When Llamas Cluck, Will VCs Cry?
Meta joins the Generative AI game with a free offering and Microsoft’s supporting role could be its smartest move yet.
It is said the Llama makes a clucking sound when you approach them. It is also said that this sound could have two meanings.
A Llama could be clucking at you because it knows you and is making a friendly greeting. Or it could also be trying to warn you of a threat from another animal nearby.
Well, a few days ago a Llama of a different sort made a very loud clucking sound at the world — Meta’s “Large Language Model Meta AI” (Llama).
And this cluck is certainly both a friendly welcome and an urgent warning, depending on who you are — founder late to the game or VC who jumped in too early.
A high-pitched battle cry
Although late to the generative AI game compared to OpenAI and Google, Meta is the first big tech to their AI model, Llama 2, free and open source.
And not only is it free, Meta claims on its Llama 2 website that it outperformed 11 other open source language models on benchmarks such as “reasoning, coding, proficiency, and knowledge tests”.
However, according to TechCrunch, “Meta says that in a range of benchmarks, Llama 2 models perform slightly worse than…GPT-4 and PaLM 2, with Llama 2 coming significantly behind GPT-4 in computer programming.”
But hey, free is still good.
Just think about Apple’s iOS versus Android. When smartphones were first made, the open source Android as a mobile phone OS was clunky to use for the non-tech savvy and rather slow compared to the smooth as silk proprietary iOS.
But because it was free, every other phone manufacturer — except Microsoft — took to it and modified it for their own products.
Within a decade, although Apple still held the prestige and brand status, there were already far more Android phone users in the world than iPhones. Google Play today is also the world’s largest app store, more than twice that of Apple’s!
Microsoft not making the same mistake three times
An interesting twist to this tale is the fact that Meta is launching Llama 2 on Azure, Microsoft’s cloud service.
Having already invested US$10 billion in OpenAI, isn’t it strange for Microsoft to be a distribution channel for a rival product?
Not at all.
They simply don’t want to make the same mistake a third time.
Microsoft tried making its own smart phone OS. It failed miserably.
When open source rose up and SaaS became the norm, Microsoft held on to proprietary software and was also slow to introduce web-based productivity tools.
But because it kept its hold on the enterprise market, the company wasn’t crushed.
Nonetheless, it wasn’t going to be slow to the next big tech trend — Generative AI.
So by aligning itself with one of ChatGPT’s biggest threats, it is hedging its own bet. It is also helping to boost the attractiveness of its cloud service — which has stayed stuck in a distant third place behind AWS and Google Cloud since forever.
VCs too fast, too furious?
But for the VCs who jumped headlong into the hordes of generative AI startups that sprouted up like mushrooms after a rainy day, this particular Llama’s cluck today could be a big warning siren.
Zuckerberg is betting that making their AI model open source will help them play catch up in developer adoption, crowdsource improvements and generate use cases for widespread consumer adoption.
This could be a repeat of the iOS versus Android story.
And if that happens, it would mean that the fledging Generative AI startups receiving billions from VCs today would be facing tough competition tomorrow. More than 80% of them supposedly use OpenAI’s paid model.
Google knows this.
In a leaked internal memo two months ago, a Google researcher concluded that the company has “no moat” in the industry, and “neither does OpenAI”, because of the threat of open source models.
History will prove if he is right, but right now the situation reminds me a lot of the crypto craze. More than one VC had admitted to me that they cannot rationalize the valuation decisions for crypto startups and have to just go along with the flow.
These are also the same VCs who literally just seconds ago, were talking about owning your core technology and having a protective moat in order to attract their funds.
With so many Generative AI startups today dependent on somebody else’s AI model and paid API service to even exist, anything free and available for complete customization and modification is a major game changer.
The plot thickens.