I would love to know what the increased token count is across these models for the benchmarks. I find the models continue to get better but as they do their token usage also does. Aka is model doing better or reasoning for longer?
It's pretty wild that Brave's terms of service state as much, considering their search API is entirely derived from storing the results of other search systems. https://support.brave.app/hc/en-us/articles/4409406835469-Wh.... Aka Brave is blocking exactly what it does to Bing and Google.
I think you would be very surprised by the number of customers who don't care if the embeddings are SOTA. For every Joe who wants to talk GraphRAG + MTEB + CMTEB and adaptive rag there are 50 who just want whatever IT/prodsec has approved
It's not free because it's cheap for them to run. It's free because they are burning that late-stage VC dollars. Despite what you might believe if you only follow them on twitter the biggest input to their product, aka a search index, is mostly based on brave/bing/serpAPI and those numbers are pretty tight. Big expectations for ads will determine what the company does.
This works better because it gives a secondary set of conditions for which the decoder (generating text) is conditioning its generation. Assume instead of their demo you are doing Speech2Text for Oncologists. Out of the Box Whisper is terrible because the words are new and rare, especially in YouTube videos. If you just run ASR through it and run NER, it will generate regular words over cancer names. Instead, if you condition generation on topical entities the generation space is constrained and performance will improve. Especially when you can tell the model what all the drug names are because you have a list (https://www.cancerresearchuk.org/about-cancer/treatment/drug...)
Not mentioned in their blog posts but on the model cards on huggingface:
"Molmo 72B is based on Qwen2-72B and uses OpenAI CLIP as vision backbone. Molmo-72B achieves the highest academic benchmark score and ranks second on human evaluation, just slightly behind GPT-4o." Others are based on Qwen 7B. What happened to the Olmo chain?
I feel like I have been hearing that since V1 TPU. I think Google is the perfect solution because they are teams whose job is to take a model and TPUify it. Elsewhere there is no team, so it's no fun.
Great paper! I have a question on how much this can be attributed to portions of text which GPT has already seen. Like if the same approach was followed on a new sliver of Wikipedia which GPT had not seen would the results be the same?
Most other search engines train with a target of google or with some form of reward which is bootstrapped on google rankings. It makes Bing results implicitly have the same behavior as Google. DDG and others just use BingAPI so googles incentives pass on through.
See. When you have to convince corporate lawyers and security folks that whole switching licenses makes them uneasy. BD and legal are much happier to deal with Meta than the UAE.