Neoteric AI News Digest 12: New Season, New AI Breakthroughs

Neoteric

Let's turn your ideas into a well-crafted digital product 🚀

Published Oct 9, 2024

he summer of new AI models may have passed, but it already feels like the new season won’t disappoint. It barely started, and we’re already celebrating Geoffrey E. Hinton’s Nobel prize — it must be a sign that this fall will bring even more exciting news from the AI world!

As the big news about Mr Hinton’s Nobel prize can already be found on every cover, we’ll do the usual and serve you with a set of important and exciting, but possibly less buzzy news — in case they skipped your attention.

Without further ado, dive into the new issue of Neoteric AI News Digest and catch up on the noteworthy events of the past weeks in the AI world.

OpenAI's o1 Model Shakes Up AI Regulation Conversations

Let’s start today’s issue with OpenAI's latest model, o1, which has shifted the conversation around AI regulation. Touted as a “reasoning” model, o1 takes more time to process tasks, showing improvements in areas like physics and math, even without significantly more compute power than its predecessor, GPT-4o. This raises questions about how we evaluate AI risk, particularly in light of regulatory frameworks like California's SB 1047, which ties safety requirements to compute thresholds.

The o1 model underscores that scaling compute isn’t the only path to improving AI performance, potentially challenging policies that assume bigger equals more powerful or riskier. Nvidia’s Jim Fan suggests that future AI systems may rely on smaller, reasoning-focused cores rather than the compute-heavy architectures that dominate today. This could render current regulations less relevant to emerging AI models.

Sara Hooker from Cohere adds that model size or compute power alone shouldn’t be the sole metrics for determining risk. However, most AI regulations, including California’s, were built to adapt over time, with room for adjustments as the technology evolves. The challenge now is finding better ways to assess risk in AI models that don’t follow traditional scaling patterns.

California’s Battle Against AI Deepfakes Faces Legal Setbacks

As we’re on regulations — California recently passed a series of laws aimed at combating AI-generated deepfakes in political content. Governor Gavin Newsom signed the key law, AB 2839, following the viral spread of an AI-altered video of Vice President Kamala Harris. The deepfake, which manipulated her voice to make it sound like she called herself an incompetent candidate, sparked significant concern about the dangers of AI in elections. Newsom’s new law directly targets individuals who knowingly distribute misleading AI-generated content during election periods, with penalties ranging from takedowns to financial fines.

However, the law quickly ran into obstacles. A federal judge blocked AB 2839 after a lawsuit argued that the deepfake in question was satire protected by the First Amendment. The judge ruled that the law could chill free speech, particularly in political content like parody and critique. The case, centered on this Harris deepfake, has put AB 2839 on hold, and its future remains uncertain.

To learn more about this case, read the articles from Cointelegraph and TechCrunch.

While Newsom advocates for stronger regulation of AI in election-related matters, he remains cautious about over-regulating legislation. He recently vetoed SB 1047, a controversial bill that proposed mandatory safety testing and kill-switch requirements for large AI models. Newsom argued that while the bill had good intentions, it could stifle innovation in California’s thriving AI sector without effectively addressing the most critical risks posed by the technology. His decision illustrates the balance California is trying to strike between innovation and regulation as AI continues to evolve.

Wanna know more? Here’s a full article from Cointelegraph.

Is Nvidia’s Bold New AI Model a Game Changer?

While California tackles AI regulation challenges, Nvidia has taken a big step in AI innovation by releasing its NVLM 1.0 model, a powerful open-source competitor to proprietary systems like OpenAI's GPTs. The 72-billion-parameter model, NVLM-D-72B, closely competes with the leading LLMs such as Llama 3 70B, Claude 3.5 Sonnet, GPT-4o, and Gemini 1.5 Pro — often performing nearly on par, and even surpassing them in certain evaluations. Notably, it consistently beats GPT-4 on most key benchmarks.

Nvidia’s decision to make the model open-source breaks with the trend of keeping advanced AI systems closed, granting unprecedented access to cutting-edge AI technology. This move could accelerate research and development across the AI community, allowing smaller teams to compete with industry giants. However, concerns around misuse and ethical risks are already surfacing, with debates about how to balance innovation with responsible use.

This release could pressure other tech leaders to follow suit, pushing the AI industry (yet again) toward more open collaboration. Nvidia’s open-source model marks a significant moment in AI’s evolution, with far-reaching implications for the future.

For more, read the full article on VentureBeat and check Nvidia’s paper on the new model’s capabilities.

KoBold Metals Raises $491M to Power the Energy Transition with AI

Speaking of AI innovation, did you hear about the KoBold Metals’ success? The startup, which uses artificial intelligence to discover critical minerals like copper, lithium, nickel, and cobalt, has raised $491 million in a funding round, with the aim of reaching $527 million. This infusion comes as KoBold is making waves with its recent discovery of what may be one of the largest high-grade copper deposits ever found.

KoBold’s AI-driven approach allows it to sift through vast amounts of geological data to identify promising mineral deposits, a historically high-risk endeavor. The company’s success in Zambia with its massive copper find is a testament to the potential of AI in transforming the energy transition. With around 60 other exploration projects in progress, KoBold plans to develop the Zambian resource itself, marking a strategic shift as the company moves beyond discovery into development.

Investors like Bill Gates, Jeff Bezos, and Jack Ma have backed KoBold’s vision of using AI to accelerate the energy transition, and with a potential valuation of $2 billion in this round, the company seems well on its way.

For more on this, check out the full article on TechCrunch.

AI Joins the Fight Against Cancer: The $40M Cancer AI Alliance

Hungry for more good news? The growing influence of AI in healthcare has taken another significant step forward with the formation of the Cancer AI Alliance (CAIA). Backed by $40 million in cash and resources from major tech players, this alliance brings together the expertise of renowned medical institutions like Fred Hutchinson, Johns Hopkins, Dana Farber, and Sloan Kettering to push the boundaries of precision medicine.

The Cancer AI Alliance aims to solve a critical issue in cancer research: the difficulty of sharing data across institutions due to regulations, data formats, and privacy concerns. By leveraging federated learning, a secure AI-driven method of collaboration, these institutions can collectively train models on private data without the need for raw data sharing. The ultimate goal is to speed up breakthroughs in cancer treatment by enabling more efficient use of data between research centers.

With tech giants like Microsoft, AWS, Nvidia, and Deloitte contributing resources and expertise, the alliance is expected to be operational by the end of 2024 and aims to produce its first actionable insights by 2025. If successful, this could be a game-changer for cancer treatment. It’s the kind of innovation that shows how powerful AI can be when it’s used for good.

Want more insights? Read the full article on TechCrunch.

Cutting-Edge AI Applications Transforming Healthcare

Building on AI’s role in healthcare, there are even more exciting applications already making an impact. A recent article published on Cointelegraph highlights six inspiring AI innovations that are transforming the healthcare industry. Let’s take a closer look at them!

GE Verisound AI is revolutionizing ultrasound imaging by enabling even non-experts to capture high-quality heart images, allowing for earlier disease detection. Atomwise is using AI to accelerate drug discovery, predicting the efficacy of drug candidates much faster and at a lower cost. In radiology, Behold.ai and Enlitic are using AI to analyze medical images, drastically reducing the time needed for diagnosis and improving accuracy.

Other examples include Voiceitt, an AI-driven solution helping individuals with speech impairments communicate more effectively, and Merative (formerly IBM Watson Health), which uses AI to assist in clinical decision-making, helping doctors personalize treatments and streamline healthcare operations. WELL Health Technologies is also making an impact by automating healthcare delivery, from scheduling appointments to remote patient monitoring.

Amid all the concerns surrounding AI safety, fair use, and its broader societal impact, it’s refreshing to see how AI is making a tangible, positive difference in the healthcare industry. These advancements show the incredible potential AI has to improve lives, offering real solutions that bring us closer to more efficient, personalized, and accessible healthcare.

For more, check out the full article on Cointelegraph.

Bridging Language Gaps with ChatGPT

While AI is driving innovations in healthcare, it's also solving critical communication challenges in public services. One standout example comes from Minnesota, where the Enterprise Translation Office (ETO) is leveraging ChatGPT to bridge language gaps and improve access to government resources for non-English speakers.

With over 20% of the state’s residents speaking languages other than English, including Spanish, Somali, and Hmong, timely and accurate translations are crucial. By incorporating ChatGPT, the ETO has significantly reduced translation turnaround times from weeks to under 48 hours — sometimes as little as two hours for urgent cases.

The new ChatGPT-powered workflow also improves cultural relevance, using custom GPTs to handle language nuances. Since the program's beta launch, it has processed over 3,000 requests and saved Minnesota over $100,000 each month. Additionally, the ETO is now piloting ChatGPT’s voice capabilities for real-time interpretation, which could further expand access to essential services.

Minnesota’s approach demonstrates the positive impact of a well-chosen AI use case, offering faster, more inclusive services to its diverse communities.

For more details, read the full article on OpenAI.

AI’s Growing Power Demand is Straining the U.S. Grid

As AI continues to transform various sectors, it's also creating new challenges. Its rapid rise is pushing the boundaries of the U.S. power grid, as data centers supporting AI operations scramble to secure the electricity needed for massive computing demands.

According to a recent report, tech companies are moving from one market to the next in search of enough power to support facilities that can consume up to a gigawatt each — enough to power an entire city like San Francisco. Some requests are four to five times larger, but many are being told they’ll have to wait years, if not until the next decade, to access the necessary power.

In states like Utah, California, and Virginia, utilities are already rationing power or halting new data center requests due to overburdened grids. This growing demand is triggering debates over how to fund the billions of dollars required for grid upgrades. American Electric Power, for instance, faces the challenge of accommodating data centers that could double electricity demand by 2028, with three New York Cities' worth of requests beyond that date.

The AI boom coincides with other pressures on the grid, such as electric vehicle infrastructure and manufacturing growth spurred by tax incentives. As AI data centers are expected to consume up to 9% of U.S. electricity by 2030, finding ways to balance this demand with grid capacity and investment is becoming critical.

Wanna dive into details? Read the article on The Wall Street Journal.

That’s it for this issue of Neoteric AI News Digest! AI is moving fast, but we’re always here to keep you in the loop with all the latest developments — so be sure to come back in two weeks for another round of essential AI updates and news.

P.S. Looking for a trusted tech partner for your project? We’ve been building AI-powered software since 2017 🚀 Check our AI development services and see how we can help you!