- AI For All
- Posts
- AI Leaders Assemble on Capitol Hill
AI Leaders Assemble on Capitol Hill
PLUS: Microsoft Phi-1.5 and Stable Audio
AI For All
September 15, 2023
Hello readers,
Welcome to another edition of This Week in the Future where we’ll be going over the Capitol Hill AI Summit as well as two new models from Microsoft and Stability AI. This week certainly emphasized the parallel trends of increased AI accessibility and growing AI scrutiny and regulation.
As always, thanks for being a subscriber! We hope you enjoy this week’s content — for a video breakdown, check out the episode on YouTube.
Let’s get into it!
Capitol Hill AI Summit
Elon Musk (left) and Alexander Karp (right)
Capitol Hill recently hosted an AI summit, drawing in an ensemble of top CEOs from major tech giants. Among the headliners, Mark Zuckerberg's call stands out: he urges Congress to engage more actively with AI, emphasizing the dual need for fostering innovation and implementing safeguards.
“Congress should engage with AI to support innovation and safeguards. This is an emerging technology, there are important equities to balance here, and the government is ultimately responsible for that.”
The burgeoning AI landscape is complex, with a multitude of stakes to balance. The mantle, it appears, is on the government's shoulders to ensure this equilibrium.
Enter Elon Musk's intriguing proposition: the concept of an 'AI referee.' Musk's vision is clear — companies delving into AI development should be under the watchful eye of an authoritative regulatory body. This 'referee' would guarantee that corporate actions align with the broader interests of public safety.
We're inclined to concur with Musk. In an ideal world, companies would autonomously prioritize public welfare. Yet, history shows us that self-regulation often falls short, underscoring the need for external oversight.
There were some other perspectives too. IBM CEO Arvind Krishna opposed the idea of requiring licenses for AI development, and Google CEO Sundar Pichai expressed support for Washington’s involvement but wasn’t too keen on details.
On a side note, the rising trend of companies signing the White House's AI pledge is noteworthy. Though it lacks punitive measures for breaches, it's certainly becoming a favored PR move in the corporate playbook.
Unlock the Power of Data Through Expert-Led Data and AI Training
DAIN Studios’ training services are designed around real-world evidence that we have gathered through building and implementing data projects across many different sectors in Europe. Companies choose us for their training because of the depth of our experience.
For many, data—and especially AI—give rise to fear, uncertainty, and doubt. At the same time, recruiting for new data roles to plug any knowledge gaps is a challenge in a saturated labor market. This can make internal training a necessity.
With our learning and development game plan, you can foster a data-driven culture in your company. This will boost your organization’s digital transformation, meaning more decisions are based on data—rather than on intuition or feelings.
We offer our trainings in various formats: masterclasses for leaders, mentoring, workshops, and online courses. Read more and contact us here!
Microsoft’s Phi-1.5
Microsoft has unveiled the next iteration of its open-source model phi-1.5, sparking renewed interest in the AI community. Here's what sets it apart:
Quality Over Quantity: Instead of amassing vast internet data, phi-1.5 is nurtured on 'textbook quality' content, stemming from sources such as textbooks and dictionaries. The result? A high-quality, concise dataset, sidestepping the noise of internet archives.
Size & Efficiency: Sporting 1.3 billion parameters, phi-1.5 is modest, especially when juxtaposed against behemoths like Falcon 180B. Yet, its lean parameter set translates to cost-efficient hosting — great for developers who want to run these models locally.
Performance & Usability: Originating from phi-1, this version is enriched with additional NLP synthetic texts. Benchmarked against vital AI tasks, it impressively nears the performance of models under the 10 billion parameter mark.
A Model for Enthusiasts: Untouched by RLHF tuning, phi-1.5 is tailored more for hobbyists and researchers. While our personal tests yielded somewhat canned outputs, it's remarkable how this model runs smoothly on standard GPUs like the 3000 series RTX NVIDIA.
To sum it up, phi-1.5 may be the answer for those eager to delve into large language models without the hefty costs. It stands as a testament to Microsoft's commitment to making AI research more accessible.
Stable Audio: A New Music Generation Model by Stability AI
Stability AI has needed its own ChatGPT, and it might have gotten it with Stable Audio. The website is experiencing so much traffic, you might not be able to use it. So, how does the music sound? Let’s break it down.
Sound Quality: Predominantly suited for textural music genres like ambient and techno, the output is more than passable as background music or white noise. Curious? You can find samples here.
Limitations: The current interface allows for a mere 45-second clip generation, and while it surpasses predecessors like those from Meta in audio quality, it lacks the finesse in crisp instrumentals and vocals.
Behind the Scenes: Borrowing the stable diffusion technique, previously successful in image generation models like Midjourney, Stability AI adapted it for audio. The model was trained on a vast dataset of 800,000 tracks from a stock audio provider.
A Note on 'Boring' Tunes: The slightly monotonous tunes might be a reflection of its training data, given that stock audio often leans towards the generic side, serving primarily as background scores.
For those intrigued by AI-driven music, Stable Audio is a promising glimpse into the future of audio generation.
🔥 Rapid Fire
EY launches AI platform and LLM after $1.4 billion investment
Salesforce announces new Einstein 1 Platform
Anthropic partners with BCG to bring Claude to more enterprises
Coca-Cola has a new AI-generated flavor
Adobe Firefly generative AI tools are now widely available
AssemblyAI releases next-gen speech recognition model
DeepMind is close to releasing Gemini to challenge ChatGPT
Meta sets GPT-4 as the bar for its next open source AI model
🎙️ The AI For All Podcast
This week’s episode featured George Matus, the Founder and CEO of Teal Drones, who discussed the cutting-edge military and enterprise applications of AI drones and the vital role drones will play in shaping the world order as US-China relations continue to strain.
📖 What We’re Reading
This week’s handpicked insight provides data showing how businesses are leveraging generative AI to improve their B2B sales processes so that reps can spend more time doing what they do best — selling.
How GenAI Can Transform B2B Sales (link)
“As generative AI fever sweeps across the industry, it can be difficult to separate its legitimate promise for B2B sales from all the noise. But here is a leading indicator: high-performing sales organizations are 1.9 times more likely to be using AI already than under-performers.”