Dit vind je misschien leuk
LLM as a judge has become a dominant way to evaluate how good a model is at solving a task, since it works without a test set and handles cases where answers are not unique. But despite how widely this is used, almost all reported results are highly biased. Excited to share our…
your post challenged me. every one of your points is wrong but i had to think about each for a while :)
their 'superhuman' ai cleverly assigned all the work to non-default streams, which means the correctness test (which waits on all streams) passes, while the profiling timer (which only waits on the default stream) is tricked into reporting a huge speedup
GPT was actually based on ULMFiT, written by a person with no PhD (me).
Johns Hopkins University is excited to announce its new tuition promise program will offer free tuition for undergraduate students from families earning up to $200,000 and tuition plus living expenses for families earning up to $100,000. As a result of the change, students from…
The comments on software patents made me chuckle. After selling both Id Software and Oculus, my continued employment contracts included “will not participate in software patents” clauses, and yet I got asked to reconsider in both cases. It wasn’t a lot of pressure, so I don’t…
The vast gulf between all of our actual daily experience with LLM improvements plateauing, vs what some researchers claim is going on, is pretty interesting. You can't properly understand AI progress without using the models for real work every day.
First Law of @matt_levine: the funniest thing is the thing that will happen
DOGE trying to vibe-refactor COBOL code with AI tools would be the funniest way for the complete collapse of the global economy to happen
My student's comment on that LLMs generating more novel research ideas than humans paper making the rounds I think this says more about NLP researchers than about LLMs Ouch..
To be fair, he has no fucking idea what he is talking about. NVIDIA is a strong WFH culture, and we're doing just fine.
Australia lost one of our medical legends yesterday, with the death of Nobel Laureate Dr Robin Warren. Together with Dr Barry Marshall, they discovered the link between the bacteria Helicobacter Pylori & Peptic Ulcer Disease. It fundamentally changed how we treat ulcer disease.
I am sorry to report that Dr (John) Robin Warren (Nobel Prize in Medicine 2005) died yesterday evening (July 23 rd). Robin was cared for at Brightwater in Perth Western Australia. He had become very frail in recent years and passed away peacefully in the company of his family.
The first rule of data breaches: if it exists in a database on the Internet, it will be stolen. The second rule of data breaches: the service that lost your data will be incredibly vague about exactly what the hackers took, because it’s way worse than you imagine.
NEW: Hackers say they stole 33 million cell phone numbers of users of two-factor app Authy. Twilio (owner of Authy) confirmed "threat actors were able to identify" phone numbers, but didn't say how many. The risk is better tailored phishing attacks. techcrunch.com/2024/07/03/twi…
please stop its physically painful
This entire paper boils down to “clamp importance ratios to 1.0”, which could have been nicely communicated in the abstract. I can only assume that academic publishing has a lot of motivating forces I am happily ignorant of. arxiv.org/abs/1606.02647
It is a very strange feeling to see an aspect of something myself and a bunch of other people were randomly posting about in August show up in a Google DeepMind Paper in November: not-just-memorization.github.io/extracting-tra…
Don't listen to me. I don't understand language model fine tuning. I'm merely the 1st author of the paper "Universal Language Model Fine Tuning", which explained 5 years ago how to fine tune universal language models.
I recommend talking to the model to explore what it can help you best with. Try out how it works for your use case and probe it adversarially. Think of edge cases. Don't rush to hook it up to important infrastructure before you're familiar with how it behaves for your use case.
United States Trends
- 1. #StrangerThings5 268K posts
- 2. Thanksgiving 693K posts
- 3. BYERS 62.3K posts
- 4. robin 97.3K posts
- 5. Afghan 300K posts
- 6. Reed Sheppard 6,334 posts
- 7. Dustin 89.7K posts
- 8. Holly 66.8K posts
- 9. Podz 4,837 posts
- 10. Vecna 62.7K posts
- 11. Jonathan 76.3K posts
- 12. hopper 16.6K posts
- 13. Lucas 84.8K posts
- 14. Erica 18.5K posts
- 15. National Guard 675K posts
- 16. noah schnapp 9,172 posts
- 17. derek 20.2K posts
- 18. Joyce 33.7K posts
- 19. Nancy 69.6K posts
- 20. mike wheeler 9,855 posts
Something went wrong.
Something went wrong.