If Sutskever's opinion were not relevant enough, it has been seconded by other famous AI researchers, like Brockman, arguably making a point through example, and not just words
I guess someone has made the point of hiring twice as many people, which is not a great point for the mythical man-month, people × 2 does not work, you need 2x people
But who has that stamina? Chollet points at sustainability
I share the same experience, also pulling one-nighters, but it is a type of credit from the future, the arguable productivity gains are paid with interests later
This is hardly new it is becoming a tradition in AI and tech in general, with job ads
Why the focus on putting more hours?: Diminishing returns
The way to keep the productivity in AI is putting more hours, doing more of the same: graduate descent, putting more computing on the hyperparameter slot machine,…
And that is not bad, it keeps the wheels spinning, it squeezes some more results, and helps some people and companies to get great success, money, fame
Props to them! Keep it up!
Eventually, in software that backfires, and in AI it triggers winters
Honestly, a winter may be even be necessary, to cool down, relax, stop all the buzz and noise around deep learning and find a paradigm shift that takes AI higher peaks
Not enough has been written about how the Dunning-Kruger effect is a self-fulfilled prophecy by the Thomas theorem:
1. Redefining mediocre as excellent
2. Creating a kakonomy (lemons ≻ peaches), with positive feedback loop & Matthew effect
3. Resulting in a race to the bottom
The Matthew effect, e.g. network effect, impedes beating the mediocre, e.g. blub paradox and market dynamics
The peaches result not indistinguishable from the lemons but actually worse, possibly even "harbingers of failure", even at an axiological or first principles level
Differently from other multipolar traps, the main causes are epistemic and the strength of numbers; not a moral hazard. If there is a moral decision, it is beyond the comprehension of the mediocre majority
The few peaches, irrelevant as they may be, face a complicated decision:
Rather, consider you are increasing the probability of: 1. recurrence into a version of the virus that may kill you (even before you are old), and 2. saturating the healthcare system, that you may need, for example after a car accident
For every system, what allows it to operate in a wider range of situations in the future and for longer time, contributes to its health, it is beneficial (game & value theory)
Easy to see in systems like: body, mind, economy, ecosystems,…
This fits with the usual definition of adaptability to the environment, and the less common ability of pursuing an objective function in a wide range of situations, but there's more to it
The objective function is more specific: increasing health
The adaptation is not necessary
Intelligence cannot be understood without upper and lower holistic levels
Parasites have by definition a detrimental effect on the health of their hosts, less health means less intelligence for their strategy for survival, e.g. if they cause the death of their host, they may die
If you make questions for a job interview without due attention, you may ask something different from what you think you are asking, and you will effectively filter out people that know better than you.
Please do, otherwise meetings will be very frustrating later on.
If you want to do new things, you have to be your own boss
Work 9 to 5 on boring repetitive tasks to pay the bills, then be your own boss to grow
HR depts will rarely risk putting you on something new, they will tell you to do what you have already done