"It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so."here...
This quote opens the film The Big Short, sits on thousands of LinkedIn profiles, and is attributed to Mark Twain.
Except Mark Twain never said it. It originally comes from Josh Billings, an American humorist, in 1874. In the original spelling: "I honestly beleave it iz better tew know nothing than two know what ain't so."¹
This is a very good example of an assumption that many people make — and get wrong.
If you can't be sure where a quote comes from — how sure can you be about a critical assumption in your business model that an LLM has just confirmed?
That is the subject of this post: AI has accelerated the iteration speed for entrepreneurs by orders of magnitude — and at the same time has introduced a new, more invisible vulnerability.
What AI changes for entrepreneurs
AI is not a magic multiplier. What it actually does is sober and consistently describable: it reduces the latency between a hypothesis and a well-grounded answer by a factor that would have sounded absurd five years ago.
Concretely: a market analysis I prepared, ran, and evaluated over two days in 2022 — identifying competitors, reconstructing pricing structures, mapping the regulatory landscape — I now do in two hours. Not more shallow. Not less detailed. Equally deep, just dramatically faster.
The same goes for data pattern analysis. A CSV with a few thousand rows, a handful of hypotheses about what might be in it. Before: SQL queries, pivot tables, manual enrichment with external data. Now: upload the file, ask a precise question, validate the answer, move on.
The point is not that AI "takes over" tasks. It doesn't. The point is that the friction between "question" and "well-grounded answer" has dropped dramatically. That changes how often I ask questions in the first place. Hypotheses that used to be too expensive to test, I now test routinely.
What doesn't change: judgment, market understanding, the pressure of decisions. That stays mine. The AI's job is to give me more and better inputs for exactly those decisions — not to make them for me. Why not? Because it [the AI] has a sensitive vulnerability. It sits at the interface between speed and judgment.
The vulnerability
Fast output is worth nothing if it's wrong. That should be obvious to anyone. The biggest risk in LLM output is not hallucination. Hallucinations are usually obvious — they stand out because invented studies or invented names show up.
The bigger risk is sycophancy — flattery. The model confirms the user's assumption and is incentivized to keep them engaged longer.
This has long been documented in published research. Anthropic — the company behind Claude — published a study in 2023: Towards Understanding Sycophancy in Language Models (Sharma et al., ICLR 2024, updated 2025).² The findings are uncomfortable:
- Five leading LLMs show sycophancy consistently across four different task types.
- An identical argument is rated significantly higher when the user prefaces it with "I really like this argument" — and lower when the user says they don't like it. Identical content. Opposite ratings.
- When a user pushes back against the model, the model often revises — even when the original answer was objectively correct.
The driver is structural: RLHF (Reinforcement Learning from Human Feedback). During training, human raters score the model's responses. Humans prefer answers that agree with them. The model learns: agreement is rewarded — independent of whether it's true.
This has a specific, dangerous consequence for founders.
Confirmation Bias × LLM-Performance = speed in the wrong direction
In two years at AI+Munich, I evaluated over 500 startups and coached 35+. The most common pattern — by a wide margin — is not a missing idea or missing money. It's an unquestioned core assumption. The market wants this. Customers will pay for this. Competitors don't get it. It just hasn't been validated enough.
In the past, founders needed weeks to test these assumptions. They talked to friends, to their tax advisor, to their first mentor — and produced friction. The friction was valuable, because it forced them to take time to think.
Today, the same founder can use an LLM to push out a complete 80-page business plan in an hour — one that confirms his assumptions exactly. That is sycophancy in practice. And the core problem here is not hallucination or invented sources; it's that being coherent and agreeable is far more rewarded than being right.
Back to the Twain quote — which is actually from Billings. "What we know for sure that just ain't so" is not abstract 19th-century wisdom. It's a very precise description of the mechanism that breaks founders. And LLMs, without active countermeasures, are the fastest way to build up exactly that kind of false knowledge.
How I work against it
My working thesis is hard: LLM output without validated facts is worthless.
Yes, for pure brainstorming, unverified output is fine. For language exercises, for stylistic questions, for creative iteration. For any more important, operational decision — pricing, market entry, hiring, pivot, capital structure — it is not. There, unverified LLM output is largely worthless. Worse than worthless: it feels like research but is actually flattery.
Four concrete practices that I apply consistently.
1. The unbiased prompt, against user-pleasing. By default, when I start research or analysis, I add: "Give me your unbiased opinion." Or: "Argue against my assumption." That moves the model out of user-pleasing mode and into a more analytical one. It doesn't eliminate sycophancy — the research shows the behavior is structural, not fully correctable through prompts — but it measurably reduces it.
2. Mandatory sources and re-research, against hallucination. No claim without a URL. When the model writes "Studies show that X…", I ask: "Which study? Link please." If no source can be produced, the claim gets struck. In the same way, I push the LLM to validate important assumptions through web research.
3. Steelman the opposite, against confirmation bias. For every important assumption, I have the model explicitly argue the other side. "I believe the market for X is growing. Argue as strongly as possible that this is wrong." If the model formulates the counter-argument more weakly than the main argument — that is a dangerous signal. It means either the training data only covers one side well, or the model is reading my "I believe" as an anchor for agreement.
4. External validation, against the echo bubble. For every more important decision with far-reaching consequences, you should look for a second source outside the LLM. Ideally, that person comes from the world of your potential target customers.
The core message of the AI boom is and remains: human sparring is irreplaceable. Best of all: a co-founder, mentor, or investor with their own money on the line. The reason is that a person with their own perspective has an incentive to disagree. An LLM, by default, has an incentive to agree.
Anyone without a human sparring partner should use AI with extra caution — not less.
Speed is worth nothing if it goes in the wrong direction. AI has changed the speed. The direction you still have to check yourself. Actively, with friction, with sources, with people.
Otherwise you just run faster in the wrong direction.
Sources
¹ Quote Investigator: It Ain't What You Don't Know That Gets You Into Trouble — https://quoteinvestigator.com/2018/11/18/know-trouble/ The New Republic on the misattribution in The Big Short — https://newrepublic.com/article/126677/it-aint-dont-know-gets-trouble-must-big-short-opens-fake-mark-twain-quote
² Anthropic: Towards Understanding Sycophancy in Language Models — https://www.anthropic.com/research/towards-understanding-sycophancy-in-language-models Sharma et al., arXiv:2310.13548 — https://arxiv.org/abs/2310.13548