Brandon Hoffman’s Post

aiDAPTIV+ is Phison's groundbreaking solution designed to make large language model (LLM) training accessible and cost-efficient for small and medium-sized companies all while maintaining control of sensitive machine learning (ML) data. Dive into the details in this video: https://lnkd.in/g7nRiy4X #GTC2024 #LLM #AI #Phison 

We want to take a moment to get personal. When we set out to create our AI Mastery course, we had a specific person in mind. Well, actually…three specific people 🙋♀️🙋♂️🙋 See if any of these sound familiar: The business leader who knows AI is the future, but feels like they're playing catch-up in a game with no rulebook The technical expert who's got the AI skills, but struggles to translate their expertise into the business world The curious go-getter who wants to use AI without getting lost in a labyrinth of technicalities Sound like anyone you know? Then make them aware of our AI Mastery https://lnkd.in/eiyUh3BJ

Generalized AI models will only go so far. Everyone has access to the same models, so how will you be different? the real value will come from leveraging data that only you have. the next wave of AI will benefit companies that have a robust data infrastructure that can feed data into smaller AI models effectively and effeciently.

🤖 The AI that tells you how reliable (or not) it is Everybody's heard about Generative AI's hallucinations and generally inaccurate answers. That's why the launch of Cleanlab's Trustworthy Language Model (TLM) is a game-changer. This robust model not only provides high-quality outputs but also quantifies its confidence in each answer. With TLM, we're no longer in the dark about the reliability of AI-generated responses. 🔍 For instance, TLM confidently asserts that 100 + 300 equals 400 with a confidence score of 93.8%. But when asked a more complex calculation, its confidence drops to 24.5%, indicating the need for human verification. This level of transparency is unprecedented and is set to revolutionize how we use Large Language Models for automation and answering questions. 🚀 Currently TLM is in 'Private Beta' mode, but hopefully the approach of showing reliability with LLM answers will soon become a standard. #Cleanlab #AI #Innovation #TrustworthyAI #TLM Image credit: Cleanlab

Companies are spending a fortune on AI solutions and tech, DBMT just published a new White Paper on how to get the most out of AI/LLM solutions with a Knowledge Domain - bridging the gap between Language and Knowledge. This is a must read: https://lnkd.in/efXC37b5

This is a helpful animated article in the FT explaining how LLMs like GPT-4 work in simple terms. However, one of the first comments was someone who posted the article to GPT-4 which was quick to highlight two main problems with LLMs... "...the article fails to address the issue of bias present in generative AI models... ...which can have harmful consequences when applied in real-world contexts such as finance, law, and public services." "...the article briefly mentions that LLMs can fabricate information and engage in "hallucination." However, it does not adequately delve into the potential ethical implications of this capability. Fabricated information generated by LLMs can mislead users and have serious consequences, especially in areas such as legal cases or journalistic reporting where accuracy and factuality are crucial." This is why Rainbird exists. Article here 👉 https://lnkd.in/evnm7uEQ Rainbird video here 👉 https://lnkd.in/eVwudi4X #GPT4, #LLM #GenerativeAI #Explainable AI #Trust #Rainbird

We must carefully consider the psychological impact of AI on individuals, as it can influence emotions and behaviors ! Including the ethical concerns is a must to promote a responsible AI use. #ResponsibleAI

Values, values, values. I cannot agree more with this article in emphasizing and knowing what you stand for before you chase AI. AI and wavering principles do not go together.

Indefinitely and in every conceivable possible path.

In the current world of rapid development, AI is transforming our interaction with each industry. Therefore, ensuring that our devices mirror the values of the humans behind them is imperative. This article is a comprehensive guide for all organisations aiming to create ethical AI services.  It highlights some categories of challenges experienced while creating a values-aligned AI and possible solutions. 1. Defining the Values: Most AI models incorporate established principles from organisations like the UN. The other alternative is to conduct internal research within the organisation and produce one's own values. 2. Writing the Values: Diverse datasets and ongoing fairness checks are crucial to ensure your code reflects your values and doesn't perpetuate biases. 3. Trade-offs: Ethical considerations must be weighed against business goals. One approach? Segmenting the market to align solutions with specific needs. 4. Third-party AI Models: Partnering with reputable vendors and conducting thorough checks are vital when integrating pre-built AI models. 5. Regular Human Feedback: Building a feedback mechanism to report all the bugs is necessary. Eventually, the goal is to create safe and ethical AI that is efficient and can be trusted by its users. I recommend checking out the full article linked below.