Even without true RAG, it’s pretty darn good. Usually have to prompt 2-3 times and send more details, but also my use cases were pretty easy. What I thought was interesting was that in every use, I did NOT mention the product brand or model number. I just said “fix this thing”
LLMS
-
Attention as Pooling: Superior to Max Pooling in Neural Networks
By
–
Most people don't understand that attention is a pooling operator. A much better (more FLOPs) and less hard-coded and peaky compared to "max" operator in the traditional max pooling, because of soft-attention with learned matmuls and dot products. However, attention and pooling
-
Groq API Access Available via Console with Paid Tier Coming
By
–
API access is available through http://
console.groq.com
If you're asking for paid tier rates, that's coming next month -
Groq acknowledges benchmark analysis from Artificial Analysis
By
–
Thanks for the benchmark @ArtificialAnlys !
-
Hugging Face Idefics2: 8B Vision-Language Model Revolution
By
–
Hugging Face Presents Idefics2: An 8B Vision-Language Model Revolution: Hugging Face’s latest offering, Idefics2 heralds a new era in multimodal AI models. With enhanced… https://
analyticsvidhya.com/blog/2024/04/h
ugging-face-presents-idefics-a-vision-language-model-revolution/?utm_source=dlvr.it&utm_medium=twitter
… #DataAnalytics #DataScience #DataDriven #CTO #IoT #AI #ITDirector #BigDataAnalytics -
Implicit Bias Revealed in GPT-4o and Latest LLMs
By
–
Many benchmarks today fail to observe bias in GPT-4o. But are such models truly unbiased? This new paper reveals implicit bias in the latest LLMs such as GPT-4o and LLama Author @baixuechunzi is on alphaXiv answering questions about her latest work: https://
alphaxiv.org/abs/2402.04105
v2
… -
Long Context Window Progress: From 65k to 1M Tokens in One Year
By
–
quite incredible to see the goalposts of long context move in the past 1 year. May 2023: asking @jefrankle and @abhi_venigalla about their 65k+ "Llongboi" model May 2024: @markatgradient casually extending Llama 3 to >1m tokens with ~perfect NIAH and mainstream ai engineers
-
Mini Model Surpasses Mixtral on MMLU Benchmark Score
By
–
Soooo what do you think about this MMLU score? Specifically for the Mini model, which is higher (68.8) than Mixtral's (68.4).
-
Using AI voice models for article summarization
By
–
After the next voice model drop: use AI to read a summarised article