- GenAI360 - Weekly AI News
- Posts
- Anthropic Extracts Praise and Sarcasm, Images That Sound, OpenRLHF for >70B Models
Anthropic Extracts Praise and Sarcasm, Images That Sound, OpenRLHF for >70B Models
Plus, Musk vs. LeCun, and models from Cohere, Mistral, & Huawei
Before we start, share this week's news with a friend or a colleague:
Key Takeaways
xAI raised $6 billion in a funding round while Grok will be updated with multimodal capabilities, Alexa will receive an update that makes it more responsive, and Microsoft is planning to enhance Copilot with automation agents.
When pitted against other models like ChatGPT, Gemini, and Claude, Perplexity was ranked first in a survey carried out by The Wall Street Journal.
Cohere’s Aya-23-35B model outperforms others in multilingual benchmarks like M-MMLU and Huawei’s Arabic LLM has an automatic speech recognition capability of 96% accuracy.
Anthropic’s study showed that sparse autoencoders can extract interpretable, human-like features from Claude 3 Sonnet, even though the process is computationally expensive.
Uni-Mol Docking V2 drastically improves binding pose prediction for drug design, addressing previous model issues and surpassing its predecessor.
Got forwarded this newsletter? Subscribe below👇
The Talk of the Day: AI Godfather vs Chief Twit; Codestral-22B by Mistral
Had a nice Memorial Day? Elon Musk spent it trading zingers with Meta Ai's chief Yann LeCun, as xAI is ramping up its hiring efforts post-raise (LeCun has declined a tongue-in-cheek invitation to work at xAI). We won't go into much detail but this is our favorite part of the exchange.
We were hoping this would be over by the time we finished researching for this post…
Kinda funny given that one of these people made CNNs, although according to the other, Tesla doesn't use them as much.
… but it's still going on.
In other news, Mistral released a 22B parameter model, trained on over 80 programming languages, with better performance than Llama v3 70B or CodeLlama 70B across some benchmark tests.
It's better on average, driven by superiority at Python, Java, PHP
The Latest AI News
While Humane, Rabbit, and Google were under fire last week, xAI and Amazon announced major updates to their AI offerings. Meanwhile, Microsoft also announced a new version of Copilot and plans to build a $1 billion data center in Kenya.
Some new LLM releases also popped up, including an Arabic LLM from Huawei and Cohere’s Aya 23 models.
Which Set of News Should We Start With First? |