> In a snake-eating-its-own-tail irony, a 2023 analysis found that between 33% and 46% of workers on the platform were using large language models to complete their tasks,
I assume AI use by workers has risen to the point where it renders Mechanical Turk pointless.
This likely means those consuming the outputs of Mechanical Turk don't have a good way to measure the value (aka quality) of the outputs.
If they did - then they shouldn't care whether it's a human or a LLM. And if it's a LLM - then the cost will roughly correlate to the MIN(cost of the LLM, cost of a human) to do the task.
I think the "state of the art" of measuring the quality of outputs was to send the same task to multiple "agents" and only accept answers if over a certain amount agree. With some human review and reputation scoring sprinkled on top. It was a while since I was in this field though
This approach does work when there's a clear answer but what about tasks where the correct answer is multi-modal? Incentivizing agreement works only for tasks where there's clear answers.
The problem is bigger. Outside of coding, there is no real way to reinforce a model with pass/fail cycles until it stops hallucinating. This is why customer service uses will always have a problem. This compounds as you chain agents together.
It's like the speed of light - to get to that point, you need exponentially more energy, and you will never ever get there.
> then they shouldn't care whether it's a human or a LLM.
I imagine that the whole point of posting a task to Mechanical Turk nowadays is that you want it to be completed by humans. Either because you are after the small discrepancy between AI and human performance, or because humans are the object of your investigation.
Yeah, I was doing this kind of Artificial Artificial Artificial Intelligence back in 2012 to make some extra $$$. Glad they finally "patched" that hole ^^.
That's just Artificial Artificial Intelligence, the triple negative implies they built an automated system to impersonate humans who impersonate an automated system (which ultimately imitates a human).
Artificial Artificial Artificial intelligence is when the chat bot is out of capacity, so a person in India is writing the response that gets returned by the LLM which gets pasted into Mechanical Turk.
Are they? If one uses artificial in the sense of "fake" then a human pretending to be a machine AI would count as an artificial AI. The only scenario where this doesn't hold is if you are using "artificial" in the sense of "not created by nature".
If you get someone to mow your lawn, do you not care if (a) they use an automated machine that you could rent by the hour for far less, and/or (b) at the end your lawn is actually in a good shape?
I have very low expectations for $2 per hour. If I would feel the need to use such service I would expect them to cut all corners. If you have a subscription for a decent LLM, know how to get my data into it and know how to write a prompt that does the job you might have 3 skills I don't or I might not have time to do it.
If the lawn looks great you get to do it again. I could ask the guy you send how much you are paying him and I could ask you how much the mower costs but it wouldn't necessarily be a good use of my time.
Back around 10 years ago, I gained a new manager who had previously managed mechanical turk. It was already recognised as a dead end back then.
I remember him talking about getting a mandate from Amazon Security to upgrade from the long EOL MySQL 4.0 to MySQL 5.something, and that it was almost impossible to get any resources committed from leadership to even do it despite the fact it was security requiring it (which usually resulted in everyone jumping before stopping to ask how high to jump). I want to say he ended up doing it himself? Something like that..
All existing extremely minimal headcount was tied up in a massive technical debt of KTLO work, and proposals to resolve those issues similarly met resourcing road-blocks.
Maybe the most unambiguous "ai will automate work" example I've seen yet.
Absolutely does not imply the workers are automated since they can now use the current models to do more complex tasks at the vast number of new AI training data startups.
Turk was simply not designed for greater complexity tasks and so much of their lunch has been eaten by startups specifically built to collect AI training data.
I turked for a bit trying to make some extra cash leading up to my wedding, but it was a very time-inefficient way to make money. I think I managed to wring 10 or 20 bucks out of it tops after plugging at it for a month.
I can see a high value startup, that will provide Human Intelligence with real Humans, locked in the room, with no network, books, LLMs and monitored 24x7 with cameras.
This has very little to do with “AI replacing jobs” and much much more to do with a bad product getting obsoleted by better ones.
Human labeling is a two sided marketplace and so as any marketplace startup knows, both sides need to be constantly nurtured otherwise the system can collapse as worsening quality leads to churn and a vicious cycle that empties out the platform.
In labeling, you need to understand the limitations of individual work and fatigue, keep your pipeline bursting with awesome and consistent work, and improve the platform to make customer experience great.
AMT has been totally languishing in all these respects. Pay is terrible, dishonesty rampant, etc. It was a bad product, no need to pedestalize it or turn it political
I assume AI use by workers has risen to the point where it renders Mechanical Turk pointless.
reply