I build it on my own over a weekend, on my own. I just wanted to confirm it can be done therefore will exist, that is all. Personally, I decided not to peruse it because I am old and lazy and don't want to compete against a16z and sequoia funded adderall filled teenagers." - 8 months forward I was right: If I could slap a bunch of python together and create business as a service that actually functions, in a weekend as just me, that's a feature on someone elses product! :)
Backups are a PITA I wanted to go exactly this route but even though I had VMs and compute I can't let any production data hit it without bullet proof backups.
I setup a cron job to store my backups to object storage but everything felt very fragile because if any detail in the chain was misconfigured I'd basically have a broken production database. I'd have to watch the database constantly or setup alerts and notifications.
If there is a ready to go OSS postgres with backups configured you can deploy I'd happily pay them for that.
I feel exactly like OP, I was visiting Ask before actually making my own "Whats the point?" post. I think the real issue is this huge community of people who have gone completely gone awry with using LLMs in their own loops and constantly posting and talking about it.
I see dozens of people on HN just posting about how amazing it is to write/compose software now. They're making more software than ever and having the time of their lives. When I read those and I actually go and explore that software they're just OSS tools, I wonder why would anyone want to use this? If everyone was doing as they were they're just asking their LLM to do it instead of looking out for a tool. Even better they'll just ask their LLM to make a tool to accomplish whatever those authors are building.
Then it's this whole new religion of human out of the loop. You feel like you've either gone stale or insane because every one now says adding human into the loop worsens the productivity gains from a model. I highly disagree, I haven't used a single model that handles a substantially complex task flawlessly. If you mention anything about that people don't shutup about harnesses.
Don't get me wrong, I use LLMs, I use them quiet frequently. However it's this obsessive attitude towards them that makes it impossible to get funding or research for anything that's not at least tangentially related them. It's completely burned me out professionally, academically and psychologically.
But that's exactly what you should be doing, technically. Human in the loop is a dead concept, you should never need to understand your code or even know what changes to make. All you should be concerned about is having the best possible harness so your LLM can do everything as efficiently as possible.
If it gets stuck, use another LLM as the debugger. If that gets stuck then use another LLM. Turtles all the way down.
You were confidently wrong for judging them to be confidently wrong
> While EMMA shows great promise, we recognize several of its challenges. EMMA's current limitations in processing long-term video sequences restricts its ability to reason about real-time driving scenarios — long-term memory would be crucial in enabling EMMA to anticipate and respond in complex evolving situations...
They're still in the process of researching it, noting in that post implies VLM are actively being used by those companies for anything in production.
I should have taken more care to link a article, but I was trying you link something more clear.
But mind you, everything Waymo does is under research.
So let's look at something newer to see if it's been incorporated
> We will unpack our holistic AI approach, centered around the Waymo Foundation Model, which powers a unified demonstrably safe AI ecosystem that, in turn, drives accelerated, continuous learning and improvement.
> Driving VLM for complex semantic reasoning. This component of our foundation model uses rich camera data and is fine-tuned on Waymo’s driving data and tasks. Trained using Gemini, it leverages Gemini’s extensive world knowledge to better understand rare, novel, and complex semantic scenarios on the road.
> Both encoders feed into Waymo’s World Decoder, which uses these inputs to predict other road users behaviors, produce high-definition maps, generate trajectories for the vehicle, and signals for trajectory validation.
They also go on to explain model distillation. Read the whole thing, it's not long
But you could also read the actual research paper... or any of their papers. All of them in the last year are focused on multimodality and a generalist model for a reason which I think is not hard do figure since they spell it out
You've successfully hacked the collective HN hive mind. I can't go a week without either seeing your post on the frontpage, someone mentioning you, your comment branching into a huge thread or obligatory pelican riding bike SVG.
I don't personally have a taste for LLM comparison posts but your consistency has paid dividends. SimonW is tattooed in my eyelids, a name I shall never forget. Wishing you all the best.
I mean how far Rusts own clippy lint went before any LLMs was actually insane.
Clippy + Rusts type system would basically ensure my software was working as close as possible to my spec before the first run. LLMs have greatly reduced the bar for bringing clippy quality linting to every language but at the cost of determinism.
reply