Proprietary LLM builders need to experience a valuation haircut as open source LLMs take share from proprietary LLMs.

Proprietary LLM builders (OpenAI, Anthropic, Google, Microsoft, Amazon), have enjoyed lofty valuations over the past several years. Given the rise of open source competitors - which are on par with proprietary models from a performance standpoint and can be operated at a fraction of the cost - the proprietary model builders should suffer a valuation haircut.

I believe that open source LLM builders such as DeepSeek and META will win the day and that 80% of LLMs and SLMs in production 5 years from now will be open source language models.

https://open.substack.com/pub/tek2day/p/valuation-haircut-is-due-for-proprietary?r=1rp1p&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false

Originally published Dec 21st 2024. View the video version here on YouTube: https://youtu.be/6P0jQxZTuQo?feature=shared

We demo NotebookLM for a YouTube video, for a TEK2day article and for an EPS call transcript. Watch the video version of this episode here: https://youtu.be/wjqMhBdTxSQ?feature=shared

Watch the video version of this episode here: https://youtu.be/dc68lkZ1Bxo?feature=shared

At some point cost and payback period will factor into frontier LLM building, especially as use cases are not well defined.

We are at the $1 billion LLM level today. $10 billion will likely be the cost of developing frontier LLMs by 2026, $100 billion by 2027 and $1 Trillion by 2028 should the current pace of development continue.

In episode 507 we make the case for smaller, “baseline” language models that are industry domain-specific, trained with opensource data as well as with proprietary enterprise data. These baseline models could power various applications and services and also be used to train third-party models. This scenario would create a natural selection/survivorship process for language models whereby smaller models power well-defined use cases that address specific commercial needs. This path makes more economic sense than developing ever larger monolithic LLMs in a vacuum.

Watch the video version here: https://youtu.be/v-2vsjFEXdo?feature=shared

We walk you through Google's AI-based podcast generation tool. Watch the video version of this podcast episode here: https://youtu.be/y2CNtayIBrU?feature=shared

View the video version of this podcast episode here on YouTube: https://youtu.be/dbwP3r6Nvqw?feature=shared

Read the related TEK2day article here: https://tek2day.substack.com/p/backtesting-the-tek2day-founder-ceo

See the backtested portfolio here: https://www.portfoliovisualizer.com/backtest-portfolio?s=y&sl=3zih7QmbLhWF4jG4AUChu1

See this podcast episode on X: https://x.com/JonathanMaietta/status/1828490305815093585

Watch the video version of this podcast episode on YouTube: https://youtu.be/7XHKgYLYZP0?feature=shared

Read the related Substack article here: https://tek2day.substack.com/p/founder-ceos-vs-comps

Watch the video version of this podcast episode on X: https://x.com/JonathanMaietta/status/1828445724075254096

Eric Schmidt's comments regarding Agentic AI were overly bullish in our view whereas Deepmind co-founder Demis Hassabis has a more grounded perspective.

View Schmidt's remarks on our Substack page: https://open.substack.com/pub/tek2day/p/former-google-ceo-eric-schmidt-speaks?r=1rp1p&utm_campaign=post&utm_medium=web

View Schmidt's remarks on X: https://x.com/JonathanMaietta/status/1826427051693502607

View Hassabis' talk on the subject of Agentic AI: https://youtu.be/pZybROKrj2Q?feature=shared

View the video version of this podcast episode on YouTube: https://youtu.be/EJRSNHKlQwo?feature=shared

View the video version of this podcast on YouTube: https://youtu.be/8Y9MCNVQlcY?feature=shared