TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Responding to recruiter emails with GPT-3

213 pointsby mbilover 2 years ago

40 comments

hcazzover 2 years ago
I&#x27;m sure some folks will see this as dishonest or unethical, but it feels more like the natural evolution of getting spammed by recruiters.<p>If recruiters are going to endlessly spam me with their templates and automation, it only seems fair that my response is also templates and automation.
评论 #32686346 未加载
评论 #32688785 未加载
评论 #32686074 未加载
评论 #32688865 未加载
Justsignedupover 2 years ago
As one article taught me about over a year ago respond by sending a well formulated email requesting the salary and job offering. Sometimes it leads to you understanding the market. Sometimes it leads to negotiation for higher salary. Sometimes an amazing new job.<p>So far this has never failed me. And is far better than ignoring. I only really ignore the crypto companies.
评论 #32687161 未加载
评论 #32686867 未加载
bilinguliarover 2 years ago
Taking into account that some sort of automated tool also generates recruiter letters and follow-ups - we witnessed a fight of two computers without any value in it for anyone.
评论 #32689264 未加载
评论 #32689229 未加载
评论 #32689255 未加载
woutr_beover 2 years ago
My response to recruiters (especially on LinkedIn) is always the same; “Please send over all relevant details for the role you had in mind (JD, company and salary)”.<p>This filters out about 90% of them, and especially those who’re just fishing for clients.
评论 #32689431 未加载
评论 #32688812 未加载
brothover 2 years ago
I felt like this blog post ended quite abruptly. I was expecting for the author to show how GPT-3 was able to formulate a response back to the recruiter not just parse out the name and company of the recruiter.
评论 #32686416 未加载
评论 #32686770 未加载
评论 #32686472 未加载
评论 #32686489 未加载
评论 #32689577 未加载
JasonFruitover 2 years ago
&gt; This solution worked well for the handful of emails I tried it on. I’m planning to run this on a cron to save myself some time and automatically maintain recruiter relationships.<p>Maintain recruiter relationships?! You&#x27;re sending a glorified autoreply that plugs in a couple values. That is not maintaining a relationship. Luckily, there&#x27;s no relationship to maintain, because the recruiter probably just blasted out an automatically-generated message to a list of contacts in the first place.<p>This sort of thing makes me sick. It&#x27;s an abominable distortion of human interaction, and it makes our society worse.
评论 #32691112 未加载
评论 #32697621 未加载
jasonhanselover 2 years ago
Someday, the recruiters will <i>also</i> start using GPT-3, and all recruitment conversations will just be transmissions between different GPT-3 instances, creating a massive hive mind of GPT-3s that brings about the singularity.
评论 #32685970 未加载
评论 #32686450 未加载
评论 #32688011 未加载
评论 #32685881 未加载
potatosalad1over 2 years ago
Now have it automatically negotiate a salary range high enough for you to consider interviewing.
评论 #32685947 未加载
评论 #32686791 未加载
评论 #32685925 未加载
mcbuilderover 2 years ago
Really smart use of GPT-3, if I read it write your prompt asks GPT-3 to use it&#x27;s magic powers of attention to figure out it&#x27;s the name and company name you&#x27;re looking for, then you further ask the response to be in the form of something you can easily extract. The first part of your prompt is the example (Apple computer), and the second part is the real email you want to parse and you give room for the AI to dump it&#x27;s answer. I really appreciate the smart ways people are coming up with using the NLP models.
stavrosover 2 years ago
I was curious how GPT-3 would fare with a more elaborate response, so I pasted a random recruiter email and asked it to reply with more information, if the information wasn&#x27;t in the email.<p>I am quite astonished by its performance, and this is coming from the guy who made an entirely AI-generated podcast (<a href="https:&#x2F;&#x2F;deepdreams.stavros.io" rel="nofollow">https:&#x2F;&#x2F;deepdreams.stavros.io</a>)<p><a href="https:&#x2F;&#x2F;www.pastery.net&#x2F;zdxyvy&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.pastery.net&#x2F;zdxyvy&#x2F;</a>
评论 #32692039 未加载
评论 #32688795 未加载
justusthaneover 2 years ago
This makes a lot more sense than having GPT-3 actually write responses, which is what I thought this was going to be. Not nearly as much fun though!<p>(But in seriousness, using GPT-3 for parsing to extract info is a neat use case I wasn’t aware of).
评论 #32686406 未加载
nullcover 2 years ago
&quot;ASK HN: My GPT-3 spam responder got hired for a job at Infosys. Now what?&quot;
评论 #32686278 未加载
评论 #32685849 未加载
dweeklyover 2 years ago
While the title doesn&#x27;t match the article content, what I really like about this is a notion of semantic regex - learning feature extraction was not a use case for GPT-3 I had imagined, and now my mind is reeling considering other ways this could be leveraged.
o_mover 2 years ago
I don&#x27;t get why you&#x27;ll need AI for this. Couldn&#x27;t you get the same result by using regex on the email from the sender?
评论 #32688827 未加载
bastawhizover 2 years ago
I explicitly list on my LinkedIn that I&#x27;m not interested in &quot;web3&quot; but that doesn&#x27;t stop the messages. This has me thinking of fun ways to waste a few minutes of these recruiters&#x27; time in return for wasting a few (many?) minutes of mine...
评论 #32685830 未加载
评论 #32687331 未加载
评论 #32686482 未加载
bitshiftfacedover 2 years ago
This brings to mind Permutation City (1994), where there exists interactive junk emails. The protagonist uses a program that automatically responds to these emails in a virtual environment in order to determine which ones are spam.
andaiover 2 years ago
Does this violate OpenAI&#x27;s terms of service? Or is it OK because you&#x27;re the only user so it isn&#x27;t a &quot;production application&quot;? They are quite clear that any AI produced text needs to be explicitly labeled so that nobody could mistake it for human written.<p>(I think this is hilarious, but just letting you know so you don&#x27;t get banned or something.)
评论 #32689391 未加载
CrociDBover 2 years ago
I like the plot twist where the GPT-3 use is not to generate the answer, but to query data from the email.
评论 #32687894 未加载
kesslykaesover 2 years ago
I spent some time in web scraping, and building extractors that work reliably on websites is damn near impossible. GPT-3 automatically extracting out the right data from text is kinda crazy. Imagine scraping LinkedIn without dealing with selectors that change every other day...
bradwoodover 2 years ago
I find my lowtech solution works best. Just don&#x27;t reply.
baryphonicover 2 years ago
This will be truly wild when most of the recruiters are GPT-3 bots as well.
shrimpxover 2 years ago
<p><pre><code> Hi {recruiter_name or &quot;&quot;}, </code></pre> This should be<p><pre><code> Hi{&quot; &quot; + recruiter_name or &quot;&quot;},</code></pre>
评论 #32698633 未加载
armchairhackerover 2 years ago
Has anyone tried responding to phishing &#x2F; scams with GPT-3? It seems like a much better use case, scammers are much more annoying and unethical. And you don&#x27;t have to worry about GPT3 giving a weird response (which idk but the author doesn&#x27;t seem to realize, machine learning even GPT is unreliable)
评论 #32686001 未加载
评论 #32688134 未加载
daniel-cussenover 2 years ago
I recently turned all the predictive email, the &quot;Nice to see you too! &#x2F;&#x2F; Yeah it was fun. &#x2F;&#x2F; See you!&quot; robot shit gmail proposes. I don&#x27;t want those proposals. The bias they introduce into my human (when it&#x27;s a human) interaction is impossible to undo, I turned all all that shit off, if I want to write &quot;Nice to see you! Yeah it was good. &#x2F;&#x2F; See you!&quot; I didn&#x27;t even read off the screen what I wrote the first time around, I just literally closed my eyes and typed from memory, incorrectly but so be it. I rather actually push the buttons that I supposedly type, do the work I allegedly do, type every character I supposedly type when it&#x27;s prose. It&#x27;s not a waste but even if it were there are much worse things than absolute waste.
评论 #32688437 未加载
nicbouover 2 years ago
I rarely get recruiter emails anymore, but SEO outreach emails are exactly the same.<p>I have a Gmail filter for them, and if something slips through, I block the sender. It takes a second and it ends the problem for good.<p>I would suggest doing the same with recruiters. Bot messages don&#x27;t deserve human courtesy.
daniel-cussenover 2 years ago
Sourceress on the flip side. And they use the long responses to their long emails to train more. Eh.<p>Recruiters have been doing this for years, stringing along applicants--ideally into turning down real offers from real people writing for real--with emotionally touching, thought-out...shit at random. To get to the point the applicant prematurely rejects all other offers and then--20 percentile salary offer, non-negotiable, &quot;and then we&#x27;ll see,&quot; and you know and I know you&#x27;re homeless if you don&#x27;t take it, but that never gets spoken out loud. Dude they look at all your financials, they build the equivalent of fraud detection to see how much they can lowball you, check your credit history (you knew that, they asked you for it), then they wanted all this endless information from other previous employers (but you never talk to their former employees, no boss&#x27;s resume of references that worked for him to see how they really felt about it). So in response, first, declare you forfeit all stock options, unconditionally. Then, ask to talk to the people you&#x27;ll be working with. Then, ultimatum offer, and if you end up on the street because of procedurally generated emotional commitment (which turns out is a liability in court, coming soon), hey, Daniel Cussen will still talk to you, I&#x27;ll still listen to you, I&#x27;ll give you when you beg, I&#x27;ll vouch for you as a witness, I&#x27;ll remember your name, I&#x27;m doing all those things where I live every day with all the awesome beggars--they have so many stories--that took the bait in some capacity.<p>I&#x27;ll treat you just like if it were the flip side, as good as I wish you treated me if I got evicted.<p>And after that, accept that it&#x27;s a job and subordination. And then do the job, it&#x27;ll suck it&#x27;s work, come on, most of the money still goes to people&#x27;s salaries (this is in a transition and is shifting dramatically, money spent on training is absolutely exploding at the expense of salaries, to my and therefore your monetary benefit if you end up on the street, <a href="https:&#x2F;&#x2F;www.fgemm.com" rel="nofollow">https:&#x2F;&#x2F;www.fgemm.com</a> coming soon).
palataover 2 years ago
&gt; Assuming I get four such emails per week and that it takes two minutes to read and respond to each one, automating this would save me about seven hours of administrative work per year.<p>Am I the only one thinking that it&#x27;s obviously not how it works? Not even considering the time it took implementing that.<p>I mean it&#x27;s a fun project, but I don&#x27;t see the need to justify it with &quot;it will make me more productive&quot;.<p>Fun is reason enough.
corytheboydover 2 years ago
I wanted to hate it so bad, but I think it&#x27;s a harmless idea that maybe gives the author a 5% advantage in future job searches. Why not?
ascariover 2 years ago
While it’s a nice concept I agree it’s probably not the best fit for the use case. I also don’t understand the point of recruitment spam. What’s the purpose of sending 5 follow up emails. I find it absolutely annoying and would never respond back to an unsolicited email. Surely had they enough applicants for an appealing job, they wouldn’t have chased that much.
评论 #32689267 未加载
评论 #32689185 未加载
a1371over 2 years ago
I am curious, what is the success rate with this? Do you get a valid JSON every time? Does it contain the right info every time?
评论 #32686824 未加载
wodenokotoover 2 years ago
Kinda surprised author uses GPT-3 to generate JSON and then a template to go from JSON to the reply.<p>Why not have GPT-3 generate the reply?
评论 #32688929 未加载
ivanvanderbylover 2 years ago
Bonus points if someone can apply this to answering and filtering spam calls, like an automated receptionist.
ACV001over 2 years ago
this is like using a CPU to prop a window open - it does the job, but it is not the right tool
djohnstonover 2 years ago
I&#x27;ve only recently started playing with these models, it didn&#x27;t occur to me you could ask it to parse content and return a structured representation. That&#x27;s pretty dang cool.
评论 #32719673 未加载
napoleonIVover 2 years ago
I have a few Alfred snippets to respond to recruiter emails. I just copy the first name of the recruiter and that use the appropriate snippet.
agf96over 2 years ago
Extracting NAME and ORG could not be done with a lightweight Named Entity Recognition model from HuggingFace?
EGregover 2 years ago
Can I generate a captain picard conversation vs Nigerian scammers using GPT-3?
personjerryover 2 years ago
I&#x27;m waiting for &quot;Responding to HN comments with GPT-3&quot;
orastorover 2 years ago
In your example you end up with &quot;Hi ,&quot; instead of &quot;Hi,&quot; if there is no name matched
评论 #32688918 未加载
jenaimarreover 2 years ago
Thanks for this. Moving forward I will reply with this (without trying to guess the values for the variables):<p><pre><code> Hi &lt;recruiter name&gt;, Thanks for reaching out! I’m not interested at this time, but I’ll keep &lt;your company&gt; in mind. </code></pre> It&#x27;s low cost and effective.
评论 #32687462 未加载
评论 #32688445 未加载
评论 #32687662 未加载