Question 1

How accurate are current deepfake detection technologies?

Accepted Answer

_Detection tools fail more often than they succeed right now._

Current deepfake detection technology isn't very good yet. The best deepfake detector is only 65% accurate. Even Microsoft's Azure Cognitive Services was fooled over 78% of the time. This means detection tools fail more often than they succeed. The future of these detectors will mirror bot detectors, spam detectors, or any other cyber threat detectors. It will be an arms race between those creating deepfakes and those trying to detect them, similar to how spam and virus detection evolved over time.

— [Deepfakes - The Danger Of Artificial Intelligence That We Will Learn To Manage Better](https://www.forbes.com/sites/lutzfinger/2022/09/08/deepfakesthe-danger-of-artificial-intelligence-that-we-will-learn-to-manage-better/) · _Forbes_

Question 2

How did AI models respond when asked to create a plan for a U.S. president's third term?

Accepted Answer

_AI models compiled scattered ideas into a cohesive third term outline._

One of my students, Dane Troup, prompted OpenAI, Grok, DeepSeek and Gemini with a question framed as research for a political thriller novel. Google's Gemini did not answer, claiming it had to learn, though dishonestly. All the other models responded. The main ideas they produced have all been discussed before: amending the 22nd Amendment, reshaping the Supreme Court, and bombarding society with fake news. No single model provided a complete list, but together they offer a solid outline and plan.

— [Trump's Third Term — The Role Of AI In Creating A Strategy](https://www.forbes.com/sites/lutzfinger/2025/03/31/trumps-third-termhow-ai-can-help-to-overthrow-the-us-government/) · _Forbes_

Question 3

How did we learn that location data from our phones was being sold?

Accepted Answer

_Pandemic dashboards exposed a hidden ecosystem of location tracking._

During the pandemic, many companies created dashboards showing aggregated movement data. This revealed that smaller companies like Unacast and Placer.ai gathered location information via hundreds of apps on your phone. Additionally, phone companies Sprint, AT&T, T-Mobile, and Verizon sold our location data. There is an entire ecosystem of apps that share a lot of information about us. WhatsApp is probably not the biggest privacy offender we have in our phones when you consider this broader data sharing landscape.

— [WhatsApp: We Should Discuss What Our Data Is Used For, Not Who Has It](https://www.forbes.com/sites/lutzfinger/2021/01/26/whatsapp-we-should-discuss-what-our-data-is-used-for-not-who-has-it/) · _Forbes_

Question 4

How do biases in narrow AI systems create real-world harm?

Accepted Answer

_AI systems enforce human prejudices at massive scale._

Narrow AI amplifies existing biases at scale. In parole decisions, computers were meant to remove human bias, but when fed biased data, they kept more minorities in jail. Loan application systems with few women in training data couldn't correctly handle female applicants. The stop and frisk program in New York used a computer to select precincts, but it was racially unjust. When you remove skin color from the data, the system became even more skewed. There's a logical data science explanation for this.

— [Bias in, bias out: Addressing the impact of bias in AI](https://www.lutzfinger.com/transcript/2023-06-07-bias-in-bias-out-addressing-the-impact-of-bias-in-ai/) · _The Edge_

Question 5

How do you spot social media bots?

Accepted Answer

_Detection heuristics._

> Figure 6-4. Subset of a social network, created using LinkedIn Labs Since bot networks will first create their own group by linking to one another before they start to infiltrate the rest of the network, especially in the beginning, you will see a highly dense network with very low cohesion to the rest of the network. However, those measures using network topology only work in the beginning as the bot network is formed. Once the bots…

— [Finger Dutta Ask Measure Learn (1)](https://www.lutzfinger.com/book/) · _Book — Ask, Measure, Learn_

Question 6

How does Catalog prevent large language models from hallucinating when retrieving enterprise information?

Accepted Answer

_Stopping hallucinations by feeding LLMs the right information first._

The way to avoid hallucination is to stop the model from reaching into the model layers for information. If you give it structured or unstructured information in raw format and then ask it to summarize or list, it gives you exactly what you provided. Without that context, it makes things up because it's doing probabilistic retrieval from its training. This is retrieval augmented generation. You get the information first, augment the prompt with it, then generate the response based on that specific context.

— [Large Language Models: one interface to rule them all?](https://www.lutzfinger.com/transcript/2023-08-15-large-language-models-one-interface-to-rule-them-all/) · _The Edge_

Question 7

How might data regulation affect innovation and global competition?

Accepted Answer

_Regulation may slow innovation but remains necessary for trust._

Regulation of data use will force companies to ask for consent before developing new algorithms. This extra step will inadvertently slow down innovation and might be a factor in the global innovation competition. China can correlate and analyze datasets without regulatory constraints, creating a competitive advantage. Despite this challenge, we need to regulate our markets and expect international partners to comply with the same rules. We should push for clear regulations and international norms on transparency, accountability, and data ownership.

— [WhatsApp: We Should Discuss What Our Data Is Used For, Not Who Has It](https://www.forbes.com/sites/lutzfinger/2021/01/26/whatsapp-we-should-discuss-what-our-data-is-used-for-not-who-has-it/) · _Forbes_

Question 8

Is AGI near?

Accepted Answer

_Lutz's stance on the AGI timeline._

> Both have been hot topics in AI, sparking both fear and excitement. Sam Altman predicted AGI would arrive in 2025, and Elon Musk 2026. Both are more about hype than reality. In 2025, there will be no AGI but large language models will find their "killer app." What are AGI and the Singularity?

— [No AGI But A 'Killer App' — 2025 AI Prediction](https://www.forbes.com/sites/lutzfinger/2025/01/03/no-agi-but-a-killer-app2025-ai-prediction-110/) · _Forbes_

Question 9

Is OpenAI going bankrupt?

Accepted Answer

_Lutz's take on OpenAI economics and moats._

> OpenAI — "Experts predict bankruptcy" — was attention-grabbing, a deeper look reveals that the company has simply seen a drop in its users. Should you fear ChatGPT will disappear? No, quite the opposite.

— [OpenAI Isn't Going Bankrupt, But It Has A Business Model Problem](https://www.forbes.com/sites/lutzfinger/2023/08/18/is-openai-going-bankrupt-no-but-ai-models-dont-create-moats/) · _Forbes_

Question 10

Is the content AI produces about controversial topics actually new information?

Accepted Answer

_AI summarizes existing knowledge but creates no fundamentally new content._

AI simply summarizes. The actual content is not new. Knowledge has simply become more readily accessible, for good and for bad. These ideas are nothing new. They have been part of political fiction, academic debate and public discourse for years. What's different is that AI systems can now surface them instantly and in a cohesive manner. The risk is not in having large language models but in having humans act on plans that may not be in the best interest of the majority.

— [Trump's Third Term — The Role Of AI In Creating A Strategy](https://www.forbes.com/sites/lutzfinger/2025/03/31/trumps-third-termhow-ai-can-help-to-overthrow-the-us-government/) · _Forbes_

Question 11

Should you trust OpenAI blindly?

Accepted Answer

_Vendor risk and architectural diversification._

> Stargate's $500B investment — message to the competitors — meant to intimidate potential challengers with overwhelming financial backing. AI is extremely dangerous; only a select few should be allowed to develop it — OpenAI's storyline to use safety concerns as justification for a regulatory setup that creates barriers to entry and gives them more monopolistic power. From now on, models should respect IP rights — OpenAI ignored IP…

— [Is OpenAI A Victim — DeepSeek And IP Rights — A Playbook For Leaders](https://www.forbes.com/sites/lutzfinger/2025/01/29/is-openai-a-victim---deepseek-and-ip-rights---a-playbook-for-leaders/) · _Forbes_

Question 12

What about deepfakes?

Accepted Answer

_How society will adapt to synthetic media._

> As AI lowers the barriers to content creation, abuse is also becoming more feasible. Deepfakes — AI-generated images, videos and audio designed to mimic real people — pose serious risks. Already in 2022, I demonstrated how easily they can be created. This increases the risk of misinformation, identity theft and political manipulation.

— [Goku, A Chinese Open-Source Model, Challenges OpenAI And The U.S.](https://www.forbes.com/sites/lutzfinger/2025/02/11/goku---a-chinese-open-source-model-challenges-openai-and-the-us/) · _Forbes_

Question 13

What are some of the dangerous ways deepfakes could be misused beyond pornography?

Accepted Answer

_From pornography to courtrooms, the threat landscape is expanding._

While early deepfakes were focused on pornography, there are more terrifying use cases that concern me. These include creating fake alibis in courtrooms, which could undermine the justice system. Extortion is another serious threat, where fabricated videos could be used to blackmail individuals. Terrorism is also a potential danger. These applications show why the simplicity of creating deepfakes is so concerning when we think about abuse and manipulation.

— [Overview Of How To Create Deepfakes - It's Scarily Simple](https://www.forbes.com/sites/lutzfinger/2022/09/08/overview-of-how-to-create-deepfakesits-scarily-simple/) · _Forbes_

Question 14

What are the fundamental limits of large language models that people should understand?

Accepted Answer

_Language models face limits in reasoning, planning, and true understanding._

First, language is only part of human knowledge. We know more than we can describe. A four year old has had way more sensory impressions than any large language model, despite not being able to read. Second, LLMs can't truly reason or explain their decisions because they're built on neural networks, which work like gut feel rather than rule based systems. They predict the next word based on probability distributions, not logical reasoning. Third, there's the data quality problem. And finally, LLMs are fundamentally probabilistic, so they can't reliably predict specific individual choices or plan multi step processes independently.

— [A reality check on AI](https://www.lutzfinger.com/transcript/2024-07-10-a-reality-check-on-ai/) · _The Edge_

Question 15

What are the main limitations of large language models that knowledge graphs help address?

Accepted Answer

_LLMs are powerful but flawed. Knowledge graphs provide the missing structure._

Large language models have several undesired behaviors. They hallucinate, generating false information. They need guardrails to prevent answering certain prompts inappropriately. They have gaps in their encoded knowledge and sometimes fail to retrieve information even when it exists, often due to how prompts are formulated. The prompt itself can cause the model to deprioritize relevant information. Knowledge graphs help with all these issues by adding structure and control. They make models more reliable and accurate at every step, from training to retrieval.

— [The synergy between LLMs and knowledge graphs](https://www.lutzfinger.com/transcript/2023-08-24-the-synergy-between-llms-and-knowledge-graphs/) · _The Edge_

Question 16

What are the main problems with using simple GPT wrappers for enterprise AI applications?

Accepted Answer

_Why basic ChatGPT integrations fail in business settings._

GPT wrappers have several critical shortcomings that prevent them from working well in enterprise settings. They lack system integration, so they cannot check calendars or access other tools. They have no context about user preferences or past interactions. Security is missing, which means they could accidentally reveal private details. They lack guardrails to handle controversial questions appropriately. They offer no user control over outputs. And they suffer from hallucinations, where the AI sometimes makes things up. These issues make simple wrappers unsuitable for real enterprise deployment.

— [AI Agents In 2025: What Enterprise Leaders Need To Know](https://www.forbes.com/sites/lutzfinger/2025/01/05/ai-agents-in-2025-what-enterprise-leaders-need-to-know/) · _Forbes_

Question 17

What are the real AI risks that Europe should be concerned about?

Accepted Answer

_Human misuse, not AGI, is the real danger._

AI has risks, but they are not the ones Sam Altman wants you to believe. No, AGI is not here, and no, AGI is not taking over. The real risk is human misuse of increasingly powerful AI. Instead of regulating technology itself, laws should target harmful outcomes of AI while allowing innovation. Europe has focused too much on AI oversight rather than fostering innovation. As Macron warned, we are regulating ourselves out of the market. Instead of embracing new business models, incumbents fight legal battles over IP rights.

— [Macron's AI Investment — But Europe Will Need To Do More To Catch Up](https://www.forbes.com/sites/lutzfinger/2025/02/10/europe-is-behind-in-ai---heres-how-to-catch-up-and-innovate/) · _Forbes_

Question 18

What are the three main problems with training data that can cause AI to get things wrong?

Accepted Answer

_Three critical failures that undermine machine learning accuracy and fairness._

Training data can fail in three ways. First, it might not be sufficient. The data may lack context or situations the algorithm will encounter in the real world. Second, it might be biased. This means data is systematically missing certain perspectives or groups. For example, if texts predominantly show nurses as women, the algorithm learns that pattern without understanding societal goals like gender equality. Third, the data might be outright wrong. Labels may be inaccurate or based on performances rather than reality, leading algorithms to learn incorrect associations.

— [It's The Data, Stupid! Why AI Might Get It Wrong.](https://www.forbes.com/sites/lutzfinger/2020/05/28/its-the-data-stupid-why-ai-might-get-it-wrong/) · _Forbes_

Question 19

What deepfake risks does Goku introduce and how should society respond?

Accepted Answer

_Goku makes deepfakes easier, requiring skepticism and AI literacy investments._

As AI lowers the barriers to content creation, abuse is also becoming more feasible. Deepfakes, AI-generated images, videos and audio designed to mimic real people, pose serious risks. This increases the risk of misinformation, identity theft and political manipulation. With tools like Goku making hyper-realistic content easier to produce, responsible AI use and deepfake detection must become priorities. Society must become more skeptical of digital content and invest in solutions to identify and counteract fakes. The only effective cure is to invest in AI literacy.

— [Goku, A Chinese Open-Source Model, Challenges OpenAI And The U.S.](https://www.forbes.com/sites/lutzfinger/2025/02/11/goku---a-chinese-open-source-model-challenges-openai-and-the-us/) · _Forbes_

Question 20

What did the Harvard study with executives reveal about relying on AI for predictions?

Accepted Answer

_Executives using ChatGPT made worse predictions due to overconfidence and bias._

Harvard conducted an experiment with nearly 300 executives and found that those who relied on ChatGPT for stock price forecasts grew more optimistic and overconfident, and ultimately made worse predictions than peers who worked with other humans to discuss the logic. The study shows how AI can amplify cognitive biases and distort judgment in high-stakes decisions. This demonstrates how easily the past can be a wrong predictor for current times, and why putting faith in AI without human judgment can lead to poor outcomes.

— [AI Fail — Trump Says 'Merz' —YouTube Transcribes 'Merkel'](https://www.forbes.com/sites/lutzfinger/2025/08/18/ai-fail---chancellor-merz-is-not-chancellor-merkel/) · _Forbes_

Question 21

What framework helps assess whether imperfect AI decisions are acceptable?

Accepted Answer

_Match your AI quality threshold to the stakes of the decision._

The framework asks: How good is your decision quality, and how important is the decision? How bad is a false decision? We discussed two opposite examples at Station F. One company did predictive maintenance for infrastructure where errors are costly, requiring high quality predictions. Another did social media outreach for hiring where low quality matters less since you already have low close rates. You can mitigate both. For low-stakes, automate with feedback loops. For high-stakes infrastructure, use augmentation first, guiding engineers and creating Jira tickets.

— [Building AI products: 5 lessons from our founders' workshop](https://www.lutzfinger.com/transcript/2024-08-01-building-ai-products-5-lessons-from-our-founders-workshop/) · _The Edge_

Question 22

What fundamental capability do AI transcription models lack that causes errors like this?

Accepted Answer

_AI generates likely sequences but cannot verify facts or cross-check reality._

AI transcription models don't verify facts. They don't pause and think, 'Merkel is no longer Chancellor, so that must be wrong.' They don't access live knowledge graphs or cross-check reality. They just generate the most likely sequence of words based on historical data. That's why AI feels brilliant in some moments and brain-dead in others. It reflects the averages of the past, not necessarily the truth of the present. The crucial gap is that these models cannot apply logic or verify their outputs against current reality.

— [AI Fail — Trump Says 'Merz' —YouTube Transcribes 'Merkel'](https://www.forbes.com/sites/lutzfinger/2025/08/18/ai-fail---chancellor-merz-is-not-chancellor-merkel/) · _Forbes_

Question 23

What happened in Eric Loomis's case regarding algorithmic transparency and trade secrets?

Accepted Answer

_Trade secrets blocked a defendant from analyzing his sentencing algorithm._

Eric Loomis was sentenced to seven years in prison partly based on an algorithmic assessment labeling him high risk. When he wanted to independently analyze the algorithm, the private company responsible denied access, claiming they needed to protect their trade secrets. Wisconsin's Supreme Court sided with the private company. This case illustrates how companies can behave as much of a black box as the algorithms themselves, preventing defendants from understanding or challenging the automated systems used against them.

— [President Biden Is Man, Woman And 40 Years Old - Why We Need Algorithmic Transparency](https://www.forbes.com/sites/lutzfinger/2021/02/04/president-biden-is-man-woman-and-40-years-old---why-we-need-algorithmic-transparency/) · _Forbes_

Question 24

What happened when Meta's AI system changed True Classic's ad creative?

Accepted Answer

_When AI replaces your best ad, conversion rates can tank._

In late October 2025, Meta's system replaced True Classic's best-performing ad with an AI-generated image. The original ad featured a younger male model that fit their 30 to 45 year old male audience. Meta's AI replaced it with an older female model. The result was that conversion rates tanked. This example shows what happens when AI makes creative decisions without human oversight and judgment about what actually works for a specific audience.

— [AI Cuts Creative Teams: What Companies Must Learn About AI Leadership](https://www.forbes.com/sites/lutzfinger/2026/02/13/ai-cuts-creative-teams-what-companies-must-learn-about-ai-leadership/) · _Forbes_

Question 25

What is the hallucination problem with large language models and why does it matter for businesses?

Accepted Answer

_LLMs invent answers when they don't know, risking business credibility_

Large language models will make things up when they don't know the answer. I demonstrated this by asking ChatGPT to summarize a completely fake Cornell website that doesn't exist. It generated a summary saying my course was the most amazing and made very good logical arguments, but it was completely fake. For businesses, this is a huge problem. If a customer asks a shop for something it doesn't have, the model might just make it up and promise red shoes with blinking laces. That's why proprietary data matters. You need to ground the model on your local data.

— [AI as an interface](https://www.lutzfinger.com/transcript/2023-04-24-ai-as-an-interface/) · _The Edge_

Question 26

What is the ultimate lesson about human responsibility in the age of AI summarization?

Accepted Answer

_AI summarization is neutral; human judgment and action remain essential._

We are not doomed. Human agency remains essential. Just because something sounds smart doesn't mean it is smart. We humans are the ones who must evaluate and act upon any plan or threat. We humans are needed. Summarization via AI is, in itself, neutral. What matters is how we, as humans, choose to use it. Language models can synthesize scattered theories into clear action plans, which can be both enlightening and unsettling, but humans must decide what to do with them.

— [Trump's Third Term — The Role Of AI In Creating A Strategy](https://www.forbes.com/sites/lutzfinger/2025/03/31/trumps-third-termhow-ai-can-help-to-overthrow-the-us-government/) · _Forbes_

Question 27

What is your perspective on AGI and when it might happen?

Accepted Answer

_AGI remains limited by data and lacks human agency for decisions._

AGI means artificial general intelligence, models that learn and develop capabilities by themselves, becoming vastly more intelligent than humans. But I disagree with this focus. AGI still has the data problem and only one type of prediction modeling, while humans have emotional prediction and other ways of understanding. Crucially, we humans have agency. We decide what data to clean, what to feed models, when to correct them. Models can't make those decisions themselves. I'm not against smarter models, but why not focus them on expert tasks like predicting climate change? I'm focused on using data and machine learning to improve things, not worried about some ominous AGI taking over.

— [A reality check on AI](https://www.lutzfinger.com/transcript/2024-07-10-a-reality-check-on-ai/) · _The Edge_

Question 28

What kinds of personal characteristics can be predicted from Facebook data?

Accepted Answer

_Facebook data reveals far more about us than we realize._

Michal Kosinski showed in 2013 that using Facebook data could accurately predict sexual orientation, ethnicity, personality traits, intelligence, happiness, addictive substances use, parental separation, age, and gender. The model even managed to separate between Democrat and Republican in 85% of cases. This research demonstrates how powerful social media data can be for profiling individuals, which contributes to why people feel uncomfortable when they don't know how their data is being used without their knowledge.

— [WhatsApp: We Should Discuss What Our Data Is Used For, Not Who Has It](https://www.forbes.com/sites/lutzfinger/2021/01/26/whatsapp-we-should-discuss-what-our-data-is-used-for-not-who-has-it/) · _Forbes_

Question 29

What lessons from social media regulation are shaping today's AI discussions?

Accepted Answer

_Social media's harms taught us to regulate AI before it's too late._

Julie Owono notes that the conversation we're having today about AI is probably the lessons we've learned from social media. When Facebook launched, there was this famous phrase, move fast and break things. At the time, we didn't think about the negative externalities. Innovation was central. Now we are all in this room to discuss regulation, which is historically a quite interesting and big step in the right direction. We may not all agree on the details, whether lawmakers understand technology or technologists understand the importance of protecting democracy, but there is agreement virtually around the world.

— [SZ 139: AI in the EU Ft. Yaad, Gerard, Florence, Julie, & Lutz](https://www.lutzfinger.com/transcript/2024-12-09-sz-139-ai-in-the-eu-ft-yaad-gerard-florence-julie-lutz/) · _Silicon Zombies_

Question 30

What pattern appears consistently in data about AI adoption and concern?

Accepted Answer

_Survey data reveals a consistent gap between concern and usage_

People express concern about AI in surveys. They worry about job displacement, privacy, hallucinations, and dependency. And then they use it anyway. This pattern shows up consistently in the data. The gap between what people say they're worried about and how they actually behave is not hypocrisy. It's simply how humans work. We routinely accept risks when the benefit feels immediate and the risk feels diffuse or abstract.

— [Chris Hemsworth Super Bowl Ad Shows Why We'll Use AI Even If We Fear It](https://www.forbes.com/sites/lutzfinger/2026/02/10/even-if-ai-could-kill-chris-hemsworth-convenience-beats-fear/) · _Forbes_

Question 31

What problems does Lutz see with the 'human in the loop' concept in Biden's plan?

Accepted Answer

_Speed and deception challenge the viability of human oversight._

Biden's plan says the President will decide when to use military AI, ensuring accountability through human involvement. But this isn't always realistic. Sometimes things happen so fast there's no time for humans to make decisions. Luxury cars already use AI to tighten seat belts when sensing a crash. Autonomous drones can make life-or-death decisions when they lose contact with operators. Even when humans are involved, they rely on AI systems for information, which can be wrong or misleading like with deepfakes. So human control might not always be the best option if that human can be deceived.

— [Biden's AI Plan And Why Central Control Might Not Work](https://www.forbes.com/sites/lutzfinger/2024/10/26/bidens-ai-plan--why-central-control-might-not-work/) · _Forbes_

Question 32

What problems with RAGs still need to be solved before widespread adoption?

Accepted Answer

_The observability and control challenges that remain in RAG systems._

RAGs don't completely stop hallucination yet. The model uses both the trained corpus and new data, but we don't have a very good solution to manage that difference. Who decides when new information should override training data? Is the LLM taking that decision or the RAG? We lack observability and control. For example, I created a document saying I'm Captain America, but ChatGPT rejected it as fake. What if that was actually true breaking news? Amazing startups will help create this observability, solve access control, and manage the difference between what the corpus was trained on and corporate datasets.

— [Are we one step closer to using LLMs in practice?](https://www.lutzfinger.com/transcript/2023-07-21-are-we-one-step-closer-to-using-llms-in-practice/) · _The Edge_

Question 33

What two essential skills must educators focus on when using AI?

Accepted Answer

_Understanding both AI's capabilities and its limitations in teaching._

We must focus on two essential skills: how to use AI effectively, and how to recognize its limits. AI models like ChatGPT excel at predictable, patterned tasks but struggle with novelty and edge cases. AI is a powerful tool, but it's still just a tool. The future of education is not about letting ChatGPT teach students directly. It's about educators designing learning paths and then using AI to scale their impact. The core responsibility remains with educators.

— [AI In Education — Up-level Your Teaching With AI By Cloning Yourself](https://www.forbes.com/sites/lutzfinger/2025/03/31/ai-in-educationup-level-your-teaching-with-ai-by-cloning-yourself/) · _Forbes_

Question 34

What unintended consequences can quality metrics create in healthcare?

Accepted Answer

_Mortality metrics can discourage doctors from treating complex cases._

As Michael Howell reminded me, the real challenges are in finding the metrics that matter and preventing unintended consequences with any given metric. Take a simple metric on mortality for surgery: patients should not die during surgery. But what if the patient's situation was complex to start with? There are cases where a patient did not receive care and died because giving him appropriate care could have resulted in a worse quality score for the doctor.

— [Healthcare And Data - How To Measure Quality](https://www.forbes.com/sites/lutzfinger/2023/06/27/healthcare-and-datahow-to-measure-quality/) · _Forbes_

Question 35

Why did celebrity bots like Meta's Snoop Dogg bot fail to gain traction?

Accepted Answer

_Why copying celebrities creates an uncomfortable user experience_

The celebrity bots fell into the uncanny valley problem. People got annoyed that the bot was a bit like Snoop Dogg but not really Snoop Dogg. When you go to someone like Snoop for authenticity, a bot cannot deliver that authentic experience. Fans know they never spoke privately to Snoop Dogg, they only know his public persona. The bot might be fun at first, but it becomes creepy when something tries to be too human without actually being human. Zuckerberg eventually shut down the celebrity bots because they didn't get enough traction.

— [The future of AI bots: winning with personalization, conviction, and cross-platform integration](https://www.lutzfinger.com/transcript/2024-08-22-the-future-of-ai-bots-winning-with-personalization-conviction-and-cross-platform/) · _The Edge_

Question 36

Why did many generative AI projects fail to scale in 2024?

Accepted Answer

_The two fatal flaws in most 2024 AI implementations._

Many companies rushed to try generative AI in 2024, and some even used chained models. But most projects failed to scale for two main reasons. First, they were test projects not integrated into system infrastructure, meaning they existed in isolation rather than connecting to existing business systems. Second, they were missing guardrails and quality controls necessary for production deployment. Without proper integration and controls, these pilot projects could not move beyond experimentation to deliver real business value at scale.

— [AI Agents In 2025: What Enterprise Leaders Need To Know](https://www.forbes.com/sites/lutzfinger/2025/01/05/ai-agents-in-2025-what-enterprise-leaders-need-to-know/) · _Forbes_

Question 37

Why did you mention Google's 2008 prediction capabilities and WIRED's 'end of theory' article?

Accepted Answer

_History shows prediction capability doesn't eliminate need for theory._

In 2008, excited from Google's ability to predict the next click, WIRED magazine proclaimed 'the end of theory,' saying that all that was needed is data. That proved to be wrong. Now we have better models for the same approach with transformers doing next-token prediction, but we still have not seen the end of theory. This parallel shows that impressive prediction capabilities, whether for clicks or tokens, don't replace the need for underlying theory and understanding. It's a reminder that we've been at this place before with prediction hype.

— [Healthcare's AI Lesson: Autocomplete Isn't Understanding](https://www.forbes.com/sites/lutzfinger/2026/01/16/healthcares-ai-lesson-autocomplete-isnt-understanding/) · _Forbes_

Question 38

Why did YouTube transcribe 'Merz' as 'Merkel' when Trump introduced the German Chancellor?

Accepted Answer

_AI leans toward historical patterns, not current reality._

YouTube's AI transcription system made this mistake because for 16 years, 'German Chancellor' almost always meant Angela Merkel in the training data. When the model heard 'Merz,' which is acoustically close but far less common, it leaned toward the familiar, high-probability continuation: 'Merkel.' The model plays the odds based on historical patterns. It's like predictive text suggesting 'Year' after 'Happy New' instead of 'Birthday.' The AI isn't wrong in a statistical sense, it's just out of sync with the real-world moment.

— [AI Fail — Trump Says 'Merz' —YouTube Transcribes 'Merkel'](https://www.forbes.com/sites/lutzfinger/2025/08/18/ai-fail---chancellor-merz-is-not-chancellor-merkel/) · _Forbes_

Question 39

Why do AI models still need human involvement even with unsupervised learning?

Accepted Answer

_Even advanced models can't escape the need for human judgment._

We need humans because there isn't always a true north for what's correct. Supervised learning works when there's a clear label, like Obama was president. That's a fact. But when you get into perception, like whether a joke is funny or which word to use about Obama's policies, it depends on context and perspective. The same happens with generated images. Art often creates things society didn't expect, so at what point is something random versus amazing? Since there's not a true label for certain things, we constantly need people to give feedback on the model and provide guardrails.

— [AI and the human factor](https://www.lutzfinger.com/transcript/2023-07-07-ai-and-the-human-factor/) · _The Edge_

Question 40

Why do people use AI even when they express concerns about it in surveys?

Accepted Answer

_The gap between AI anxiety and behavior reflects human nature_

The gap between stated anxiety and actual behavior is not hypocrisy. It's just how humans work. We accept risks all the time when the benefit feels immediate and the risk feels diffuse. People worry about job displacement, privacy, hallucinations, and dependency. But the cost of non-adoption feels more concrete than the cost of risk. If you don't use AI and your competitor does, you lose. If you don't use AI and your colleague does, you fall behind. The fear is real, but the friction of not using it feels worse.

— [Chris Hemsworth Super Bowl Ad Shows Why We'll Use AI Even If We Fear It](https://www.forbes.com/sites/lutzfinger/2026/02/10/even-if-ai-could-kill-chris-hemsworth-convenience-beats-fear/) · _Forbes_

Question 41

Why does Lutz compare requiring AI explainability to regulating cars not to scare horses?

Accepted Answer

_Some regulatory demands are technically impossible and counterproductive._

Large language models rely on probability distributions and they tend toward the statistical middle, making it difficult to explain AI's reasoning. Asking for explainability is like asking cars not to scare horses, which was once a real regulatory demand. If regulations require the impossible, we risk stifling AI development. Regulation should focus on what is feasible rather than demanding something the technology fundamentally cannot provide. This is part of why we are over-regulating.

— [Emmanuel Macron On AI: We Are Regulating Ourselves Out Of The Market](https://www.forbes.com/sites/lutzfinger/2024/10/22/emmanuel-macron-is-right-about-ai-over-regulatingi-have-no-doubt/) · _Forbes_

Question 42

Why does OpenAI argue that AI is extremely dangerous and should be restricted?

Accepted Answer

_Safety rhetoric as a competitive moat building strategy_

OpenAI's storyline that AI is extremely dangerous and only a select few should be allowed to develop it is meant to use safety concerns as justification for a regulatory setup that creates barriers to entry and gives them more monopolistic power. This is part of Sam Altman's classical playbook to protect their business. All of OpenAI's core statements over the past few months come down to one focus: OpenAI tries to protect their market position.

— [Is OpenAI A Victim — DeepSeek And IP Rights — A Playbook For Leaders](https://www.forbes.com/sites/lutzfinger/2025/01/29/is-openai-a-victim---deepseek-and-ip-rights---a-playbook-for-leaders/) · _Forbes_

Question 43

Why don't people trust Facebook with their data anymore?

Accepted Answer

_Trust erodes when data sharing is careless and opaque._

We don't trust Facebook anymore because we learned that they gave away our data way too carelessly. The real issue isn't about who has access to our data, but rather what the data is actually used for. Since we don't know what happens with our data once Facebook has it, trust has eroded. The outcry over WhatsApp's updated privacy policy, which disclosed data sharing with Facebook, reflects this deeper mistrust. It doesn't matter what Facebook does at this point because the trust is already broken.

— [WhatsApp: We Should Discuss What Our Data Is Used For, Not Who Has It](https://www.forbes.com/sites/lutzfinger/2021/01/26/whatsapp-we-should-discuss-what-our-data-is-used-for-not-who-has-it/) · _Forbes_

Question 44

Why is education the most important defense against deepfakes?

Accepted Answer

_Like Nigerian Prince scams, awareness makes people harder to fool._

Education is the best protection against deepfakes, just like with Nigerian Prince scams. Many people won't fall for those scams anymore because they've heard about them. We need the same degree of education for deepfakes. Examples like Channel 4 in the UK creating a deepfake of the Queen's speech bring deepfakes into public discussion. At Cornell, I start each term with a one minute deepfake video welcoming MBA students. It's fun, but it's also a reminder that great technology can be abused. User awareness is critical.

— [Deepfakes - The Danger Of Artificial Intelligence That We Will Learn To Manage Better](https://www.forbes.com/sites/lutzfinger/2022/09/08/deepfakesthe-danger-of-artificial-intelligence-that-we-will-learn-to-manage-better/) · _Forbes_

Question 45

Why is emotion detection training data problematic, according to the article?

Accepted Answer

_Algorithms learn performed emotions, not real ones, from limited datasets._

Emotion detection algorithms were often trained on the JAFFE dataset, which contains images of 10 Japanese female models performing seven facial expressions. The problem is the algorithm doesn't learn how someone actually looks when angry, but rather how those specific women performed an angry expression. This distinction is critical. The models were acting out emotions rather than genuinely experiencing them, so the training data captures performance, not authentic emotional states. This makes the foundation of emotion detection fundamentally flawed.

— [It's The Data, Stupid! Why AI Might Get It Wrong.](https://www.forbes.com/sites/lutzfinger/2020/05/28/its-the-data-stupid-why-ai-might-get-it-wrong/) · _Forbes_

Question 46

Why is energy the AI race?

Accepted Answer

_Energy and infrastructure as constraints._

> What wins the AI race? UX? Access to users?

— [The Missing Moat In AI: Your Eval Data](https://www.forbes.com/sites/lutzfinger/2026/05/26/the-missing-moat-in-ai-your-eval-data/) · _Forbes_

Question 47

Why is memory the next AI battleground, not prompts or agentic systems?

Accepted Answer

_AI memory systems create new risks and control challenges._

Memory is essentially part of our context window. Models now start to keep information about our memory. I asked ChatGPT about using AI in student interviews and it scolded me, saying as a faculty member I should be extremely careful. But even worse, it noted down in my memory that I wanted to use that tool. That's pretty much a good idea for a big disaster. How do I as an organization ensure that my memory is kept in the correct form? What is the memory you want to capture?

— [From Google to Cornell: Rethinking AI Workflows | Lutz Finger | Build AI Podcast 17](https://www.lutzfinger.com/transcript/2026-04-16-from-google-to-cornell-rethinking-ai-workflows-lutz-finger-build-ai-podcast-17/) · _Build AI_

Question 48

Why isn't autocomplete understanding in healthcare AI?

Accepted Answer

_LLM limitations in clinical settings._

> I just got back from the JP Morgan conference. A few learnings that apply to any industry, not just healthcare:Humans + AI will change the industry. Not AI alone.Cash flow matters. Not AI hype.The real AI revolution is operational.One of the most hyperbolic promises in healthcare in the last month was the claim that: "AI is going to design our drugs and will replace researchers." In my view, this is not true.

— [Healthcare's AI Lesson: Autocomplete Isn't Understanding](https://www.forbes.com/sites/lutzfinger/2026/01/16/healthcares-ai-lesson-autocomplete-isnt-understanding/) · _Forbes_

Question 49

Why shouldn't dealerships use ChatGPT or other AI tools to create content for GEO?

Accepted Answer

_AI-generated content harms your visibility in generative search results._

Do not ask ChatGPT, Gemini, or other tools to create content for you. It will not work. All model providers are extremely careful about trying to figure out slop, which is AI-generated content. The model doesn't learn from content it just created. This is the snake eating its own tail. If you create slop, you will lose out. Use your human team to bring content in. Your content needs to be authentic, novel, and unique. You can use AI tools to help organize your authentic content, like generating Q&As from your existing newsletters, but not to create the original content.

— [Lutz Finger | From SEO to GEO: How AI is Changing How Car Buyers Find You](https://www.lutzfinger.com/transcript/2026-05-22-lutz-finger-from-seo-to-geo-how-ai-is-changing-how-car-buyers-find-you/) · _Three Key Insights_

Q&A, Risks

How accurate are current deepfake detection technologies?

How did AI models respond when asked to create a plan for a U.S. president's third term?

How did we learn that location data from our phones was being sold?

How do biases in narrow AI systems create real-world harm?

How do you spot social media bots?

How does Catalog prevent large language models from hallucinating when retrieving enterprise information?

How might data regulation affect innovation and global competition?

Is AGI near?

Is OpenAI going bankrupt?

Is the content AI produces about controversial topics actually new information?

Should you trust OpenAI blindly?

What about deepfakes?

What are some of the dangerous ways deepfakes could be misused beyond pornography?

What are the fundamental limits of large language models that people should understand?

What are the main limitations of large language models that knowledge graphs help address?

What are the main problems with using simple GPT wrappers for enterprise AI applications?

What are the real AI risks that Europe should be concerned about?

What are the three main problems with training data that can cause AI to get things wrong?

What deepfake risks does Goku introduce and how should society respond?

What did the Harvard study with executives reveal about relying on AI for predictions?

What framework helps assess whether imperfect AI decisions are acceptable?

What fundamental capability do AI transcription models lack that causes errors like this?

What happened in Eric Loomis's case regarding algorithmic transparency and trade secrets?

What happened when Meta's AI system changed True Classic's ad creative?

What is the hallucination problem with large language models and why does it matter for businesses?

What is the ultimate lesson about human responsibility in the age of AI summarization?

What is your perspective on AGI and when it might happen?

What kinds of personal characteristics can be predicted from Facebook data?

What lessons from social media regulation are shaping today's AI discussions?

What pattern appears consistently in data about AI adoption and concern?

What problems does Lutz see with the 'human in the loop' concept in Biden's plan?

What problems with RAGs still need to be solved before widespread adoption?

What two essential skills must educators focus on when using AI?

What unintended consequences can quality metrics create in healthcare?

Why did celebrity bots like Meta's Snoop Dogg bot fail to gain traction?

Why did many generative AI projects fail to scale in 2024?

Why did you mention Google's 2008 prediction capabilities and WIRED's 'end of theory' article?

Why did YouTube transcribe 'Merz' as 'Merkel' when Trump introduced the German Chancellor?

Why do AI models still need human involvement even with unsupervised learning?

Why do people use AI even when they express concerns about it in surveys?

Why does Lutz compare requiring AI explainability to regulating cars not to scare horses?

Why does OpenAI argue that AI is extremely dangerous and should be restricted?

Why don't people trust Facebook with their data anymore?

Why is education the most important defense against deepfakes?

Why is emotion detection training data problematic, according to the article?

Why is energy the AI race?

Why is memory the next AI battleground, not prompts or agentic systems?

Why isn't autocomplete understanding in healthcare AI?

Why shouldn't dealerships use ChatGPT or other AI tools to create content for GEO?