________________________________________________________
AIM Daily XO by Amit Naik
Monday, Mar 18, 2024 | Was this email forwarded to you? Sign up here
_________________________________________________________________
While Elon Musk has kept his promise of open sourcing Grok, the developer community is reacting with mixed feelings. There are questions on his motives behind releasing it and concerns regarding the nature of this open-source initiative, as it lacks the training dataset and detailed methodologies.
In a recent blog, xAI announced the launch of the base model checkpoint from the Grok-1 pre-training phase, which concluded in October 2023. “This means that the model is not fine-tuned for any specific application, such as dialogue,” read the post.
“We should just call it open-weight models at this point,” said a user named Swalsh on HackerNews. While the models’ weights (the parameters developed during the training process) are made available, there is a lack of transparency regarding the training data.
As per the Grok-01 model card released last year for its fine-tuned model (which powers Grok on X): “The training data used for the release version of Grok-1 comes from both the Internet up to Q3 2023 and the data provided by our AI Tutors.”
Compared with Meta’s Llama and Mistral, which seem to have thorough documentation and deployment code with instructions, Grok appears to have just released a raw base model from the Grok-01 pre-training phase. This is as good as not releasing it in the first place.
On the bright side, Grok-1 is just a large model (~314B), which matches GPT-3.5 released two years ago, and is about the same level as much smaller models like Mixtral (~47B) and Qwen (~72B).
Clearly, it was released only to take a dig at OpenAI. But, the move seems to be backfiring as the folks at OpenAI are taking the weight seriously and quite literally.
However, there are a few early enthusiasts experimenting with fine-tuning Grok – especially the GPU-rich folks like Perplexity AI and others.
Read: Elon Musk’s xAI Open Sources Grok
The Most Frustrating Programming Language
“English is the hottest new programming language” – nothing irks a software developer more than this latest quip in tech circles. But apart from English, several programming languages also frustrate software engineers and developers. Some earn the ire for their syntax and style, while some for the community built around them. Read the full story to find out.
Decoding BharatGPT
BharatGPT is all about bringing India (Bharat) to a leadership position in tech. “My key motivation behind BharatGPT is to not lose this ideological or civilisational war in the tech space,” said IIM Indore professor Aditya Maheshwari, in an exclusive interaction with AIM.
Get to know more about BharatGPT here.
Healthify is Hiring
About two-and-a-half months ago, Bengaluru-based health and fitness startup Healthify (formerly HealthifyMe) introduced Ria 2.0, a multimodal generative AI-powered virtual health coach with multilingual conversational capabilities.
To strengthen its products, Healthify is actively looking for talented folks for various roles, including senior AI engineer, machine learning engineer, business analyst, and senior business analyst.
Check out the interview process here.
_________________________________________________________
Welcome to AIM Daily XO, where AI's newest tales come alive. Every day, we send out an email packed with the top four AI stories, each one handpicked to spark curiosity and inspire. Crafted by a fresh voice daily, our stories blend insight with a touch of fun, making sure you're always in for a delightful read. Dive into the heart of AI with us, where every email is a new adventure.
AIM Daily XO: Your daily dose of AI, served with a side of creativity.
_________________________________________________________________
Комментариев нет:
Отправить комментарий
Примечание. Отправлять комментарии могут только участники этого блога.