Commentary on X
@BlackHC We have seen an exponential increase in intelligence per cost This has been a basic consistent principle since the start with GPT-3.5 The author claims to be a "tech investigator", it's like letting a flat earther write about space exploration https://epoch.ai/data-insights/llm-inference-price-trends
12:51 PM · Jul 16, 2026@BlackHC @GaryMarcus Sorry I'll be more direct This article is blatantly distorting its own data and citations and @TheAtlantic should be ashamed of its fact-checking standards
1:45 AM · Jul 16, 2026@BlackHC @GaryMarcus @TheAtlantic And this isn't an one off, this author has been distorting data on the subject for long
3:03 AM · Jul 16, 2026@anko_979 @GaryMarcus @TheAtlantic Agreed. Def shocked that The Atlantic published it as is
1:47 AM · Jul 16, 2026View all
The Atlantic @TheAtlantic says generative AI is "an engineering disaster." I had Claude fact-check all 23 checkable claims against primary sources: 7 check out · 7 need context · 9 don't hold Verdict: the economics hold up. The computer science doesn't (I checked it too) 🧵
8:44 AM · Jul 16, 2026The argument rests on a chart from @EpochAIResearch that serving more tokens costs more But its x-axis is tokens per second PER REQUEST! Costs rise here because serving each user faster forces small batches. More volume at the same speed makes tokens cheaper. That's not it!
8:44 AM · Jul 16, 2026