FloridaBoi

Basically the cash burn rate for inference is much higher than reported and the media’s focus on training obfuscates it. All of the companies involved hide what the actual revenues are and if they even cover (or are trending towards covering) running costs of the models.

Ed Zitron with FT attempt to derive what those actual numbers are since they’re all over the place but cannot confirm hardly anything. It seems that at best revenues are 2/3 cash burn rate but are more like 1/2 and this is without considering the massive obligations that OpenAi is committing to.

The article gives no indication as to how the model works (other neural networks) so it could be a fluke especially since it is not physics based.

The arstechnica piece has this at the bottom:

“It’s not immediately clear why the GFS performed so poorly this hurricane season,” Lowry wrote. “Some have speculated the lapse in data collection from DOGE-related government cuts this year could have been a contributing factor, but presumably such a factor would have affected other global physics-based models as well, not just the American GFS.”

Scroll through the major events timeline

“If the U.S. doesn’t transition to new energy vehicles quickly,” auto industry expert Michael Dunne of Dunne Insights told Newsweek last month. “Detroit will cede the global market and be reduced to a niche supplier of gas-powered pickup trucks and SUVs.”

The latest research suggests that as the datasets being fed to AI models continue to grow, attacks become easier, not harder.

“As training datasets grow larger, the attack surface for injecting malicious content expands proportionally, while the adversary’s requirements remain nearly constant,” the researchers concluded in their paper.

Pretty good summary and critical analysis of the last 3-6 months of AI mania