The Worst Data Science Advice I've Ever Seen

แชร์
ฝัง
  • เผยแพร่เมื่อ 28 ม.ค. 2025

ความคิดเห็น • 42

  • @ElizabethMacias5
    @ElizabethMacias5 3 หลายเดือนก่อน +7

    Very sensible advice. I am a beginning data scientist (5 months in) and am experiencing most of your observations, especially the one about driving business value.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      Glad it was helpful!

  • @kyleebrahim8061
    @kyleebrahim8061 3 หลายเดือนก่อน +6

    Very true on part 2, I realised this with multiple linear regression, in order to understand the how predictions are made there needs to be base knowledge on vectors and matrices, which ties back to algebra

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน +1

      completely agree!

  • @mikkiwittenberg7883
    @mikkiwittenberg7883 3 หลายเดือนก่อน +2

    Love the last advice point on focusing on impact!!! Thanks!!

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      glad you liked it!

  • @chessfighter-r5g
    @chessfighter-r5g 3 หลายเดือนก่อน +1

    Rbf kernel also very good , exponensional euclidean distance + dropping duplicated to not allow interferring long sides + gradyan boosting good tandem

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      sounds good :)

  • @stephenbernardo7288
    @stephenbernardo7288 3 หลายเดือนก่อน +2

    Finally! That's what I thought of. The funny thing is: these people who claim AI will take over jobs do not understand AI. Number 2 sounds about right. I knew it! I now know what to look for when I take another masters is related to Data Sciecne (not necessarily an MS in Data Science).

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน +1

      glad you agree!

  • @mytechnotalent
    @mytechnotalent 3 หลายเดือนก่อน +4

    Finally common sense speak. Thank you Egor!

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      glad you liked it!

  • @isalutfi
    @isalutfi 3 หลายเดือนก่อน +2

    Thank you for this video, Egor

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน +1

      thank you!

  • @chessfighter-r5g
    @chessfighter-r5g 3 หลายเดือนก่อน +1

    Is job market narrowing or future extend more ?

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      i feel its picking up slightly in the UK!

  • @kshitijk14
    @kshitijk14 3 หลายเดือนก่อน

    Very Practical & Helpful

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      Glad it was helpful!

  • @kmdsummon
    @kmdsummon 3 หลายเดือนก่อน

    The problem with data science compared to any other job, is the job is is not strict and tolerates a lot of mistakes. You may make a mistake or bias in your “analysis”, which I see pretty much every time DS does and nothing will change: there would be a mistake decision, experiment conducted, verified it’s not working and shut down. Manager will not recheck anything, since it’s not his job. If you compare this type of job with lawer or doctor - you just plain wrong.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน +1

      this is true to a certain extent, but every bit of analysis or modelling I do does get reviewed.

  • @cheahanqi964
    @cheahanqi964 3 หลายเดือนก่อน +1

    wonderful sharing

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      Thank you! Cheers!

  • @andreahoehmann1939
    @andreahoehmann1939 3 หลายเดือนก่อน

    Minute 2:17: I'm supposed to be able to explain how the system works under the hood? I've never seen any employer, any manager or stakeholder think about that! Everyone thinks the software will do its job correctly and without fail. I studied mathematics, but that wasn't a hiring criterion.

    • @fernandofuentes7617
      @fernandofuentes7617 3 หลายเดือนก่อน

      for some is good to know that. for example, the behaviour of algorithms in high dimensional data, o mixed numerical and categorical data, but in general of course you are not arithmeting matrices all day long

    • @andreahoehmann1939
      @andreahoehmann1939 3 หลายเดือนก่อน

      @@fernandofuentes7617 You are 100 percent right. But it has always been easier to explain to customers or employers: "The infallible software produced the result!" than to explain the algorithm to them.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      its more to ensure they trust you, if you don't know how it works and can't explain certain decisions then it doesnt look good

    • @andreahoehmann1939
      @andreahoehmann1939 3 หลายเดือนก่อน

      @@egorhowell I agree with you absolutely, but in 20 years I have never seen a manager who was interested in how a software works under the hood. I would have been happy to meet one. I completed two subjects at university, maths and data science, and for a long time I believed that it made a difference to applicants who had only taken a short IT course. I no longer believe that.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      Fair point. I have been asked by stakeholders a few times to explain a decision, so it has been useful for me!

  • @mehek6440
    @mehek6440 3 หลายเดือนก่อน +2

    Second like is mine🎉

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน +1

      Thank you!

  • @DownToEarthMind
    @DownToEarthMind 3 หลายเดือนก่อน +3

    your assessment is only true to some degree. top data scientist will keep their jobs (complex domain know-how, AI model creation, etc.) but many low to intermediate task will be taken over by AI. So you either have to be very good or face AI competition in the job market.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      Thanks for sharing

    • @ozan4702
      @ozan4702 3 หลายเดือนก่อน

      I dont think this will happen anytime soon.

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      I agree. But time scales are hard to judge.

  • @andydataguy
    @andydataguy 3 หลายเดือนก่อน +1

    Data science and ML engineering are the first jobs that I'm replacing in my startup. Not sure what you're talking about 3 years bro.
    RAPTOR + REMO + ACE + STORM + LangGraph + l33tSkillz = AGI with just GPT-o1-mini fine tunes and Sonnet3.5
    O1-preview wasnt even necessary, but it made it to where now it can optimize our DRL agents policy classes and market regimes even better than before.
    The improvements will increase with each new model.
    Most people won't have this capability anytime soon. But some people have it right now.
    The idea it'll take 3 years to propagate with exponentially better technology is just a willlllddddd take

    • @andydataguy
      @andydataguy 3 หลายเดือนก่อน

      Nothing against data science... I literally call myself andydataguy lol
      But creating a startup is hard and capital is more limited than ever
      Technical founders are totally going to choose the $15k option over the $185k per year option anyway of the week.
      The agents get exponentially better and exponentially cheaper.
      The issues with ai today can almost entirely be solved with better automation. The rest is just knowledge management and devops.
      ...and being crazy enough to not fall for the lie that AI is somehow not currently a threat to all white collar jobs
      People need to be preparing man. It's genuinely concerning more people don't take this serious.
      And unfortunately the ones who are preparing will be the new wealthy class that everyone will hate even if we're trying to warn people...

    • @daffydooms7953
      @daffydooms7953 3 หลายเดือนก่อน

      web3 end of comment.

    • @ramg4699
      @ramg4699 3 หลายเดือนก่อน +1

      Yeah. sure buddy 😁

    • @egorhowell
      @egorhowell  3 หลายเดือนก่อน

      thanks for you opinion :)