Josh Vera
@joshvera.com
Owner of @paradiselostnyc in East Village. GitHub alum.
*looks at Damodaran's NVDA analysis*
*looks at the board*
*looks at my hardware needs*
Yeah, that's a buy.
*looks at the board*
*looks at my hardware needs*
Yeah, that's a buy.
January 28, 2025 at 4:40 PM
*looks at Damodaran's NVDA analysis*
*looks at the board*
*looks at my hardware needs*
Yeah, that's a buy.
*looks at the board*
*looks at my hardware needs*
Yeah, that's a buy.
As a consumer of LLMs, I nearly dropped $30k on MacBooks before realizing I could probably only get 7 words per second out of them. Maybe ok for a hobby, but I really need to drop $300k for H200s to do anything meaningful at scale. R1 agents talking to each other would saturate that. Seems worth it.
January 28, 2025 at 4:40 PM
As a consumer of LLMs, I nearly dropped $30k on MacBooks before realizing I could probably only get 7 words per second out of them. Maybe ok for a hobby, but I really need to drop $300k for H200s to do anything meaningful at scale. R1 agents talking to each other would saturate that. Seems worth it.
Until we crack program synthesis, we're living in probabilistic land where we can't prove a logical conclusion is true. DeepSeek is on the case with their open source Prover model. OpenAI might be ahead with O3, but many labs are pushing the boundaries of what's possible with grounded reasoning.
January 28, 2025 at 4:40 PM
Until we crack program synthesis, we're living in probabilistic land where we can't prove a logical conclusion is true. DeepSeek is on the case with their open source Prover model. OpenAI might be ahead with O3, but many labs are pushing the boundaries of what's possible with grounded reasoning.
The research was there, it just wasn't evenly distributed. So I'll throw my hat in with the tech hubris. I'm not saying they haven't printed money. But that hubris still seems quite costly from this vantage point.
January 28, 2025 at 4:40 PM
The research was there, it just wasn't evenly distributed. So I'll throw my hat in with the tech hubris. I'm not saying they haven't printed money. But that hubris still seems quite costly from this vantage point.
Re: tech hubris - 4 years ago you couldn't beg GitHub/Microsoft to work on ML code generation with vector embeddings. I should know, I was on the Semantic Code team at the time. Fast forward and they're playing catch-up through acquisition, and restructuring the whole org around AI/Copilot.
January 28, 2025 at 4:40 PM
Re: tech hubris - 4 years ago you couldn't beg GitHub/Microsoft to work on ML code generation with vector embeddings. I should know, I was on the Semantic Code team at the time. Fast forward and they're playing catch-up through acquisition, and restructuring the whole org around AI/Copilot.