Home

학습 콘텐츠

Why did #Anthropic decide not to release #Mythos, its most powerful #AI model yet? #cybersecurity - Video học tiếng Anh

Why did #Anthropic decide not to release #Mythos, its most powerful #AI model yet? #cybersecurity

학습 모드 선택:

Highlight:

3000 Oxford Words4000 IELTS Words5000 Oxford Words3000 Common Words1000 TOEIC Words5000 TOEFL Words

자막 (55)

0:00Anthropic just pulled off one of the

0:02most unusual moves in AI history. The

0:04company built what may be its most

0:06powerful AI model ever called Mythos and

0:08then chose not to release it. So what's

0:10really going on? There are three main

0:13explanations. First, the official

0:15reason. Mythos is extraordinarily

0:17capable in cyber security. The model

0:19found thousands of serious software

0:22vulnerabilities, including flaws in

0:23major operating systems and browsers

0:26that humans had missed for years. That

0:28led Anthropic to launch project

0:30Glasswing, which gives trusted partners

0:32access to the model so they can patch

0:34systems before these capabilities become

0:36widely available. The AI security

0:39institute found that OpenAI's GPT 5.5

0:42model is similarly capable. So this

0:44undermines the official stance a bit.

0:46Second, the compute theory. Mythos is a

0:49huge expensive model to run. Anthropic

0:51has been dealing with massive demand for

0:53Claude and has openly talked about

0:55expanding its compute capacity. Some

0:58experts have argued that Anthropic

1:00simply didn't have enough infrastructure

1:01to launch Mythos at scale. Now, after a

1:04string of new compute deals, Axios

1:06reports Anthropic is preparing to

1:09release Mythos level models more

1:10broadly. Coincidence? I will let you

1:13decide. Okay. Third, the competitive

1:16theory. AI companies increasingly worry

1:18about something called distillation.

1:20When a Frontier model is released,

1:22rivals can collect its outputs and use

1:25that data to improve their own systems.

1:27Anthropic may simply want to keep its

1:29best capabilities out of competitors

1:31hands for as long as possible,

1:33especially from open- source rivals and

1:35fastmoving Chinese AI labs. The bigger

1:38takeaway is this. For years, the AI race

1:40was all about releasing bigger and

1:42better models as fast as possible.

1:44Mythos suggests the next phase may be

1:47different. The future may be less about

1:49who builds the smartest model and more

1:52about who can deliver these products the

1:54most efficiently to the biggest

1:56audience. To read more about anthropic

1:58mythos and the next phase of the AI

2:01race, head to Business Insider.