Alex Ruiz’s Post

Great fireside chat here at LONDON TECH WEEK 2024 (which has a chunky AI content strand). Yoav Shoham (AI21 Labs) spoke about how an LLM is like deploying a car engine in terms of how much needs to be wrapped around the LLM to make it into a car (in order to truly useful in an enterprise environment). He also spoke about the fundamental need for reliability in order to build trust in the use of an LLM-based Generative AI solution. Great questions from Alexandra Ebert at MOSTLY AI.

Ever had an 'Aha!' moment? Had one today during the 'Language Is Code' workshop! Demystified the daunting concept of AI. It was not just enlightening, it was fun too! Check out this highlight.

The future of seismic data processing is here, and it’s powered by AI! Excited to share the most recent PGS feature story which showcases a fascinating First Break paper by PGS experts 👉 @Julien Oukili, @Jon Burren, @Bagher Farmani, @Jyoti Kumar, @Steven Cochran, @Martin Bubner and @Denis Nasyrov Find out how PGS is shaping the future of seismic data processing 💻 and the transformative impact of machine learning on seismic data processing and its real-world applications🌏 https://ow.ly/jsi850QFEuy

At Montoux we're used to working with the many industry specific acronyms that actuaries use in their day to day work. But lately we've been using a new acronym, a lot....RAG. So what is Retrieval-Augmented Generation, aka RAG? "Retrieval-augmented generation (RAG) is a technique for enhancing the accuracy and reliability of generative AI models with facts fetched from external sources" Curious to know more, including how the term Retrieval-Augmented Generation was coined? Check out this great blog post from Nvidia https://lnkd.in/dZ_PhPcF

Next on my reading list. 👇I just enjoy reading different usecases and when things are explained in practice. And hey.. with good cup of coffee of course. 😉 Bernard Marr and Generative AI In Practice 👇

Check out this post about Generative AI!

Buu Lam (F5 DevCentral) catches up with Michelle Balderson (ISSQUARED) at BSides Edmonton on looking at the past to understand the future for things like AI. Connect with Michelle on LinkedIn to get a copy of her presentation!

This is exceptional work. Please take the time to give this a read. "Openness in the AI field matters, not just for model biases, but for the structural biases in the ecosystem. An ongoing problem is that curation by statistics amplifies many of those structural biases. Driven by investments going into the trillions, datasets and AI models, which are too complex or large to be truly understood, are being deployed with a break-neck speed. This is why advocating for dataset transparency is so important if AI systems are ever going to be accountable for their impacts in the world." #aiethics #ai #ethics #dataset #aitraining

"It contains less about how humans see the world than it does about how search engines see the world. It is a dataset that is powerfully shaped by commercial logics." "These statistics tell us less about the composition of the originating dataset - Common Crawl - than they do about the shortcomings of the language detection model the LAION-5B developers chose to use." "This split tells us something specific about the worldview that LAION-5B contains; a perspective that is carried into AI models that are trained on it." "For models trained on LAION-5B, English (and English-speaking culture) is valued more than the other 107 languages combined." "The creators of SAC are transparent about the shortcomings of the set, specifically the fact that the scores were submitted by users who were both WEIRD (Western, Educated, Industrialized, Rich, and Democratic) and developers of AI art, a demographic they describe as leaning toward "nerdy" and "esoteric."" "Furthermore, they admit that most of the ratings in the dataset were submitted by a "handful of users," whose "aesthetic preferences dominate the dataset."" "The top 50 reviewers, responsible for more than 7.5 million total reviews, appear to fall neatly in the WEIRD demographic. Of the 41 users who share location info, 95% are in the US, Canada, or Europe. They are, mostly, middle-aged photography enthusiasts from small American cities." "The concepts of what is and isn't visually appealing can be influenced in outsized ways by the tastes of a very small group of individuals, and the processes that are chosen by dataset creators to curate the datasets." "Openness in the AI field matters, not just for model biases, but for the structural biases in the ecosystem. An ongoing problem is that curation by statistics amplifies many of those structural biases."