Trust: Zuck, Meta, and Llama 4
April 17, 2025
Sorry, no AI used to create this item.
CNET published a very nice article that says to me: “Hey, we don’t trust you.” Navigate to “Meta Llama 4 Benchmarking Confusion: How Good Are the New AI Models?” The write up is like a wimpy version of the old PC Perspective podcast with Ryan Shrout. Before the embrace of Intel’s intellectual blanket, the podcast would raise questions about video card benchmarks. Most of the questions addressed: “Is this video card that fast?” In some cases, yes, the video card benchmarks were close to the real world. In other cases, video card manufacturers did what the butcher on Knoxville Avenue did in 1951. Mr. Wilson put his thumb on the scale. My grandmother watched friendly Mr. Wilson who drove a new Buick in a very, very modest neighborhood, closely. He did not smile as broadly when my grandmother and I would enter the store for a chicken.
Would someone put an AI professional benchmarked to this type of test? Of course not. But the idea has a certain charm. Plus, if the person dies, he was fooling. If the person survives, that individual is definitely a witch. This was a winner method to some enlightened leaders at one time.
The CNET story says about the Zuck’s most recent non-virtual reality investment:
Meta’s Llama 4 models Maverick and Scout are out now, but they might not be the best models on the market.
That’s a good way to say, “Liar, liar, pants on fire.”
The article adds:
the model that Meta actually submitted to the LMArena tests is not the model that is available for people to use now. The model submitted for testing is called “llama-4-maverick-03-26-experimental.” In a footnote on a chart on Llama’s website (not the announcement), in tiny font in the final bullet point, Meta clarifies that the model submitted to LMArena was ‘optimized for conversationality.”
Isn’t this a GenZ way to say, “You put your thumb on the scale, Mr. Wilson”?
Let’s review why one should think about the desire to make something better than it is:
- Meta’s decision is just marketing. Think about the self driving Teslas. Consequences? Not for fibbing.
- The Meta engineers have to deliver good news. Who wants to tell the Zuck that the Llama innovations are like making the VR thing a big winner? Answer: No one who wants to get a bonus and curry favor.
- Meta does not have the ability to distinguish good from bad. The model swap is what Meta is going to do anyway. So why not just use it? No big deal. Is this a moral and ethical dead zone?
What’s interesting is that from my point of view, Meta and the Zuck have a standard operating procedure. I am not sure that aligns with what some people expect. But as long as the revenue flows and meaningful regulation of social media remains a windmill for today’s Don Quixotes, Meta is the best — until another AI leader puts out a quantumly supreme news release.
Stephen E Arnold, April 17, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
