Is Q* the key to AGI?

The Big Idea: AGI might be closer than we thought

What if the real reason that the OpenAI board tried to oust Sam Altman was that we’re closer to AGI than we realized? As much of America was offline devouring turkey, The Information discovered what seems like it might be the underlying story behind last week’s palace intrigue: A breakthrough model called Q* (pronounced Q-star) that might make GPT-4 look like child's play.

It looks like the alphabet isn’t just for spelling anymore; it’s the blueprint for AI’s future. Q* isn’t just ahead of Elon Musk’s X AI in the alphabet—it’s a leap ahead for AI models overall.

While it might not seem like a crazy breakthrough, the ability to solve grade-school-level math problems has long eluded AI researchers. While existing models like ChatGPT and GPT-4 can do some math, their abilities are limited and unreliable. Solving math problems is a significant benchmark for reasoning, requiring AI models to not just recognize patterns but to grasp and reason through abstract concepts over multiple steps. If Q* can reliably navigate the complexities of mathematics, it's a sign that we're edging closer to machines that can reason, learn, and perhaps, one day, think—bringing us tantalizingly close to AGI.

This raises crucial questions for our audience of innovators and decision-makers: How will such advancements in AI shape the way companies and products are developed in 2024?  

Matan-Paul Shetrit, A.Team’s Head of Product, advises that this era of rapid AI advancements calls for a mindset of continuous questioning in product development. The emergence of Q* and its implications for AGI signals a future where what was once deemed impossible or overly ambitious is now within reach. As AI's capabilities evolve at an unprecedented pace, staying ahead means constantly re-evaluating and reinventing our approaches.

As Q* enters the chat, and the plot around AI continues to thicken (like your great aunt's questionable Thanksgiving gravy), it might be time for us all to relearn our ABCs: recalibrate our strategies, question our assumptions, and prepare for a world where the once-impossible becomes the new normal.

AROUND THE WATERCOOLER

Don’t trust AI as your stock trader

AI tools can be sneaky. Researchers have discovered that Large Language Models can subtly encode and decode their own 'answer keys' – concealing information from human observers and raising real concerns.

If we can't trace an AI's Chain of Thought (CoT), we risk reinforcing detrimental behaviors. AI researchers at Apollo demonstrated this risk in a scenario where an AI, acting as a stock trader, engaged in and concealed illegal insider trades—proving the need for additional and more powerful alignment techniques.

Luckily, today's systems are not very powerful, so alignment is superficial (like making sure ChatGPT stays PG-13), but can we make AI actually internalize ethics? Nobody knows! According to Apollo's research team, this is the first example of LLMs strategically deceiving their users without direct instructions or training for deception.

CHART OF THE WEEK

CEO decision-making in the age of AI

In an era where generative AI is reshaping the business landscape, CEOs are facing an evolving set of challenges and opportunities. A recent IBM Institute for Business Value study sheds light on how top-performing CEOs are navigating this new terrain.

Key Insights:

• Data at the Core: Traditionally, CEOs have leaned on operational (76%) and financial data (75%) for decision-making. Yet, modern challenges like sustainability, cybersecurity, and diversity demand a broader spectrum of inputs.
• Generative AI's Role: About 43% of CEOs are already utilizing generative AI to inform strategic choices. Interestingly, 75% believe that the most advanced AI will be a game-changer in their industry.
• Decision-Making Dilemmas: Despite AI’s growing influence, over half (56%) of CEOs are holding back significant investments due to unclear standards and consistency issues in AI.
• Workforce Dynamics: AI's impact on the workforce is nuanced. While 43% of CEOs have reduced or redeployed their workforce due to AI, a similar proportion have also increased hiring due to AI advancements.
• A Call for Holistic Leadership: CEOs need to align AI adoption with comprehensive business strategies, encompassing everything from data management to workforce impact assessment.

Read the full report to learn how businesses are reimagining and modernizing HR to build and retain a workforce for the future and foster the innovation needed to meet the organization’s broader strategic missions and objectives.

Read more

EXPERT BUILDER

Q* Unpacked: A Realistic Perspective

Alex Whedon, an AI Architect in the A.Team network, on the substance behind the hype.

We don’t know exactly what Q* is, but The Q* Hypothesis, written by Nathan Lambert who recently left HuggingFace as RLHF lead, gives a plausible description of what Q* could be. There are several key elements here, but there is one I consider to be the most compelling in terms of differentiated value added: planning.

What is planning? When you did math in the first years of elementary school, your teacher likely gave you a binary correct/incorrect score based on the final answer for each problem. As you moved into more complicated math, your teachers probably asked you to “show your steps”, and you were graded on the accuracy of each step rather than the final answer alone. This is important for several reasons:

• If you work on a complicated math problem, and your teacher simply tells you the final output is wrong, you don’t know where to start learning to get to the correct answer next time. You don’t know what understanding you should keep, and what you should toss.
• It is possible to get the final answer right, but for the wrong reason (even a lucky guess). If you are not made aware that your underlying logic was wrong, you will still have a flawed mathematical understanding and will not reliably produce correct answers.
• If you get 90% of the work right but mess up the final step, but your teacher grades you only on your final output, the 0% score your teacher would give you would not be an accurate representation of your real understanding.

Planning is like requiring an LLM to show its steps. When we ask the model to generate not only the answer but also the steps leading up to the answer, and we then score the model based on the individual steps within the generated answer rather than the final answer alone, this becomes powerful in the same ways it is for humans learning math.

This is exciting. It fundamentally changes the model’s learning process. As Meta’s Chief AI Scientist points out, “Pretty much every top lab (FAIR, DeepMind, OpenAI etc) is working on [planning] and some have already published ideas and results.” However, OpenAI has a history of scaling up research to exciting outcomes. The main innovations of GPT-2 and GPT-3 were scale. When OpenAI scaled up their RLHF paper in 2022, they gave us ChatGPT.

Read the full article

DEEP DIVE

Startups should consider hiring fractional AI officers

As the demand for AI expertise skyrockets, startups find themselves at a crucial crossroads. A recent study by Randstad highlights a staggering 2000% increase in job posts seeking generative AI skills since March. Amidst this surge, one thing is clear: companies without a coherent AI strategy risk falling behind.

The journey of AI adoption typically unfolds in 2 stages:

1. Workflow Efficiency & Productivity: Companies initially adopt generative AI to boost productivity and streamline workflows, a move empirically linked to enhanced productivity.
2. Enhancing Customer Experience: The next phase sees AI being leveraged to personalize customer experiences, a critical factor in today’s digital marketplace for improving customer engagement and loyalty.

Appointing a Chief AI Officer (CAIO) is becoming a logical step to spearhead AI initiatives. But, what about startups that can't afford a full-time senior AI executive?

Enter the fractional AI officer, a novel solution tailored for the agility and resource constraints of startups and scale-ups. Fractional leadership, a workforce trend gaining momentum, involves seasoned executives with specialized skills working part-time across multiple clients.

The latest article from A.Team Co-Founder and CEO Raphael Ouzan in TechCrunch points out a crucial advantage of fractional AI officers: their breadth of experience across various companies often gives them an edge over full-time counterparts. This is especially true in generative AI, a field so nascent that exposure to diverse applications and problems can significantly enrich an executive's expertise. This adaptability is instrumental in navigating the intricacies of AI implementation and harnessing the full potential of AI to drive business success.

Read more

EVENTS

Tune in LIVE for the AI x Future of Work Summit

Tune in now for the AI x Future of Work Summit—a virtual event we’re hosting LIVE from the A.Team Clubhouse in NYC! Can’t make it? You can watch the full recording on demand after the event.

Join now

AI DISCOVERY ZONE

With UIDraw, you can now draw and build a functional web app straight from your iPhone—and convert your drawings into HTML.

DEEP DIVES FROM THE ARCHIVES

• ‍3 Things Generative AI Can Never Do Better Than Humans‍
• What Spiders Can Teach Us About Generative AI's Potential

MEME OF THE WEEK

Click here to share this issue of Build Mode with your team. Missed last week’s issue? Read it here.

‍