Meta AI – The Long Road

Llama put out to pasture.

• Having completely ceded the open source to China, Meta’s Muse Spark is an attempt to reassert a leadership position in AI, but the model remains unremarkable and there is a long way to go before Meta could be considered to be a leader once again. .
• Meta Muse Spark is the first release since the shake up precipitated by the disappointment of Llama 4 and apart from a good showing benchmarks there is little that is remarkable here.
• The product press release dressed up as a scientific paper describes a model for which no data is given but from which one can infer that it is probably somewhere between 100bn and 500bn in terms of parameters.
• Larger models are in development, but even this smaller “thinking” version scores extremely well on the usual benchmark tests (see here).
• This immediately makes me suspicious that the model has been fine-tuned to beat benchmarks rather than execute real-world tasks, but this will become obvious once the model is generally available.
• This also implies that Llama has been consigned to the dustbin of history, leaving Chinese models and Mistral as the only realistic open source option.
• In a sensible move, the model has also been fine-tuned to address tasks that are most often used when interacting with social media, which makes complete sense as this is how Meta’s AI will be monetised.
• Unlike its competitors, it is not making the AI generally available for a fee, meaning that the hundreds of billions of dollars that are being invested will have to be recouped through growth in the monetisation of its existing ecosystem of apps.
• This is why it has included the ability to offer healthcare tips and a shopping mode to facilitate price comparisons.
• This would appear to be Amazon’s backyard, but Amazon is also struggling with its AI, and so this could be an area where Meta could make some progress.
• The other area where Meta is focusing is on smart glasses, where it is by far the dominant provider of hardware, and the advent of AI has improved the user experience so much that AI glasses are selling like hot cakes.
• There is plenty of competition coming in this space, and here, Google’s offering may gain some traction thanks to the inclusion of the well-known and well-liked Google ecosystem of apps in the glasses themselves.
• This is a big challenge for Meta, as almost everyone I know would prefer to get walking directions to their glasses using Google Maps as opposed to Meta’s navigation service.
• However, Google has proven to be fairly inept at making hardware, which gives Meta some breathing space, but its deepening ties with Samsung could help fix this problem.
• The net result is inconclusive as the model is not blowing the doors off, but to be fair to Meta, it was never intended to do so.
• Instead, this is the first stage of the reset, and the promise is that there is much more to come.
• Given the amount of money being spent here, there needs to be an awful lot more, and I am not convinced that it will be enough to justify the investment.
• This is why I remain cautious on both Meta’s outlook in AI and on the valuation of its shares, where shareholder funds are being poured down an AI-related black hole.
• There are much more interesting places to invest.