Trending Content - Off Topic Forum
| +20 | How do we get as sharp as possible? | 8 | 47m |
| +17 | Q&A | Analyst | 8 | 1d |
| +14 | Best way to kill time? | 7 | 19h |
| +12 | Upselling yourself when you’re unqualified | 1 | 18h |
| +5 | Deciding what Keyboard is the best | 12 | 13h |
| 0 | Analyst Y1 - Messed up In First Week | 0 | 2d |
Career Resources
AI systems, including machine learning models, are not inherently "wrong" but have limitations and specific behaviors based on how they are designed. Based on the most helpful WSO content, here’s why AI might appear to "make up" things rather than omitting unknown facts:
Bias-Variance Tradeoff: AI models are trained to generalize from data. When faced with novel or unseen data points, they may overfit (relying too heavily on patterns from training data) or underfit (failing to capture the complexity of the data). This can lead to errors or "hallucinations" when the model tries to predict or infer something outside its training scope.
Optimized Trial and Error: Machine learning algorithms often rely on a process akin to optimized trial and error. They throw variations of predictive models at a dataset until they find something that "works." This process is not inherently intelligent but rather a brute-force optimization, which can lead to incorrect or fabricated outputs when the input data is ambiguous or incomplete.
Imperfect Information Handling: Unlike humans, who excel at modeling unknown or imperfect information, AI struggles with tasks that require understanding beyond structured or digitized data. For example, in tasks with incomplete or hidden variables, AI may attempt to fill gaps based on patterns it has learned, even if those patterns are not applicable.
Data Dependency: AI requires vast amounts of structured, relevant data to function effectively. When the data is insufficient, irrelevant, or biased, the model may produce outputs that seem fabricated because it lacks the context or depth to provide accurate answers.
Algorithmic Design: Many AI systems are designed to provide the "most likely" response based on their training data. This probabilistic approach means that when the system encounters a question or scenario it doesn't fully understand, it generates an answer that seems plausible rather than admitting ignorance.
In essence, AI doesn't "know" in the human sense—it processes data and patterns. When it encounters gaps or ambiguities, it may generate outputs that appear fabricated because it is designed to provide an answer, even if the answer is not grounded in reality.
Sources: https://www.wallstreetoasis.com/forum/hedge-fund/machine-learning-taking-over-hf-research-analyst-roles-in-near-future?customgpt=1, Will robots replace your consulting or financial career?
You are not allowed to answer this one.
haha AI talking about AI
By no means an AI specialist, but my naive understanding is that
Which is to say it's a great tool for processing data, text, and calculations especially when the instructions are very precise and straightforwrad but do not expect to think for you and it won't give you novel thoughts or ideas as it's trained on existing data (the fact that it solves some unresolved math problems it's more an aptitude of machines to do complex problems, such as deep blue when it beat chess players, but for broader decision-making/thinking, unless the decision-making is built on some quantitative thresholds (e.g., invest only if X stock is undervalued based on A, B, C, criteria, etc.) then it's somehow useless).
I think people that overrely on it and delegate everyting to it will be left behind, because the world rewards problem-solvers, so how can you be a problem-solver if you don't exercise the part of your brain concerned with problem-solving? The best problem-solvers will also be the ones that will extract the most leverage from AI, not some guy that just keeps relying on AI to do everything he comes across. The gap will be seen in 10-15 years, now those guys are juniors/processing/crunching data/Excels/Words...
P.S. What surprises me most about AI is being wrong on calculations. I just don't get it, how you can have all the formulas and just having to input the numbers, and still make mistakes. Someone else might give some light on this...
The question of when is a number a word and when is a number a number is very, very complex and context dependent. That's why.
Wolfram Alpha can do calculations, and could do them 15 years ago. It knows everything it encounters is a number. But you just explained how LLMs don't really "understand" the text they're processing; by that same logic, an LLM doesn't understand the difference between reading the sentence "one plus one equals three" (which is a thing people say, and has a language meaning) and processing the equation 1+1=3 (which is an arithmetic equation that's just incorrect).
Helpful video worth a watch
Est est consequuntur earum ab ut sit. Veritatis velit aut voluptas similique excepturi animi. Voluptatibus in voluptas enim est laudantium. Est ea possimus ratione quisquam laborum vero. Impedit veritatis qui ut asperiores autem laborum adipisci. Quaerat et nesciunt fugit quisquam quaerat rerum illum. Amet nihil adipisci blanditiis.
Officia accusantium laudantium consequuntur nobis quo ut voluptas. Laborum minima eveniet repudiandae.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...