📆 May 24th 2024📆 

The last few weeks have been awash with conferences, scandals and mistakes stealing the headlines. Today, we focus back into expert opinions and the shape of the industry.

• Jann LeCun was speaking this week about the future vision for AI and how it’s not all about LLMs.

• Anthropic report progress on solving the black box problem which brings us a step closer to understandable AI.

• Humane, once lofted up as the next product revolution, has expressed their desire to find a buyer. Does this mark the top of the hype cycle? Will we see similar news about other LLM based start-ups?

• Check out the Other News Rollup section for a quickfire review of other top stories in AI.

Let’s dive in for all the details!

💫Want to Support my Work?💫

I really enjoy researching and writing about AI - help me to keep delivering better and more expansive content.

• Buy Me a Coffee ☕️- AI is really fuelled by coffee - keep the tank filled

• Spread the Word 🗞️- Share on social or directly with your friends

• Give me Feedback 💬 - Tell me what’s important to you

📰 News 📰 

AI Pioneer LeCun's Advice on Next-Gen AI Systems

There is no doubt that the current hype cycle has been driven by LLMs, the transformer architecture and OpenAI’s release of ChatGPT about 2 years ago. But this is only one strand of AI capabilities. I’m not discounting the vital role that LLMs (more specifically multi modal LLMs) will play but I believe these will eventually be more the interface than the core component of an AGI system.

Yann LeCun was recently advising AI developers to focus on next-gen AI systems instead of LLMs. He believes LLMs have limitations and suggests exploring alternative AI approaches. LeCun also criticizes LLMs for not reaching human-level intelligence and advocates for objective-driven AI. Speaking at VivaTech in Paris, his words have sparked debate about the role of LLMs continuing to dominate the AI landscape. As above however, I agree with him. It’s naïve to say that a single technology will be “the technology” and we should not forget the diverse set of AI paths ahead.

Cracking the AI Black Box: Anthropic's Research Announced

The Black Box problem is not a new one. When we run a trained machine learning model we are basically giving it input and receiving an answer. While we are responsible for training the model, explaining exactly why and with what criteria it arrived at the answer is very difficult. This is an important element of AI ethics and trust, so labs like Anthropic continuing to research and develop new techniques help move the industry forward. Breakthrough help make AI more trustworthy and integrated into our daily lives.

Anthropic researchers have made progress in understanding artificial neural networks, particularly large language models like Claude, by identifying specific features encoded into the model. By manipulating these features, they aim to make LLMs safer and reduce bias. While their work is promising, it does not claim to have fully solved the black box problem, and there are limitations to their approach.

Their findings to date can be found here

Humane AI Pin Maker Seeks Buyer

Perhaps too early to say but is this a worrying sign for the current cohort of start-ups? Humane is one of a number of companies basing their success off LLM advancements. While promising, the technology still has it’s challenges when it comes to hallucinations and resource needs. On top of all that, we regularly see announcements from the big tech giants that wash over a start-up’s moat. This is a reversal of the traditional trends where the small and nimble start-up can move faster and be more disruptive to the slow moving giants.

Despite raising $230 million from investors, Humane’s AI Pin received poor reviews, with issues like slow response times, irrelevant answers, and poor battery life. It’s still early days but definitely a space to watch as we move past the peak of the hype cycle.

Read More At: Engadget

🔨 AI Powered Tools 🔨 

• Adobe Firefly - Generative AI tool designed for creatives to easily produce high-quality content across various media, including images, videos, and 3D objects. Integrated into Adobe's Creative Cloud, Firefly enhances creative workflows with text-based editing and generation.

• Scale - Accelerates the development of AI applications by providing high-quality data labelling and model training services. It serves industries such as autonomous vehicles, e-commerce, and finance.

• Stable Artisan - Generative AI tool available as a Discord bot, enabling users to create and edit images using Stable Diffusion 3. It offers features like image generation, background removal, and video creation.

🗞️ Other News Rollup 🗞️ 

Hollywood Partnerships - Meta and Google offer millions to Hollywood for AI video licensing, aiming to advance video generation tools.

OpenAI Training Data Deal - OpenAI partners with News Corp for $250 million deal to train ChatGPT AI models with news content.

Smart Glasses Update - Ray-Ban Meta smart glasses now share photos on Instagram and play Amazon Music hands-free. New AI features added.

Nvidia's GenAI Success - Nvidia sees explosive growth in GenAI start-ups, faces chip supply challenges, and expands into ethernet networking for AI.

AI Deployment Made Easy - Dell Enterprise Hub simplifies on-premise AI model training and deployment, offering optimized open models for Dell platforms.

Advanced Multimodal RAG - mmRAG enhances AI assistants with LangChain capabilities, Amazon Bedrock, and fusion techniques for multimodal data processing.

Robocall Scam Fines - FCC fines scammer $6M for using AI to clone Biden's voice in illegal robocalls.

AI Export Deal - US politicians wary of Microsoft exporting AI tech to UAE's G42, fearing tech transfer to China. Related also to….

AI Export Control Bill - US lawmakers aim to regulate AI exports to prevent national security threats, impacting US-China tech competition.

LLM Bug Alert - Critical bug in AI models unaddressed, posing security threats. Lack of vendor accountability and support highlighted.

AI Glitches: Pizza Glue - Google's AI feature suggests adding glue to pizza, showcasing glitches and misinformation, raising concerns about AI accuracy.

Voice Mimicry Drama - Scarlett Johansson claims OpenAI copied her voice for ChatGPT, leading to voice pause and clarification.

Game Voice Generator - Inworld AI launches Inworld Voice, offering 58 voices for game makers with customization options. Cost-efficient tool.

AI Safety Commitments - Worldwide commitments to mitigate AI risks, EU's AI Act, and 'Frontier AI Safety Commitments' for responsible AI development.

Thank You for Subscribing

Enjoying what you’re reading? Help me get better so I can continue to provide you with the most relevant content.