The world may never know just what happened at OpenAI during last weeks whirlwind.

After the headline-grabbing drama, CEO Sam Altman was reinstated without a board seat; the companys chief scientific officer, Ilya Sutskever, returned to his post; and the nonprofits board of directors was given a proper shakeup.

But what was behind it all?

Rumors and hype are swirling around reports that OpenAI researchers created a new model, called Q* (pronounced Q-star), able to complete grade-school-level math problems. This new development, and Altmans push for commercialization, are what some observers believe to have spooked the nonprofit board, whose mission is centered around developing AI for the good of humanity.

A generative artificial intelligence (AI) model that can regularly and reliably solve math problems on its own would constitute a huge advance in the capabilities of AI systems.

Even todays most advanced and cutting-edge AI systems struggle to repeatably solve relatively simple math problems, a situation that has been both vexing AI researchers and inspiring them to push the field forward for years.

If there is an AI model out there, or under development, that can really do math even simple equations on its own, then that represents a massive leap forward for AIs applications across many industries, especially payments.

Math, after all, is a benchmark for reasoning. And the bread and butter for most AI models particularly large language models (LLMs) is pattern recognition, not logical sequence cognition.

Read also: Specific Applications of Gen AI Are Winning Play for Enterprises

LLMs are trained on text and other data that would take a human many millennia to read, but generative AI models still cant be trusted to reliably discern that if X is the same as Y, then Y is the same as X.

AI systems with the ability to plan already exist, however they are typically embedded within highly contextually limited scenarios, such as playing chess, where the rules and permutations are fixed, or controlling a robot on a grid. Outside of their defined zone of expertise, these systems, including Google DeepMinds AlphaGo and AlphaGo Zero, are limited in their planning capacity even when compared to animals like cats or mice.

Building a generative AI system that is capable of unsupervised reasoning and able to solve math problems without regular mistakes is a challenging, but important, milestone.

The name of OpenAIs alleged model, Q*, may give a hint as to how to get there. It combines two fundamental computer science techniques, Q-learning and A* (pronounced A-star).

A* was created to build a mobile robot that could plan its own actions; while Q-learning is a model-free reinforcement learning algorithm to learn the value of an action in a particular state.

Performing math reliably, for both AIs and humans, requires planning over multiple steps. But a senior NVIDIA scientist, Dr. Jim Fan, tweeted that some combination of an A* model like Googles chess rules AlphaGO and a Q-learning system like an LLM could someday get there.

Maybe that model, combining Q-learning with A*, would be called something like Q*.

See also: What Superintelligent Sentience (AGI) Means for the AI Ecosystem

Most AI models operate on weights, not contexts, meaning they operate without truly understanding what they are dealing with. To perform math, however, an in-step sequential understanding is crucial.

An AI capable of doing math reliably is an enticing concept because, as in the laws of nature themselves, math represents a foundation of learning for other, more abstract tasks.

A 2023 research paper by OpenAIs Sutskever along with other OpenAI researchers, titled Lets Verify Step by Step investigates this concept while attempting to reduce the regularity of AI models trained on the MATH dataset from producing logical mistakes. The OpenAI scientists leveraged a dataset of 800,000 step-level human feedback labels to train their model.

Getting AI models to solve math problems would represent a crucial step in the innovations ability to transform enterprise workflows and operations, helping reduce the daily labor burden on internal teams by evolving their responsibilities from doing the process to managing or overseeing it.

Within security-critical areas like finance and payments, the future-fit impact of this as-yet hypothetical capability cant be overstated.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

Visit link:
AI Systems That Master Math Will Change the World - PYMNTS.com

Google's Gemini, hailed by CEO Sundar Pichai as an exciting innovation, has been making waves since its announcement. This development, following the seismic impact of ChatGPT's launch last November, prompted Google to take decisive action, investing substantially in catching up with the generative AI trend. This concerted effort led not only to the introduction of Google Bard but also to the unveiling of Google Gemini.

We are building our next generation of models with Gemini and I am extraordinarily excited at the innovation coming ahead. I expect it to be a golden age of innovation ahead and can't wait to bring all the innovations to more people, Pichai recently said at the APEC CEO Conference.

What exactly is Google Gemini?

Gemini represents a suite of large language models (LLMs) employing training methodologies akin to those used in AlphaGo, integrating reinforcement learning and tree search techniques. It holds the potential to challenge ChatGPT's dominance as the premier generative AI solution globally.

It emerged mere months after Google amalgamated its Brain and DeepMind AI labs to establish a new research entity known as Google DeepMind. It also follows swiftly on the heels of Bard's launch and the introduction of its advanced PaLM 2 LLM.

While expectations suggest a potential release of Google Gemini in the autumn of 2023, comprehensive details regarding its capabilities remain elusive.

In May, Sundar Pichai, CEO of Google and Alphabet, shared a blog post offering a broad overview of the LLM, stating: "Gemini was purpose-built from the ground up to be multimodal, boasting highly efficient tool and API integrations, and designed to facilitate future innovations such as memory and planning."

Pichai also highlighted, "Despite being in its early stages, we are already witnessing remarkable multimodal capabilities not previously seen in earlier models. Once fine-tuned and rigorously assessed for safety, Gemini will be available in various sizes and functionalities, akin to PaLM 2."

Since then, official disclosures about its release have been scarce. Google DeepMind CEO Demis Hassabis, in an interview with Wired, hinted at Gemini's capabilities, mentioning its amalgamation of AlphaGo's strengths with the impressive language capabilities of large models.

According to Android Police, an anonymous source associated with the product suggested that Gemini will generate text alongside contextual images, drawing on sources such as YouTube video transcripts.

Challenges on the horizon

Google's extensive endeavour to catch up with OpenAI, the creators of ChatGPT, appears to be more challenging than initially anticipated, as reported by The Information.

Earlier this year, Google informed select cloud clients and business partners that they would gain access to the company's new conversational AI, the substantial language model Gemini, by November.

However, the company recently notified them to expect it in the first quarter of the following year, as revealed by two individuals with direct insight. This delay poses a significant challenge for Google, particularly amidst the slowdown in its cloud sales growth, contrasting with the accelerated growth of its larger rival, Microsoft. A portion of Microsoft's success can be attributed to selling OpenAI's technology to its customer base.

Also ReadGovernment to meet social media platforms including Meta and Google over deepfake crisis, says Ashwini Vaishnaw

Continue reading here:
What is Google Gemini? CEO Sundar Pichai says 'excited' about the innovation - Business Today

This article has been reviewed according to ScienceX's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

fact-checked

preprint

trusted source

proofread

by Peter Grad , Tech Xplore

close

A team of researchers at DeepMind focusing on the next frontier of artificial intelligenceArtificial General Intelligence (AGI)realized they needed to resolve one key issue first. What exactly, they asked, is AGI?

It is often viewed in general as a type of artificial intelligence that possesses the ability to understand, learn and apply knowledge across a broad range of tasks, operating like the human brain. Wikipedia broadens the scope by suggesting AGI is "a hypothetical type of intelligent agent [that] could learn to accomplish any intellectual task that human beings or animals can perform."

OpenAI's charter describes AGI as a set of "highly autonomous systems that outperform humans at most economically valuable work."

AI expert and founder of Geometric Intelligence Gary Marcus defined it as "any intelligence that is flexible and general, with resourcefulness and reliability comparable to (or beyond) human intelligence."

With so many variations in definitions, the DeepMind team embraced a simple notion voiced centuries ago by Voltaire: "If you wish to converse with me, define your terms."

In a paper published on the preprint server arXiv, the researchers outlined what they termed "a framework for classifying the capabilities and behavior of AGI models."

In doing so, they hope to establish a common language for researchers as they measure progress, compare approaches and assess risks.

"Achieving human-level 'intelligence' is an implicit or explicit north-star goal for many in our field," said Shane Legg, who introduced the term AGI 20 years ago.

In an interview with MIT Review, Legg explained, "I see so many discussions where people seem to be using the term to mean different things, and that leads to all sorts of confusion. Now that AGI is becoming such an important topic we need to sharpen up what we mean."

In the arXiv paper, titled "Levels of AGI: Operationalizing Progress on the Path to AGI," the team summarized several principles required of an AGI model. They include a focus on the capabilities of a system, not the process.

"Achieving AGI does not imply that systems 'think' or 'understand' [or] possess qualities such as consciousness or sentience," the team emphasized.

An AGI system must also have the ability to learn new tasks, and know when to seek clarification or assistance from humans for a task.

Another parameter is a focus on potential, and not necessarily actual deployment of a program. "Requiring deployment as a condition of measuring AGI introduces non-technical hurdles such as legal and social considerations, as well as potential ethical and safety concerns," the researchers explained.

The team then compiled a list of intelligence thresholds ranging from "Level 0, No AGI," to "Level 5, Superhuman." Levels 14 included "Emerging," "Competent," "Expert" and "Virtuosos" levels of achievement.

Three programs met the threshold of the label AGI. But those three, generative text models (ChatGPT, Bard and Llama 2), reached only "Level 1, Emerging." No other current AI programs met the criteria for AGI.

Other programs listed as AI included SHRDLU, an early natural language understanding computer developed at MIT, listed at "Level 1, Emerging AI."

At "Level 2, Competent" are Siri, Alexa and Google Assistant. The grammar checker Grammarly ranks at "Level 3, Expert AI."

Higher up this list, at "Level 4, Virtuoso," are Deep Blue and AlphaGo. Topping the list, "Level 5, Superhuman," are DeepMind's AlphaFold, which predicts a protein's 3D structure from its amino acid sequence; and StockFish, a powerful open-source chess program.

However, there is no single proposed definition for AGI, and there is constant change.

"As we gain more insights into these underlying processes, it may be important to revisit our definition of AGI," says Meredith Ringel Morris, Google DeepMind's principal scientist for human and AI interaction.

"It is impossible to enumerate the full set of tasks achievable by a sufficiently general intelligence," the researchers said. "As such, an AGI benchmark should be a living benchmark. Such a benchmark should therefore include a framework for generating and agreeing upon new tasks."

More information: Meredith Ringel Morris et al, Levels of AGI: Operationalizing Progress on the Path to AGI, arXiv (2023). DOI: 10.48550/arxiv.2311.02462

Journal information: arXiv

2023 Science X Network

See the rest here:
Researchers seek consensus on what constitutes Artificial General Intelligence - Tech Xplore

Sam Altman, cofounder of OpenAI, arrives for a bipartisan Artificial Intelligence (AI) Insight Forum for all US senators hosted by Senate Majority Leader Chuck Schumer at the US Capitol in Washington, on Sep 13, 2023.

OpenAI board is in discussions with Sam Altman to return as CEO, just a day after he was ousted. Sam was fired by the board on Friday with no notice and major investors including Microsoft were blindsided.

Sam co-founded OpenAI with Elon Musk and a team of AI scientists in 2015 with the goal of developing safe and beneficial AI. He has since been the face of the company, a leading figure in the field, and has been credited with the creation of ChatGPT.

Microsoft released a statement saying theyre still committed with their partnership with OpenAI but was caught off guard like other investors by the boards abrupt decision to oust CEO Sam Altman, leaving the companys future in doubt amidst fierce competition in the AI landscape with the rise LLMs like Google Bard, ChatGPT, and now xAI.

According to The Verge, OpenAI board is in discussions with Sam to return to the company as CEO, according to multiple people familiar with the matter. Altman, who was unexpectedly let go by the board, is undecided about his comeback and demands substantial governance alterations.

The 4 board members who voted out Sam Altman:

Helen Toner Director of strategy and foundational research grants at Georgetowns CSET, expert on Chinas AI landscape. She joined OpenAI board in September 2021.

Adam DAngelo CEO of Quora, advocate for OpenAIs capped-profit structure and nonprofit control. He joined OpenAI board in April 2018. He also crated Poe, an AI chatbot app which allows users to interact with many different chatbots (including ChatGPT, Claude, Llama, PaLM2, etc).

Tasha McCauley Adjunct senior management scientist at RAND Corporation, co-founder of Fellow Robots and GeoSim Systems. Shes also a signer of the Asilomar AI Principles, a set of 23 AI governance principles published in 2017. (Sam Altman, Iyla Sutskever, and Elon Musk also signed.)

Ilya Sutskever OpenAI cofounder, Russian-born chief scientist, co-author of a key paper in neural networks, helped lead the AlphaGo project.

Unlike traditional companies, OpenAI board is not focused on making money for shareholders. In fact, none of the board members even own shares in the company. Instead, their goal is to make sure that artificial general intelligence (AGI) is developed in a way that benefits everyone, not just a select few.

This is a very different approach than the one taken by most companies. Typically, companies are run by a board of directors who are responsible for making decisions that will increase shareholder value.

This often means maximizing profits, even if it comes at the expense of other stakeholders, such as employees, customers, or the environment.

This is a challenging task, but its one that OpenAI board is taking very seriously. They are working with some of the worlds leading experts on AI to develop guidelines and safeguards that will help to ensure that AGI is used for the benefit of all.

Follow this link:
Sam Altman In Talks to Return to OpenAI | by Derick David | Utopian ... - Medium

AI-generated image.

Predicting the weather is notoriously difficult. Not only are there a million and one parameters to consider but theres also a good degree of chaotic behavior in the atmosphere. But DeepMinds scientists (the same group that brought us AlphaGo and AlphaFold) have developed a system that can revolutionize weather forecasting. This advanced AI model leverages vast amounts of data to generate highly accurate predictions.

Weather forecasting, an indispensable tool in our daily lives, has undergone tremendous advancements over the years. Todays 6-day forecast is as good (if not better) than the 3-day forecast from 30 years ago. Storms and extreme weather events rarely catch people off-guard. You may not notice it because the improvement is gradual, but weather forecasting has progressed greatly.

This is more than just a convenience; its a lifesaver. Weather forecasts help people prepare for extreme events, saving lives and money. They are indispensable for farmers protecting their crops, and they significantly impact the global economy.

This is exactly where AI enters the room.

DeepMind scientists now claim theyve made a remarkable leap in weather forecasting with their GraphCast model. GraphCast is a sophisticated machine-learning algorithm that outperforms conventional weather forecasting around 90% of the time.

We believe this marks a turning point in weather forecasting, Googles researchers wrote in a study published Tuesday.

Crucially, GraphCast offers warnings much faster than standard models. For instance, in September, GraphCast accurately predicted that Hurricane Lee would make landfall in Nova Scotia nine days in advance. Currently used models predicted it only six days in advance.

The method that GraphCast uses is significantly different. Current forecasts typically use a lot of carefully defined physics equations. These are then transformed into algorithms and run on supercomputers, where models are simulated. As mentioned, scientists have this approach with great results so far.

However, this approach requires a lot of expertise and computation power. Machine learning offers a different approach. Instead of running equations on the current weather conditions, you look at the historical data. You see what type of conditions led to what type of weather. It gets even better: you can mix conventional methods with this new AI approach, and get accurate, fast readings.

Crucially, GraphCast and traditional approaches go hand-in-hand: we trained GraphCast on four decades of weather reanalysis data, from the ECMWFs ERA5 dataset. This trove is based on historical weather observations such as satellite images, radar, and weather stations using a traditional numerical weather prediction (NWP) to fill in the blanks where the observations are incomplete, to reconstruct a rich record of global historical weather, writes lead author Remi Lam, from DeepMind.

While GraphCasts training was computationally intensive, the resulting forecasting model is highly efficient. Making 10-day forecasts with GraphCast takes less than a minute on a single Google TPU v4 machine. For comparison, a 10-day forecast using a conventional approach can take hours of computation in a supercomputer with hundreds of machines.

The algorithm isnt perfect, it still lags behind conventional models in some regards (especially in precipitation forecasting). But considering how easy it is to use, its at least an excellent complement to existing forecasting tools. Theres another exciting bit about it: its open source. This means that companies and researchers can use and change it to better suit their needs.

Byopen-sourcing the model code for GraphCast,we are enabling scientists and forecasters around the world to benefit billions of people in their everyday lives. GraphCast is already being used by weather agencies, adds Lam.

The significance of this development cannot be overstated. As our planet faces increasingly unpredictable weather patterns due to climate change, the ability to accurately and quickly predict weather events becomes a critical tool in mitigating risks. The implications are far-reaching, from urban planning and disaster management to agriculture and air travel.

Moreover, the open-source nature of GraphCast democratizes access to cutting-edge forecasting technology. By making this powerful tool available to a wide range of users, from small-scale farmers in remote areas to large meteorological organizations, the potential for innovation and localized weather solutions increases exponentially.

No doubt, were witnessing another field where machine learning is making a difference. The marriage of AI and weather forecasting is not just a fleeting trend but a fundamental shift in how we understand and anticipate the whims of nature.

Read the original here:
For the first time, AI produces better weather predictions -- and it's ... - ZME Science