1(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[5029],{54821:function(e,t,n){(window.__NEXT_P=window.__NEXT_P||[]).push(["/100-words/day-177",function(){return n(54192)}])},54192:function(e,t,n){"use strict";n.r(t),n.d(t,{meta:function(){return s}});var i=n(85893),a=n(11151),o=n(53841);let s={author:"Andrew Miller",date:"2024-09-12",title:"Technical tests in the age of AI",description:"How do we allow for AI usage when evaluating candidates?",tags:["100 words","ai","hiring","technical tests","llms"]},MDXLayout=e=>(0,i.jsx)(o.B,Object.assign({meta:s},e));function _createMdxContent(e){let t=Object.assign({p:"p",ol:"ol",li:"li"},(0,a.ah)(),e.components);return(0,i.jsxs)(i.Fragment,{children:[(0,i.jsx)(t.p,{children:"Story time! When I was about a year into my first job at Ocado out of Uni, I was in the Netnix team (Networks & *nix). We needed to hire a few new sysadmin people and decided it was time to update the technical interview to be a bit more modern in that we wanted to see how they would actually solve some simulated problems and also do a bit of scripting as well."}),"\n",(0,i.jsx)(t.p,{children:"It was quite an involved test, we setup two laptops (one client, one server) and the test was a python flask app running on the web server with realtime feedback of the answers. Half of the test involved fixing bits of the machines and getting information from server, the other half was doing some scripting on the log files and we could watch the participant via VNC without being in the room watching them."}),"\n",(0,i.jsx)(t.p,{children:"All that to say we expected them to use the Internet to search for answers, it was very much the against the classic interview of having to know everything and acknowledged that you do research problems and tools in real life."}),"\n",(0,i.jsx)(t.p,{children:"Recently, as I may need to start hiring again in the near future, I am considering the same thing in regard to AI. Tools, including AI, used correctly are fine in my books, but how do we realistically incorporate there usage into the technical portion of the hiring process."}),"\n",(0,i.jsx)(t.p,{children:"Currently I have a couple of ideas:"}),"\n",(0,i.jsxs)(t.ol,{children:["\n",(0,i.jsx)(t.li,{children:"Have an AI generate some code with bugs in it for the candidate to fix (this isn't quite right, but a neat idea)"}),"\n",(0,i.jsx)(t.li,{children:"Create a custom GPT which is limited in scope for the test (this is too easy to get around)"}),"\n",(0,i.jsx)(t.li,{children:"Just let candidates use whatever they like, but like the original story, observe how they use the AI, if they do at all?"}),"\n"]}),"\n",(0,i.jsx)(t.p,{children:"Or something else entirely? I would love to hear your thoughts and ideas!"})]})}t.default=function(e={}){return(0,i.jsx)(MDXLayout,Object.assign({},e,{children:(0,i.jsx)(_createMdxContent,e)}))}}},function(e){e.O(0,[407,8819,4825,9774,2888,179],function(){return e(e.s=54821)}),_N_E=e.O()}]);
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.