You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Working on these next-gen intelligent AIs must be a freaky experience. As Anthropic announces the smartest model ever tested across a range of benchmarks, researchers recall a chilling moment when Claude 3 realized that it was being evaluated.
newatlas.com
Any thoughts or opinions on this new AI would be greatly appreciated, including what companies will be leading in this new model. Thanks
If you’re thinking long term investment opportunity more than ‘how much will this change the world’ - I think we need to see how the patent situation evolves in this space. Those with the large patent lawyer contingent are in a really good spot right now.
For performance, this looks incremental — decent, but incremental. I wish I knew the tests a bit better to determine if this is ‘high school / early college level learning’ that is mostly memorization (advancements) or if there is actually ‘thinking’ advancement here. The fact that you can study for intelligence test shows you how broken the concept of intelligence can be.
The only good thing about the NEW ATLAS articles is they link to actual information at the end. But before that you are exposed to their sensationalistic blather.