Zen

Source string Source string

391/10000 · 391

The string is read only.

English Actions
<g id="45"><g id="46">Find out how LLMs are built and how they work—and then you can decide how much you want to rely on them. </g></g>
<g id="47"><g id="48">2</g></g>
<g id="49"><g id="50">1 </g></g>
<g id="51"><g id="52">LLMs start by ‘learning’ 
single words.</g><x id="53" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="54"><g id="55">By analysing other data related to that word, it can predict what should come 
next in the sentence. </g></g>
<g id="56"><g id="57">After enough attempts and with enough data, LLMs can generate content that </g></g><g id="58"><g id="59">sounds</g></g><g id="60"><g id="61"> human. Some experts say they aren’t sure exactly how it works—one even said: “We just have no idea what any of it means.”</g></g>
<x id="62" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="63"><x id="64" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="65">PIGEON</g></g>
<x id="66" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="67"><x id="68" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="69">NEST</g></g>
<x id="70" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="71"><x id="72" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="73">EGG</g></g>
<x id="74" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="75"><x id="76" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="77">WING</g></g>
<x id="78" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="79"><x id="80" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="81">WORM</g></g>
<x id="82" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="83"><x id="84" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="85">CHICK</g></g>
<x id="86" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="87"><x id="88" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="89">TWEET</g></g>
<x id="90" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="91"><x id="92" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="93">CAT</g></g>
<x id="94" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="95"><x id="96" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="97">BIRD</g></g>
<g id="98"><g id="99">An LLM is only as good as the ‘diet’ of data it’s trained on. If that diet partly comes from the armpit of the internet, it could stink. </g><x id="100" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="101">The most intelligent-sounding tool isn’t necessarily all-knowing... It’s just programmed to seem like it.</g></g>
<g id="102"><g id="103">Still not that bad? Well, those data sets might </g></g><g id="104"><g id="105">also</g></g><g id="106"><g id="107"> be:</g></g><g id="108"><g id="109"/><x id="110" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[3]/{}Br[0]"/></g>
<g id="111"><g id="112">Full of hate speech, conspiracy theories or information that is just </g></g><g id="113"><g id="114">plain wrong</g></g><g id="115"><g id="116">.</g><x id="117" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[2]/{}Br[0]"/></g><g id="118"><g id="119">Biased</g></g><g id="120"><g id="121"> against certain jobs, cultures, religions, genders, or circum­stances.</g><x id="122" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[4]/{}Br[0]"/></g>
<g id="123"><g id="124">LLMs are trained on data that also leaves stuff out altogether. If there’s little or no information about a group of people, language or culture in the LLM’s training data, it won’t be able to generate any answers about them.</g></g>
<g id="125"><g id="126">LLMs also learn from what you tell them and what you type, to predict what you want to hear. This is what makes them seem like they have a ‘personality’.</g></g>
<g id="127"><g id="128">LLMs have been fed so much data that they generate essays and ace exams. But they also present things as fact that aren’t true.</g><x id="129" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="130">While some experts believe LLMs are getting ‘smarter’ on their own, others say they’re not actually understanding the words they repeat.</g><x id="131" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[1]"/><g id="132">Do you really want to trust what a chatbot tells you? </g><x id="133" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[2]"/></g>
<g id="134"><g id="135">How does the model ‘learn’ from data? Well, it’s kind of like auto- complete: </g></g>
<g id="136"><g id="137">AI </g><x id="138" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="139">or BS?</g></g>
<g id="140"><g id="141">WTF </g><x id="142" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="143">LLM?</g></g>
<g id="144"><g id="145">While it may seem like LLMs know everything on the internet, that doesn’t make them all-knowing. Here are some reasons why:</g><x id="146" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="147"><g id="148">Some of the data they are </g></g><g id="149"><g id="150">trained </g></g><g id="151"><g id="152">on might be </g></g><g id="153"><g id="154">personal</g></g><g id="155"><g id="156">, </g></g><g id="157"><g id="158">copyrighted or used without 
</g></g><g id="159"><g id="160">permission.</g><x id="161" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[6]/{}Br[0]"/></g>
<g id="162"><g id="163">Garbage In, Garbage Out</g><x id="164" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="165"><g id="166">But before this data can be useful for training the LLM, it has to be labelled and categorised by loads of human workers.</g></g>
<g id="167"><g id="168">Can you guess which kinds of data are used to train LLMs?</g></g>
<g id="169"><g id="170">Feed 
the Bot</g></g>
<g id="171"><g id="172">You probably already know that LLMs (‘Large Language Models’ like ChatGPT and Bard) are more than just search engines. </g><x id="173" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g><g id="174"><g id="175">Maybe you use LLMs to plan a trip, to get advice or even to do your homework.</g><x id="176" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[1]/{}Br[0]"/></g><g id="177"><x id="178" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[2]/{}Br[0]"/></g>
English Actions
<g id="33"><g id="34">ONLINE SHOPS</g><x id="35" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="171"><g id="172">You probably already know that LLMs (‘Large Language Models’ like ChatGPT and Bard) are more than just search engines. </g><x id="173" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g><g id="174"><g id="175">Maybe you use LLMs to plan a trip, to get advice or even to do your homework.</g><x id="176" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[1]/{}Br[0]"/></g><g id="177"><x id="178" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[2]/{}Br[0]"/></g>
<x id="90" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="91"><x id="92" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="93">CAT</g></g>
<g id="47"><g id="48">2</g></g>
<g id="49"><g id="50">1 </g></g>
<g id="179"><g id="180">These AI chatbot systems can be helpful... but they’re far from perfect. While they may seem like an ‘everything tool,’ it’s good to know what they’re best at and where they fall short. </g></g>
<g id="136"><g id="137">AI </g><x id="138" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="139">or BS?</g></g>
<g id="45"><g id="46">Find out how LLMs are built and how they work—and then you can decide how much you want to rely on them. </g></g>
<x id="74" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="75"><x id="76" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="77">WING</g></g>
<g id="127"><g id="128">LLMs have been fed so much data that they generate essays and ace exams. But they also present things as fact that aren’t true.</g><x id="129" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="130">While some experts believe LLMs are getting ‘smarter’ on their own, others say they’re not actually understanding the words they repeat.</g><x id="131" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[1]"/><g id="132">Do you really want to trust what a chatbot tells you? </g><x id="133" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[2]"/></g>
<g id="12"><g id="13">WIKIPEDIA ENTRIES</g><x id="14" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="42"><g id="43">PATENT WEBSITES </g><x id="44" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<x id="94" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="95"><x id="96" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="97">BIRD</g></g>
<g id="162"><g id="163">Garbage In, Garbage Out</g><x id="164" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="98"><g id="99">An LLM is only as good as the ‘diet’ of data it’s trained on. If that diet partly comes from the armpit of the internet, it could stink. </g><x id="100" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="101">The most intelligent-sounding tool isn’t necessarily all-knowing... It’s just programmed to seem like it.</g></g>
<g id="56"><g id="57">After enough attempts and with enough data, LLMs can generate content that </g></g><g id="58"><g id="59">sounds</g></g><g id="60"><g id="61"> human. Some experts say they aren’t sure exactly how it works—one even said: “We just have no idea what any of it means.”</g></g>
<g id="27"><g id="28">RECAPTCHA </g><x id="29" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="144"><g id="145">While it may seem like LLMs know everything on the internet, that doesn’t make them all-knowing. Here are some reasons why:</g><x id="146" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="147"><g id="148">Some of the data they are </g></g><g id="149"><g id="150">trained </g></g><g id="151"><g id="152">on might be </g></g><g id="153"><g id="154">personal</g></g><g id="155"><g id="156">, </g></g><g id="157"><g id="158">copyrighted or used without 
</g></g><g id="159"><g id="160">permission.</g><x id="161" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[6]/{}Br[0]"/></g>
<g id="30"><g id="31">VOTER REGISTRATION DATABASES</g><x id="32" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="0"><g id="1">:</g></g>
<g id="24"><g id="25">CROWDFUNDING WEBSITES</g><x id="26" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="140"><g id="141">WTF </g><x id="142" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="143">LLM?</g></g>
<g id="51"><g id="52">LLMs start by ‘learning’ 
single words.</g><x id="53" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="111"><g id="112">Full of hate speech, conspiracy theories or information that is just </g></g><g id="113"><g id="114">plain wrong</g></g><g id="115"><g id="116">.</g><x id="117" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[2]/{}Br[0]"/></g><g id="118"><g id="119">Biased</g></g><g id="120"><g id="121"> against certain jobs, cultures, religions, genders, or circum­stances.</g><x id="122" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[1]/{}CharacterStyleRange[4]/{}Br[0]"/></g>
<g id="39"><g id="40">DIGITAL LIBRARIES</g><x id="41" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>
<g id="54"><g id="55">By analysing other data related to that word, it can predict what should come 
next in the sentence. </g></g>
<x id="86" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}Properties[0]"/><g id="87"><x id="88" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Properties[0]"/><g id="89">TWEET</g></g>
<g id="15"><g id="16">NEWS WEBSITES</g><x id="17" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/></g>

Loading…

User avatar None

String added in the repository

  1. Everywhere, All the Time
  2. 05_EAT-LLM-Poster_A3-A4_EN
  3. English
<g id="98"><g id="99">An LLM is only as good as the ‘diet’ of data it’s trained on. If that diet partly comes from the armpit of the internet, it could stink. </g><x id="100" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="101">The most intelligent-sounding tool isn’t necessarily all-knowing... It’s just programmed to seem like it.</g></g>
<g id="98"><g id="99">An LLM is only as good as the ‘diet’ of data it’s trained on. If that diet partly comes from the armpit of the internet, it could stink. </g><x id="100" xid="idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0]/{}CharacterStyleRange[0]/{}Br[0]"/><g id="101">The most intelligent-sounding tool isn’t necessarily all-knowing... It’s just programmed to seem like it.</g></g>
a year ago
Browse all string changes

Things to check

Ellipsis

The string uses three dots (...) instead of an ellipsis character (…)

Reset

Glossary

English English
No related strings found in the glossary.

String information

Flags
read-only, strict-same
Source string location
idPkg:Story[0]/{}Story[0]/{}ParagraphStyleRange[0] • Stories/Story_u15cf.xml
String age
a year ago
Last updated
a year ago
Source string age
a year ago
Translation file
05_eat-llm-poster_a3-a4_en/en.idml, string 33