Anthropic developed a super powerful artificial intelligence model and named it Model 2. Based on available data, it exceeds the capabilities of its flagship Mythos product; the company has no plans to slow down its development, but it won’t release it to the public either.
Image source: anthropopic.com
The company insists that the risk of serious negative consequences associated with the operation of the anthropic model remains low, but not as low as it was when the last report was published. The company was forced to raise its overall assessment of threats related to model loss in a pessimistic scenario from “very low” to “low,” citing recent cybersecurity incidents.
The offline R&D capabilities of these models show signs of rapid growth, which contributes to technological advancement, but can also be misused. “As part of our standard R&D process, we perform internal training and evaluation on many different experimental versions of the model that we do not intend to release. Model 2 is one of them”– the company told the publication Axios.
Anthropic points out that model 2 shows “Significantly improved” among many internal tasks. Along with Mythos 5, it is “actively” used by company employees to write code, manage agents, and generate content, although the jump in quality of results is not the same as during the transition from Opus 4.6 to Mythos. “We have no current plans to release the model for external use.”the company noted. It appears that Anthropic is increasingly having difficulty understanding the capabilities of its own models and assessing threats related to its work.
If you find an error, select it with your mouse and press CTRL+ENTER.
