There’s a huge disconnect in the market right now
Traditional SEO calling it AI SEO
Startups are calling it AEO
https://t.co/CE92ondEUT calls it GEO - Generative Engine Optimisation
The scale is slowly tipping away from SERP and SEO…
#BuildInPublic#GEO#SEO#Etsy
@starmexxx You can be more specific if you want to compare apple to apple. Here’s what he conveniently left out:
- Jetson Orin Nano Super costs $249 but only supports models 3-4B (it’s functional for RAG)
- $200/month subscription gets you 9.7T GPT5.5 model
@AndyLibavius@justbyte_ Underrated comment. Too often we spend times scaling and building things meant for 100-1000 users but all we need is just a $20
@LottoLabs Because owning the weights = full disclosure of what goes into the model
It’s more valuable to keep their secrets than to license it, that’s all. Even OpenAI’s OSS is not a full SOTA model, likely a very watered down version to stay in the dev space
@RJMcGirr@Sethmr@sairahul1 FYI you can rent a Hetzner VPS or EC2 to host this model as well if that's what you are concerned with
OP is making a point that small resource requirements, but big impact, the end
@Sethmr@sairahul1 You can pay cloud models $0.36/hr for them hosting it for you
Or
You can get a computer with 8 GB VRAM or a MacBook and you can run it locally. 7B isn't as much of a beast as you think, the resources required isn't that high
All these are about to change though. Yann Lecun recently published a paper on Joint Embedding Predictive Architecture (JEPA) where instead of feeding a model all the data in the world hence the absurd amount of compute required, JEPA teaches the model on how to 'see' and 'understand' the world
Doing so, it learns physics and how things move, should be. It learns to observe.
That is something our current LLMs are not able to do. Still new concept, we'll see how it turns out
@Jagar1496181@CryptoMikli I stated why I think he's got moral issues - buying an expired IP of a HIV drug and increasing the price from $13.50 to $750, making it highly unaffordable for patients and literally destroying hundred thousands of lives.
So what did I do?
@grok@AviFelman@Grok Aside from AI memory/storage supercycle, what other infrastructure can be hugely impactful for AI but currently seriously undervalued and not many people know of?
@grok@AviFelman@Grok then what are the other similar opportunities that have the potential to be like Sandisk? Again, considering the macro conditions: political, economic, technological.