TC premium for engineers with ML background

Google DJ Okawari
May 27 22 Comments

How much higher (or lower) is comp for ML engineers relative to general SWE with same YoE, after accounting for negotiation on competing offers and difference in career growth rate due to supply/demand for subject matter expertise?

58 VOTES SELECT ONLY ONE ANSWER
VOTE VIEW RESULT

comments

Want to comment? LOG IN or SIGN UP
TOP 22 Comments
  • Amazon hypercube
    Don’t do ML if you want to be SWE. Because SW work accompanying ML is tier 4.

    Only do ML if you are willing to be pure Research Scientist (not even so called Applied Scientist to avoid tier 4 SWE work).

    Let tier 4 engineers do the SWE work.
    May 27 6
    • Amazon hypercube
      Not necessarily the engineers (for Google and similar companies) but the most SWE work (not including the ML model design/tuning/paper publication) like training jobs, automation, data management, and so on.

      Developing and releasing Tensorflow or Opencv type software is exception of course. But those probably receive heavy input and guidance from the scientist for the interesting and challenging parts.
      May 27
    • Walmart.com / R&D
      serjorah

      Walmart.com R&D

      PRE
      Grab
      serjorahmore
      Tier 4😂
      May 27
    • New / Eng Gannicus⚔️
      God Silicon Valley is full of such arrogant people.
      May 27
    • Google DJ Okawari
      OP
      My main purpose in continuing with ML is

      1. for the sake of finding a more mathy line of work than building web infrastructure

      2. enjoyment of analyzing data (and that includes cleaning, statistical inference, reading documents on applied results, etc.)

      So it seems a reasonable plan to find a product team that needs someone with ML experience, without having to serve on a purely research-focused org. What do you think?
      May 27
    • Credit Karma Oaxg376
      This is not true. Tier 4 shit I mean
      May 27
  • Microsoft Vbsm66
    It really depends on your particular situation, ML experience is obviously not all equal. The difference will range from zero to 10x.
    May 27 12
    • Google DJ Okawari
      OP
      I can't poll an entire distribution using Blind's system. The question pertains to the majority audience.
      May 27
    • Microsoft Vbsm66
      Polls won't get you the answer that you're looking for, there's no magic button that HR pushes if you have ML keywords on your resume. However, I can provide what you're looking for. There are essentially two ML engineer archetypes. The first type, engineers with industry ML experience, are basically regular engineers with some relevant experience. They receive regular TC offers, but due to the market demand they may get better job offers from companies normally outside their competency (eg get Google offers instead of Walmart, +50% TC) and get more offers overall, thus being able to improve their average TC through negotiation (+10-25% TC). The second type, engineers with perceived domain expertise, are generally PhDs with top publications and are in the highest demand. They receive top offers out of the gate pre-negotiation (roughly equivalent to the highest in-band offers, +50% TC), and their post-negotiation offers can go through the stratosphere (+200-500+% TC). Obviously, non-ML engineers can get top offers and negotiate as well, so your "median" answer will depend on the statistical distribution of people in each bucket. That is likely not useful for your purposes though.
      May 27
    • Google DJ Okawari
      OP
      I found the distribution in the poll results interesting because of its skew and mode. The tail drops off quickly, although it may be long as there are ≈15% in the top bucket (50%+); that type of aggregate estimate is useful information. Moreover, I am rather skeptical of the +200-500% figures in today's market; if you extrapolate the distribution from the poll, the proportion of those cases seems very low (close to 0). I feel more confident about statistics on 40 data points, rather than one or two cherrypicked anecdotes. And the indifference toward outliers is partly because I am unlikely to study ML for 40 hours per week if it is likely one would only find a meagre gain in TC; tbh, in the +10-20% cases, I'd rather distribute my time on some other topics for satisfaction, such as number theory and theoretical statistics. Or spend some extra time at work and get promoted faster.
      May 27
    • Microsoft Vbsm66
      The poll results fall exactly in line with my description above.
      May 27
    • Microsoft Vbsm66
      I don't think you should study ML for 40 hours per week if your goal is TC. It is unlikely that you'll become a true domain level expert with that level of studying (you need a PhD for that), and therefore the only benefit would be added leverage during negotiation.
      May 27
  • New / Eng Gannicus⚔️
    Don't try to make yourself feel superior ⚔️
    May 27 0
  • Nutanix WyOV12
    ;)
    May 27 0