Training data
The labeled dataset used to teach a machine learning system to recognize patterns; its composition determines the system's behavior and biases.
As defined in 156 | Catherine D'Ignazio on Data, Objectivity, and Bias
Other definitions
The collection of examples and information used to teach an AI system, upon which neural networks learn statistical patterns and language models develop their word-prediction capabilities
Is AI's "intelligence" an illusion? | GZERO World with Ian BremmerHistorical examples with known outcomes used to fit a predictive model.
Quant in Private Markets
Where it comes up · 15
- This is not the AI we were promised | The Royal Society6 claims
- Yann LeCun Says LLMs Have 2 Years Left…3 claims
- Is AI's "intelligence" an illusion? | GZERO World with Ian Bremmer3 claims
- Quant in Private Markets2 claims
- Zvi Mowshowitz on AI and the Dial of Progress2 claims
- What Does "Unbiased" Mean in the Digital World? (with Megan McArdle)2 claims
- 324 - Debating The Future Of Ai2 claims
- Dave Girouard - Making Better Loans - [Founder’s Field Guide, EP. 52]2 claims
and 7 more
Most important claims · 10 of 29Search all claims for “Training data” →
- A tech company training a hiring algorithm on its own history—defining success as staying 3 years with two promotions—could end up rejecting all women applicants; the algorithm would not flag that the company culture is unwelcoming to women, it would simply conclude 'women do not succeed here, throw them out,' because the algorithm is dumb: it understands 'what happened,' not 'why.'Cathy O'Neil on Weapons of Math Destruction
- Because LLMs are probabilistic and predict the most likely next item, an image model asked for a doctor might default to the majority demographic and thereby produce even more under-representation than actually exists, which is one legitimate reason to deliberately curate training data toward greater diversity.What Does "Unbiased" Mean in the Digital World? (with Megan McArdle)
- A core problem in combating misinformation with AI is the lack of training data, because AI needs many labeled good/bad examples to generalize, and society lacks agreement on what even counts as misinformation, leaving us uncertain whether deepfake detection will keep pace with deepfake generation.280 - The Future Of Artificial Intelligence
- There is a self-undermining dynamic in human creativity under AI: the more clever and human the creative work we produce (such as the Keynes/Hayek rap videos), the more high-quality material AI has to learn from and ultimately imitate, so as we try to match and outdo the machines they learn from us and continually raise the bar.Ian Leslie on Being Human in the Age of AI
- Large language models analyze the statistical relationship between words in their training data rather than analyzing concepts, ideas, or entities in the world, functioning essentially as 'autocomplete on steroids' that predict the most likely next word based on learned patterns.Is AI's "intelligence" an illusion? | GZERO World with Ian Bremmer
- AI bias can enter at three distinct points: (1) curation of the original training data, including aspirational over-representation and Left-leaning source moderation like Reddit; (2) human raters whose instructions or own selected demographic skew Left; and (3) manual hard-coding of responses—with the third being the most disturbing because it signals a deeper institutional sickness.What Does "Unbiased" Mean in the Digital World? (with Megan McArdle)
- Off-the-shelf facial recognition systems fail to detect dark-skinned faces (as Joy Buolamwini found, only recognized when she wore a white mask) not because engineers are racist but because benchmark training data was roughly 88% 'pale and male'—drawn from celebrity and political profiles—so the systems work well for white men and badly for Black women; the algorithm isn't biased so much as we are, garbage in, garbage out.156 | Catherine D'Ignazio on Data, Objectivity, and Bias
- Much of what we call human creativity is just a person having different training data and different mental connections, exploring a different region of idea-space, so their output seems creative to others; GPT is relatively weak at the synthesizing, world-reframing creativity (like coining a novel metaphor) but is not at zero and will improve.Zvi Mowshowitz on AI and the Dial of Progress
- Current LLM systems lack transparency regarding their training data sources and mechanisms of political influence, which combined with their potential to affect everyone's political opinions creates a public interest justifying mandatory oversight even for users who did not consent to the system.Is AI's "intelligence" an illusion? | GZERO World with Ian Bremmer
- Despite significant investment in driverless car technology, after 7 years since 2016 the industry remains plagued by the 'outlier problem'—scenarios not represented in training data that the systems cannot handle—exemplifying how capital investment alone does not guarantee technological breakthroughs.Is AI's "intelligence" an illusion? | GZERO World with Ian Bremmer