Meta Introduces Self-Rewarding Language Model Capable of GPT-4 Level Performance

[alimehdi242]

https://digialps.com/meta-research-introduces-revolutionary-self-rewarding-language-models-capable-of-gpt-4-level-performance/

[extheat]

The webpage here seems more like promotional blogspam. The paper has been on arxiv for 2 days now.

Previously: https://news.ycombinator.com/item?id=39051279, https://news.ycombinator.com/item?id=39061399

[dang → extheat]

Comments moved to https://news.ycombinator.com/item?id=39051279. Thanks!

[Intralexical → extheat]

The webpage in question:

> WE DEVELOP DIGITAL FUTURE THROUGH DIGITAL PRODUCTS

> WE ARE A TEAM FULL OF ENERGY, PASSION AND ENTHUSIASM IN WEB DEVELOPMENT AND WEB DESIGN.

> WE ARE VERY EXCITED TO HELP YOU THROUGH OUR WEBSITE DEVELOPMENT SKILLS.

The two Arxiv submissions had 7 points and 0 substantive comments between them, before the comments on this post were moved.

I guess blogspam works. Microcosmic, maybe.

[d3m0t3p → extheat]

indeed, and the op also mostly posted about this website previously.

[brucethemoose2]

It's a llama 70B finetune using an interesting new technique, not a GPT4 beating foundational model from Meta.

The headline is so misleading, I am flagging this.

I am seeing way to many of these on HN in general, and I think this is the straw that broke the camel's back. I am going to start flagging misleading headlines much more liberally, even if its the blog's actual title or whatever.