Software and electrical engineer with a bit of data scientist. 25 years experience. CNC, metal casting, and 3D printing enthusiast. Father and husband first.
@esrtweet The actual margin on hosting models is fine even at kimi 2.6 type prices. Training the next model and building out the next DC eat up all the profit, but thats not atypical for capital intensive growth. Worst case, they slow down a bit.
@lian75864@xai It gets used more than you think. I have 4.3 deployed in a lot of corporate use cases via API because it's cheap, has a huge context, has vision, and most importantly and unlike most other models, it doesn't hallucinate very much.
@theo I am no fan of Anthropic but I've thrown 4.8 at a lot of really complex problems today and its certainly stronger than 4.7 and much stronger than composer 2.5, which I use a lot, in virtually everything. My use case is bespoke analytics services on kubernetes with UX.
@fyrewede@Devon_Eriksen_ Yes, he is almost certainly in the top 1% for his age. That lifestyle should be (and once was) available to a much much larger share of the population in his age bracket. I say this as a gen x that has also had a pretty successful career (and my share of screw ups).
@maiab Same here. I blame it on being born nearly deaf in one ear, but that's just a guess.
Regardless, it's become an increasingly serious issue for me due to the world changing around us.
@__tinygrad__ The work you guys do is worth every penny of profit you're making on these.
Selling assembled and supported bespoke hardware like this is a really tough business and a valuable service for those that need to get a job done. People have no idea.
@staysaasy Of course they start new companies. You'd be shocked how many subsidiaries large companies have partly for this reason. It comes back to brand pollution and also the reason you incorporate in the first place (liability shells).
It's a judgement call. You use products everyday that only exist because some obsessive and highly motivated person wasn't happy with the status quo.
Here are just a few examples:
Linear: Created because founders were frustrated with bloated, slow enterprise issue trackers like Jira. They designed a sleek, lightning-fast, and keyboard-centric project management tool.
Figma: Built because designers were frustrated by fragmented collaboration models (like emailing files and using multiple plugins) in existing software suites. They pioneered the first browser-based, collaborative design platform.
Shopify: Started by founders who wanted to sell snowboarding gear online in the early 2000s but were deeply frustrated by the clunky, restrictive, and expensive e-commerce SaaS platforms of the time.
Zulip: Developed because the founders were frustrated with the information overload and unthreaded communication of tools like Slack and IRC. They designed a platform centered entirely around threaded conversations.
Notion: Built as a response to the frustration of having to use entirely different apps for wikis, tasks, and documents. The founders created an interconnected, "all-in-one" workspace.
Retool: Created because developers were tired of constantly rebuilding repetitive internal tools and dashboards from scratch when existing SaaS drag-and-drop builders were too limiting.
@tonjkb In reality, while on the road you virtually never use the tesla touch screen for anything but confirming navigation and maybe changing the music if you dont want to use your voice. There is nothing to be gained here.
@OEAMBRCH@niccruzpatane You're confusing luxury with status. There is literally nothing on the road more luxurious than a new Tesla driving you around quietly with an amazing sound system, doing it reliably for the next 300k miles, and never having to go to the gas station again.
@Vixhal The 2019 Mac Pro is my all-time favorite for case design, though I don't like things like proprietary power supplies. I bought one in 2020. Financially, it was the fastest depreciating asset I've ever purchased, losing 80% of its value in about 3 years.