Ask HN: Do you feel OpenAI is in a league of their own? | Hacker News Reader

Ask HN: Do you feel OpenAI is in a league of their own?

6 pointsSeanAnderson5 comments

How important is first-mover advantage in this space? How about the rapid pace of compounding development iteration abilities via the use of AI itself?

There are clearly a lot of incumbents in the space - I expect all the largest tech companies are rolling their own and then there's off-shoots like Anthropic. There's also the hope that open-source models will usurp the private models, similar to what happened for a while with DALL-E vs Stable Diffusion.

I'm curious how relevant others see these other players and options in this space. Do you expect OpenAI to maintain its breakaway lead for the forseeable future? Are others playing their cards close to their chest and will surprise us? Will the rate of progress slow and allow for the field to stabilize with multiple companies having effectively comparable offerings?

View on HN

> league of their own?

OpenAI isn't really open. It's an opaque black box with no transparency as to how it works, apart from surface-level details of how it works. I imagine when something similar to OpenAI, (like some skunk-works garage project) gets released on GitHub and we have full insight into how it works, OpenAI's days will be over. We need insight into these things, otherwise it's another closed ecosystem we have to blindly trust.

> Do you expect OpenAI to maintain its breakaway lead for the forseeable future?

Eventually, there will be an alternative that is just as good as O̶p̶e̶n̶AI.com. They are not in a league of their own. They are just great at marketing hype, only possible alongside Microsoft.

The top one(s) to watch in AI is DeepMind (Google), Meta, Nvidia and Anthropic as they push for the foundational breakthroughs in AI. Not just LLMs.

> Are others playing their cards close to their chest and will surprise us? Will the rate of progress slow and allow for the field to stabilize with multiple companies having effectively comparable offerings?

Yes. The under hyped ones that almost no-one is looking at and will definitely surprise the industry will be Cohere.ai and Cerebras Systems. Apple Inc. may decide to enter later with their Apple Silicon hardware and will push for smaller, local and useable AI models in Macs and iPhones focusing on 'privacy'.

Hardware compute is what separates these companies and Cerebras Systems, amongst all the hype is a hardware AI startup that is going to be acquired by a large company. Anyone would want systems to cut the time to train extremely large AI models at scale to maintain their lead. I wonder who could that be...

I wouldn't say they are "in a league of their own," but they are the clear front-runner in a race to become the dominant "generalist AI model" layer in a new kind of AI software stack that is growing and expanding into new applications very rapidly.

Every day, we're seeing (a) more and more tools and plugins that enhance the abilities of OpenAI's generalist models, (b) more and more tools for using OpenAI's generalist models as components of applications, and (c) more and more end-user applications built atop OpenAI's generalist models.

Most businesses that want to leverage a generalist AI model cannot afford to develop and train it on their own, especially if the cost of doing so continues to increase as rapidly as expected. In fits and starts, they will eventually settle on using one dominant provider with an industry-standard API.

No, not after seeing how they’re displacing Google before our very eyes. No one is in a league of their own. Especially with the availability of tools like LLMs which can be used to disrupt just about anyone.

LLMs are way too computationally expensive. I would like to see specialized and unfiltered models, but the only way that could happen is through an untapped market for such a thing, or breakthrough research to make it easier to run on consumer hardware. LLMs will only be run by million dollar companies for some time.

Reply on news.ycombinator.com