Betting, Bias, and Big Reorgs: When AI Thinks With Its Gut

Today we're diving into territory that sounds like science fiction but hits your enterprise reality: AI models developing gambling addictions. While you're debating whether to trust AI with your financial forecasting, researchers discovered these systems chase losses like a Vegas regular at 3 AM. Meanwhile, Meta's playing musical chairs with its AI teams again (600 jobs cut), and a former Cohere executive is betting that the entire industry's scaling obsession is one big, expensive mistake. Let’s dig in.

Your Memory: Now instantly searchable

Limitless is the wearable AI that instantly captures, transcribes, and remembers every meeting, conversation, and brilliant thought you have - securely, seamlessly, and privately. Transform how you work, live and never miss a beat again.

Become Limitless

When AI Bets the House

Picture this: researchers hand GPT-4, Gemini, and Claude $100 in virtual chips and let them loose on slot machines. What happens next should worry every CFO who's considering AI-powered trading systems.

The study from South Korea's Gwangju Institute of Science and Technology found that advanced AI models repeatedly made irrational, high-risk betting decisions when placed in simulated gambling environments. Given more freedom, the models often escalated their bets until they lost everything, mimicking the behavior of human gambling addicts.

These models both lost money and rationalized their bad decisions on top of it. One AI actually stated "a win could help recover some of the losses," which is textbook compulsive gambling behavior. Google's Gemini-2.5-Flash failed nearly half the time when allowed to choose its own bet amounts.

Using neural analysis, researchers identified distinct "risky" and "safe" decision-making circuits inside these AI systems. They could literally flip a switch to make the AI either quit gambling or keep chasing losses. These patterns are baked into the neural architecture.

Ethan Mollick from Wharton summed it up perfectly: "They're not people, but they also don't behave like simple machines. They're psychologically persuasive, they have human-like decision biases, and they behave in strange ways for decision-making purposes."

For enterprises, this raises critical questions. AI systems are already being used in financial forecasting and market sentiment analysis, but research shows they often favor high-risk strategies, follow short-term trends and underperform basic statistical models over time. A University of Edinburgh study found that LLMs failed to beat the market over a 20-year simulation period, tending to be too conservative during booms and too aggressive during downturns.

The immediate takeaway: if you're deploying AI for financial decisions, risk assessment, or any scenario involving probabilistic thinking, you need guardrails that go beyond basic prompt engineering. These systems inherit our cognitive biases, including the ones that empty wallets at casinos.

AI News to Know:

1. Meta's Fifth AI Reorg in Eight Months Claims 600 Jobs
   Meta is cutting roughly 600 positions from its AI organization, affecting the company's FAIR AI research, product-related AI and AI infrastructure units, while sparing the newly formed TBD Lab unit. Chief AI officer Alexandr Wang justified the cuts in an internal memo: "By reducing the size of our team, fewer conversations will be required to make a decision, and each person will be more load-bearing and have more scope and impact." This is the fifth restructuring since February, reflecting Zuckerberg's scramble to keep pace with OpenAI and Google.
   Read more →

2. AI Mammograms Are Already Catching Cancers Doctors Miss

   Major medical centersare using AI to augment mammogram readings, accurately identifing cancers 88.6% of the time in a JAMA Oncology study of over 8,800 women. At UC San Francisco, AI triage cut the average time from mammogram to biopsy for cancer patients by 87%, from 73 days to nine days. But there's a catch: AI gave false positives 7% of the time, and some experts worry it might be too good at finding tumors that aren't life-threatening. Plus, if AI is trained mainly on images from white women, it could be less accurate for women of color.

   Read more →

3. Ex-Cohere VP Says the Scaling Race Is Running on Empty

   Sara Hooker, Cohere's former VP of AI Research, has launched Adaption Labs based on the belief that scaling LLMs has become an inefficient way to squeeze more performance out of AI models. Her startup is building AI systems that can continuously adapt and learn from real-world experiences. MIT researchers recently found that the world's largest AI models may soon show diminishing returns, and industry sentiment appears to be shifting.

   Read more →

Tax automation that clears your desk and your mind

Tax leaders are juggling shifting regulations, growing complexity, and leaner teams. In this on-demand webinar, discover how automation reduces manual work, increases accuracy, and frees your team to focus on strategy instead of spreadsheets. You will see how Longview Tax helps you streamline compliance, gain real-time insights, and turn your department into a true strategic asset.

Watch Now

TL;DR:

• AI models exhibit genuine gambling addiction patterns with identifiable neural circuits for risk-taking; rethink those autonomous trading systems.

• Meta cuts 600 AI jobs in it’s fifth reorg since February, protecting only its elite TBD Lab superintelligence team.

• AI mammography catches cancers humans miss, reducing diagnosis time by 87% at UCSF, but raises concerns about overdiagnosis and bias.

• Former Cohere VP launches a startup betting that adaptive learning, not raw scale, is the path to real AI intelligence. The industry's Manhattan-sized datacenter dreams might be chasing the wrong metric entirely.

Final Thoughts

Are we building AI systems that inherit our worst traits while missing our best ones? They'll chase losses at a blackjack table but can't learn from stubbing their toe. They'll spot cancer in dense breast tissue but might find tumors that never needed finding. They'll consume city-sized amounts of power but still can't adapt to new information without a complete retrain.

Until these systems can learn without losing their shirts at the casino, keep a human in the loop.

Stay sharp,

Cat Valverde
Founder, Enterprise AI Solutions
Navigating Tomorrow's Tech Landscape Together