I think modern-day LLMs, are the “78 records” of AI, down the road, they will become much more powerful.
That is an interesting thought, but there are currently LLMs which are incredibly powerful and sophisticated. OpenAI’s GPT 5 cost over $500 Million to develop. $Hundreds of Millions have been spent by Google on various Gemini versions including $192 Million spent on Gemini Ultra 1.0. Both of them were surpassed by Claude Opus 4.6 from Anthropic. We do not know how much was spent on that model, but Anthropic is expected to spend $12 Billion in development costs in 2026 and $23 Billion in 2027. Anthropic is expected to lose $5-7 billion in 2026 and likely more in 2027.
I don't know if with this amount of money being spent if comparing current LLMs are a good comparison to 78RPM records. A bunch of these companies are making bets that likely will not pay off and could easily result in the company's total failure. What I am wondering is how much money will the Federal Government be willing to spend to keep these companies from failing.
As an interesting sidenote... according to the latest version of Gemini about $20 Billion in 2024 dollars were spent producing and manufacturing 78 records from 1898 to the late 1950s this includes all expenses including record company executives to janitors, to all of the performers. Who knows how accurate that is.