I got a bit carried away with the amount of stories in this edition, but there's so much good stuff that it’s hard to pick what to include…

View in browser 

AI'm Informed

16.11.2023

I got a bit carried away with the amount of stories in this edition, but there's so much good stuff that it’s hard to pick what to include. Enjoy!

• “It’s really disappointing to see how cavalier we’ve become about normalizing surveillance,” say the opponents, “It's allowed us to [...] be very effective in our investigation,” say the proponents of Fusus, an AI-powered surveillance system that’s quickly spreading across the US and beyond. - 404 Media

 

• The new HGX H200 AI computing platform from Nvidia is coming to system manufacturers and cloud providers in Q2 2024, promising a sizable performance leap, for example doubling the inference speed on Llama 2 70B compared to the H100. - Nvidia

 

• Two engineers at Hugging Face, working remotely from Europe and with support from the company’s other teams, develop tools and ‘recipes’ to help builders create ChatGPT-like experiences based on open-source models, including Mistral and Falcon. - HuggingFace, TechCrunch

 

• Help OpenAI steer the future of AI by providing unique data for training models – for example, they’ve partnered with the Icelandic Government to improve GPT-4’s ability to speak Icelandic, or the Free Law Project with a large collection of legal documents. - OpenAI

 

• Following CEO Jamie Dimon’s statement that AI can be integrated into “every single process” at JPMorgan, they’re jumping ahead of the competition and already working with US regulators to ensure proper controls for their genAI pilot projects – an app that generates earnings summaries and an AI helpdesk. - Bloomberg

 

• The cost of training an AI model from scratch depends on the compute requirements, and the required length of training – this article is a great in-depth look at all the factors that contribute to these requirements. - Brev.dev

 

• Building GPTs is an exciting new avenue, here are some invaluable lessons from engineer Simon Willison’s attempts – and a crucial observation that it’s difficult to get good results from using your own knowledge to empower your GPT. - Simon Willison

 

• Wish you could ship new features as quickly as OpenAI? Here are some of the secrets that lead to such achievements as having to pause new ChatGPT Plus subscriptions due to huge demand. - Pragmatic Engineer

 

• Google DeepMind’s new GraphCast model for weather forecasting makes 10-day forecasts in less than a minute on a single TPU machine (compared to hours with a conventional approach on a supercomputer with hundreds of machines), and outperforms the conventional approach on 90% of test variables. - DeepMind

 

Thanks for reading!

I'd like to reach more people with this newsletter, because I believe it's important for all of us to stay informed on where AI is headed. If you'd like to support my mission, share this newsletter on LinkedIn.

kuba filipowski
Kuba Filipowski
CEO and Co-founder at Netguru
LinkedIn
netguru-sign-rgb-01

Netguru S.A., Małe Garbary 9, Poznań, Polska 61-740, Poland

Unsubscribe Manage preferences