The Delaware court's opinion on the compensation package for Elon Musk suggests that $TSLA could simply put together a new pay package. The only difference would be that Tesla would incur an accounting charge of $25+ billion, whereas with the previous package, the charge was

If you believe that $TSLA can sell at least 3-4 million Optimus humanoid robots at $25K a piece 5 years from now, then the present value of that, assuming a 10% discount rate, is around $50 billion. That's what it would cost the company in accounting charges to issue Elon Musk a

$PLTR UBS just initiated its coverage at $80 target. 17 large customers unveiled critical details on the success of AIP. Here is what you need to know: 1. Palantir is the perfect all-in-one platform "The front end is just too unique. Palantir is really the perfect

The most beautiful thing on LLM reasoning is that the thought process is generated in an autoregressive way, rather than relying on search (e.g. mcts) over the generation space, whether by a well-finetuned model or a carefully designed prompt.

Thoughts on the eve of AGI I talked to several friends about o3 this week. Their summarized response is basically "holy crap is this actually happening?" Yes, this is actually happening. The next few years are going to be insane. This is historic stuff, galactic even. What's

2011 VIC pitch on $AMZN is one of the best educational case studies on investing in multibaggers. In 2011, this was a huge $85bn Mcap company at 100x PE with major banks calling it overvalued. How do you push past all that as a die-hard value investor? Let’s dive in 👇

LLMs are a commodity with Palantir AIP. Here's how easy it is to swap them out or use them together.

A great chart by SemiAnalysis shows the spike in demand (price) for $NVDA H100 on AWS after DeepSeek. This is something that I also discussed in my DeepSeek article: With the reduction of LLMs, you will have more usage, which benefits the whole ecosystem. Good for $AMZN, $MSFT,

Mistral AI, a French AI startup, launched a new LLM that processes tokens 10 times faster than ChatGPT-4o. You may not have heard of this company before, but it's a $NBIS client for training its AI models. Here's a quote from its CTO about $NBIS: 👇🏻👀

Post the release of R1: DRAM pricing: increasing. GPU rental pricing: increasing at GCP and Azure. GPU availability: decreasing. Jassy and Sam: compute constrained per comments yesterday. GPT 4.5 disappointment: most likely b/c they couldn’t make over 50k GPUs coherent.

Everyone in here saying "but can you build salesforce from prompt, show me" (for the record, I think we're not far...) Why hasn't it been done? Because no specific person needs all that. End users need one specific capability at any given time, not all of them, and if you can

SaaS is being dismantled as we speak! We're witnessing the slow-motion collapse of an entire business model that dominated tech for two decades. The $1.3 trillion SaaS is being quietly hollowed out from within by AI agents. Here's how I see it playing out: Phase 1 (Now): AI as

GREG ISENBERG It might be an odd thing to compare it to, but there's another angle to look at this too: Software is now content. I actually saw the same thing with video when the Canon 5D Mark II came out. When that camera dropped, so many more people had access to making high-quality

AI is currently progressing faster than any technology in history but in opposite directions at the same time; 1) Reducing inference cost per unit of intelligence for large language models (down 600-2,400x in 2 years) and 2) Increasing the total compute we can spend on the same

Software Business Models In The Age of AI: From SaaS to WaaS to RaaS I've written before about the idea of "RaaS" (Results as a Service) whereby instead of paying a subscription to access software, you pay for the actual results that the software delivers. I still think that's

We actually have 3 different ways to scale inference compute to exchange $ for more intelligence with the same model: - Series scaling of thinking tokens; Longer Chain of Thought: o1/o3 - Parallel model instances plus automated selection of the final answer: o1-pro & o3-1024k

LLMs show us that your internal dialogue is one of the most important things you have. It’s what makes you human. Give it space, curate it. Turn off the podcasts and go for a walk.