4 ms·

I wonder how much of this startup playbook is just a way to rationalize luck. Companies that succeed may not necessarily do all of these things. Companies tha

by twayt 3y ago

I wonder how much of this startup playbook is just a way to rationalize luck. Companies that succeed may not necessarily do all of these things. Companies that fail may do all of these things. A person who'd be inclined to click this link is likely new to startups and wants some kind of structure, knowing how unpredictable things can be. In some sense, this is exactly "what people want" ala YC advice. As long as they say a few milquetoast things that other established entrepreneurs cannot argue with, they have essentially used the (possibly random) initial successes they've had to build credibility in the eyes of potential founders. There is no doubt that a large part of YC's successes are due to its network effects but you first need to get lucky in order to build it. That's the part they seem to leave out.

revelio 3y ago
I don't mean to be too cynical. I think a lot of the advice in this article actually rings true and aligns with my personal experience. No advice on something as vague as success can be comprehensive and OpenAI may just be an exception. Especially as it's not a successful business yet and given their enormous raises, it'll be years before anyone can tell if they're actually able to swim on their own. For example, when I was young I worried that ideas were precious and had to be protected in case I told someone and they stole it. Now I recognize the fundamental truth Sam expresses in the article: the best ideas sound bad, and the majority of people will just roll their eyes if you tell them what you want to do. I've directly had that experience with several successful projects so this isn't just nodding along to something that sounds vaguely aspirational. Whilst OpenAI seems to violate a lot of what's in the article, in that way it's example of it. They're in their current position because groupthink within the AI space came to state that scaling up language models wasn't worth doing and academically uninteresting compared to developing more sophisticated neural architectures. OpenAI said, no, let's just try spending shittons of money and engineering on brute forcing enormous model sizes with an ordinary-ish transformer network and see what happens. Right up until they started getting these amazing results they were sort of out on their own in being so committed to that approach.
- twayt 3y ago
  Using the example that you just used: The best ideas sound bad. Really bad ideas also sound bad. Many good sounding ideas are also hugely successful businesses. Maybe the etiology of success had nothing to do with how bad sounding the idea is? Also in the case of OpenAI, being in the field, I'd strongly argue that it wasn't that crazy sounding of an idea as people might conjure in retrospect. Google had been throwing resources at LLMs and been using them in production for several years before OpenAI started to do it. MANY papers written about scaling laws and how we haven't reached the limit of scaling data / compute yet. OpenAI is only relevant because they iterated quickly on releasing a product. Absolutely nothing visionary about the technology honestly. The question that should be asked here is why OpenAI was the first to release a product a few other companies were capable of creating.. Everything OpenAI has done is low hanging fruit. The logical progression of LLMs is chat, zero-shot learning, multimodal, connect it to web / other knowledge bases, planning, personal customization, multi-agent systems, etc. There is a vast amount of research in several vectors in this area..

by twayt 3y ago

revelio 3y ago

I don't mean to be too cynical. I think a lot of the advice in this article actually rings true and aligns with my personal experience. No advice on something as vague as success can be comprehensive and OpenAI may just be an exception. Especially as it's not a successful business yet and given their enormous raises, it'll be years before anyone can tell if they're actually able to swim on their own. For example, when I was young I worried that ideas were precious and had to be protected in case I told someone and they stole it. Now I recognize the fundamental truth Sam expresses in the article: the best ideas sound bad, and the majority of people will just roll their eyes if you tell them what you want to do. I've directly had that experience with several successful projects so this isn't just nodding along to something that sounds vaguely aspirational. Whilst OpenAI seems to violate a lot of what's in the article, in that way it's example of it. They're in their current position because groupthink within the AI space came to state that scaling up language models wasn't worth doing and academically uninteresting compared to developing more sophisticated neural architectures. OpenAI said, no, let's just try spending shittons of money and engineering on brute forcing enormous model sizes with an ordinary-ish transformer network and see what happens. Right up until they started getting these amazing results they were sort of out on their own in being so committed to that approach.

twayt 3y ago
Using the example that you just used: The best ideas sound bad. Really bad ideas also sound bad. Many good sounding ideas are also hugely successful businesses. Maybe the etiology of success had nothing to do with how bad sounding the idea is? Also in the case of OpenAI, being in the field, I'd strongly argue that it wasn't that crazy sounding of an idea as people might conjure in retrospect. Google had been throwing resources at LLMs and been using them in production for several years before OpenAI started to do it. MANY papers written about scaling laws and how we haven't reached the limit of scaling data / compute yet. OpenAI is only relevant because they iterated quickly on releasing a product. Absolutely nothing visionary about the technology honestly. The question that should be asked here is why OpenAI was the first to release a product a few other companies were capable of creating.. Everything OpenAI has done is low hanging fruit. The logical progression of LLMs is chat, zero-shot learning, multimodal, connect it to web / other knowledge bases, planning, personal customization, multi-agent systems, etc. There is a vast amount of research in several vectors in this area..