vessenes

45 ms·

931.
by vessenes 1y ago
Roughly 1/10 the cost of Opus 4.1, 1/2 the cost of Sonnet 4 on per token inference basis. Impressive. I'd love to see a fast (groq style) version of this served. I wonder if the architecture is amenable.
932.
by vessenes 1y ago
Overall this seems good to me, esp. since it’s limited to 2x a game, leaving space for framing and other human elements to the calling, like make up calls, which I generally think are good. That said. Strike zone height is between 53.5% and
933.
by vessenes 1y ago
Yep. As the economy changed and large estate living became untenable in England, a group moved to Kenya to sort of extend their lifestyle.
934.
by vessenes 1y ago
I’m the other way — the cost of launching a creative / interesting software company / project just got cut to 1% or so. (I said launching. Maintaining … obviously not quite as good on the numbers). I propose software creation, and
935.
by vessenes 1y ago
I think you’re wrong. Consider the following. It’s 1995. You and your next door neighbour Jeff Bezos have both just raised $10mm from competing VCs to build amazon.com. You can choose to have a Claude API portal to the future where you pay
936.
by vessenes 1y ago
Totally agreed.
937.
by vessenes 1y ago
You have LLM derangement syndrome, and don’t understand. Say the first model cost $2 to make. On metered sales, they’ve made $10 on it. They then decide to make a $20 model, raising more money. It turns out, that model made $100. They then
938.
by vessenes 1y ago
If you talk to any of his early investors, they considered him absolutely crucial to the project.
939.
by vessenes 1y ago
Many Brits disagree with you — they’re serious about maintaining / reclaiming their class-based lifestyle in the UK - usually enabled by cheap labor — google “Happy Valley” if you’d like to see the 100 years-ago version of this.
940.
by vessenes 1y ago
Question - anyone here have one of those “most popular” VWs? The article says they’re called id.3 id.5 and id.7. I’ve ridden in an e-Golf (California compliance car basically), but not the newer ones. The e-Golf did not impress, so I’m curi
941.
by vessenes 1y ago
Notable in that this is research out of the genomics lab at Stanford - it’s likely that an ml practitioner could do better with a more hands on approach - but demonstrating some end to end work on genomics implementations as they do in th p
942.
by vessenes 1y ago
The article is basically worthless. Wealth follows a power law; saying 90x% of millionaires didn’t move is noise. The very wealthy are in fact moving; they don’t want to be subject to a tax jurisdiction that claims reach on global income. N
943.
by vessenes 1y ago
They account internally for each model separately; Dario said they even think of each model as a separate company on Dwarkesh some time ago. Inference services are wildly profitable. Currently companies believe it’s economically sensible to
944.
by vessenes 1y ago
They’re already spending as much money as they possibly can on growth, and have no further use for cash currently - they’ve been doing share buybacks this year.
945.
by vessenes 1y ago
Wait, why the quotes? NVDA sends cash, and the Coreweave spends it, no? I don’t think quotes are accurate, if they imply these transactions aren’t real, and material. At the end of the day, NVDA owns Coreweave stock, and actual, you know, p
946.
by vessenes 1y ago
So, basically a single BYD factory
947.
by vessenes 1y ago
Yeah the path forward here is going to be Apple-like vertical supply chain integration. There is absolutely no spare capacity in the infra side of electrical right now, at least in the US.
948.
by vessenes 1y ago
NVDA outstanding shares are down ~1.2% year over year; the company has been buying back its own shares with —>> profits <<— to the tune of tens of billions. Meanwhile NVDA stock is mildly up on this news, so the current owners o
949.
by vessenes 1y ago
Almost every model trained by the majors has paid for itself with inference fees. I’m not saying there isn’t a bubble, but I am saying if the researchers and strategists absolutely closest to the “metal” of realtime frontier models are corr
950.
by vessenes 1y ago
I’d like to see browser agent modes publish their scores on this. Definitely the next great benchmark
951.
by vessenes 1y ago
I too avoided it; I had an interesting interaction with the (American) call center scammer -- he called, said his story; he gave me a callback number when asked; I asked him for a web page I could verify a callback number. He quickly rattle
952.
by vessenes 1y ago
Well, he’s currently running a startup aimed at making better tooling for the space. So, he’s putting his time where mouth is.
953.
by vessenes 1y ago
George famously started a self driving hardware-as-an-addon company. He started after both claiming self driving was easy, and reportedly taking up an Elon bounty/bet in the seven figure range to make self driving AI for Tesla.
954.
by vessenes 1y ago
Agreed. That said, in general a 1M context model has a larger usable window than a 260k context model.
955.
by vessenes 1y ago
We are nearly infinitely far away from saturating compute demand for inference. Case in point; I'd like something that realtime assesses all the sensors and API endpoints of stuff in my home and as needed bubbles up summaries, diaries,
956.
by vessenes 1y ago
Unfortunately, no. The industry is moving super quickly, and spinning up new ideas on the backs of old ones at a fast rate. If you want to understand what's going on, I think the best thing to do is some intro courses, train and design
957.
by vessenes 1y ago
Thanks for writing back! I appreciate the plan to integrate the two architectures. On that front, it might be interesting to have a future research section - like what would be uniquely good about this architecture if scaled up? On ‘usefuln
958.
by vessenes 1y ago
Meh. Well, at least, possibly “meh”. Upshot: Gaussian sampling along the parameters of nodes rather than a fixed number. This might offer one of the following: * Better inference time accuracy on average * Faster convergence during traini
959.
by vessenes 1y ago
I like the idea of credentialing by relying on the separation of search corpus and training - including links to the global coverage of this event, a critical turning point in how ethical AI can be most helpful to humanity. I’d like to talk
960.
by vessenes 1y ago
Curious: are you trolling? Or do you really think doing math shouldn’t be a priori allowed to humans?

More ›