Post Aa8OpAyPzPcWsBoRHM by fuzzychef@m6n.io
(DIR) More posts by fuzzychef@m6n.io
(DIR) Post #Aa8Cx3DbVXHJghu7Wq by simon@fedi.simonwillison.net
2023-09-25T16:39:14Z
0 likes, 0 repeats
Here's a thought-provoking take:"We already know one major effect of AI on the skills distribution: AI acts as a skills leveler for a huge range of professional work. If you were in the bottom half of the skill distribution for writing, idea generation, analyses, or any of a number of other professional tasks, you will likely find that, with the help of AI, you have become quite good."Ethan Mollick - https://www.oneusefulthing.org/p/everyone-is-above-average
(DIR) Post #Aa8EmtfeLnS75ZJB8i by simon@fedi.simonwillison.net
2023-09-25T16:59:48Z
0 likes, 0 repeats
For me this is the single most positive potential future outcome of the current leap forward in terms of AI capabilities: a tool that helps human beings level up, as opposed to replacing themA big challenge here of course (as Ethan outlines further down in that post) is whether organizations will chose to do more with their existing teams, or will downsize them to try to achieve the same with less people
(DIR) Post #Aa8Fnv4CkHZ5f7qSLA by simon@fedi.simonwillison.net
2023-09-25T17:11:27Z
0 likes, 0 repeats
I'm particularly interested in translation (between human languages) because that's been quietly impacted by leaps forward automation for 5-10 years at this pointI'd love to see some credible studies on what the impact of that has been on the human translator industry: has there been mass unemployment, or are existing translators getting more work but it's more "clean up this AI-generated version", or is it a mixture of both?
(DIR) Post #Aa8G0waiOC4k9eGJYO by fuzzychef@m6n.io
2023-09-25T17:13:42Z
0 likes, 0 repeats
@simon This is exactly how mechanization has proceeded since the beginning of the industrial revolution. Owners will use new tech to decrease labor costs, and it's foolish to expect them to go against their financial interests. Only labor action can make increased productivity result in increased compensation.
(DIR) Post #Aa8GEB7LgxPQyJpxGS by fuzzychef@m6n.io
2023-09-25T17:15:04Z
0 likes, 0 repeats
@simon Disagree. AI does not allow mediocre staff to become good. It allows completely unskilled staff to become mediocre.
(DIR) Post #Aa8GSbx0itvuFQspuK by NIH_LLAMAS@mastodon.social
2023-09-25T17:15:53Z
0 likes, 0 repeats
@simon oh golly geezballs i cant imagine which direction that will go
(DIR) Post #Aa8Gf8OTvUzcLLhjSi by simon@fedi.simonwillison.net
2023-09-25T17:18:44Z
0 likes, 0 repeats
Just found this excellent piece about exactly that:-"How human translators are coping with competition from powerful AIProductivity is up and real wages are down, but humans are still in the game."https://www.slowboring.com/p/how-human-translators-are-coping
(DIR) Post #Aa8GqSHKKuNzCRkvlw by pettter@mastodon.acc.umu.se
2023-09-25T17:19:35Z
0 likes, 0 repeats
@simon A mixture of both, and translators overall hate it, is what I've heard from translator friends. Important to note that machine translation for anything but short snippets still is not actually very useful except in the heads of managers and AI hypists.
(DIR) Post #Aa8H3SkFueRAPZUPlQ by simon@fedi.simonwillison.net
2023-09-25T17:21:53Z
0 likes, 0 repeats
@fuzzychef That's a pretty definitive statement - what leads you to believe that's the case?The study Ethan ran is a bit weird in that the study participants all worked for Boston Consulting Group, so they presumably represented a pretty high functioning initial group
(DIR) Post #Aa8HE27zDPbp2qd7Bo by peter@thepit.social
2023-09-25T17:23:28Z
0 likes, 0 repeats
@simon i've been a translator for 10 years and i'm currently switching careers. work has slowly dried up, and wages are terrible. i would never recommend anyone become a linguist. good luck finding people to train your LLMs in 20yrs.
(DIR) Post #Aa8HPovW7eyxZrPyk4 by mia@hcommons.social
2023-09-25T17:23:39Z
0 likes, 0 repeats
@simon I'd also be interested in the impact on language learning, and the related loss of different ways of thinking that learning languages gives you. It must all have a chilling effect on people who've chosen not to learn creative skills that seem AI-replaceable
(DIR) Post #Aa8HtSe1X1639rMxea by thejacenallen@mastodon.social
2023-09-25T17:32:23Z
0 likes, 0 repeats
@simon lol
(DIR) Post #Aa8M4ojg8Y4gftgWvo by recker5@mastodon.social
2023-09-25T18:21:31Z
0 likes, 0 repeats
@simon >whether organizations will chose to do more with their existing teams, or will downsize them to try to achieve the same with less peopleOf course they will downsize, that's what capitalism demands. But that's not a bad thing. AI will enable us to work less and achieve the same productivity.All in all AI will make our lives much easier and more comfortable - in the same way the industrial revolution did, after it lost its horrific working conditions in the beginning.
(DIR) Post #Aa8MbGdXw8fGTWI2i0 by fuzzychef@m6n.io
2023-09-25T18:27:10Z
0 likes, 0 repeats
@simon This is based on my own experience managing people on writing tasks in a work environment, and my spouse's experience editing manuscripts generated by ChatGPT.Looking at the study, the qualifications of the human graders of output are not defined. If those graders were generalists, and not experts in the material they were grading, then they would not have been able to assess actual quality as opposed to merely polish ...
(DIR) Post #Aa8Mp9Ole5Qy50zti4 by simon@fedi.simonwillison.net
2023-09-25T18:29:30Z
0 likes, 0 repeats
@recker5 Capitalism demands growth and profit - if a new technology means someone can 10x their workforce to 100x their profitable output, the incentive is for them to hire more people
(DIR) Post #Aa8N2NsZ7OV0ktFH04 by fuzzychef@m6n.io
2023-09-25T18:29:34Z
0 likes, 0 repeats
@simon ... for example, my experience with talk proposals written by LLM is that they have better grammar and spelling than inexperienced writers would have on their own. But the actual content of the proposal isn't better, and is in many cases worse because of the LLM generating BS filler.A grader who had no real expertise in judging talk proposals might judge the LLM one as being better than it is, though.
(DIR) Post #Aa8N2RVJdmHJzwOlJg by simon@fedi.simonwillison.net
2023-09-25T18:30:59Z
0 likes, 0 repeats
@fuzzychef I think using LLMs effectively takes a substantial amount of skill - that's one of the many open questions for me: how quickly can we help people develop that skill to the point where it's a net boost to their productivity
(DIR) Post #Aa8N2SPKHX56nebUcS by fuzzychef@m6n.io
2023-09-25T18:32:19Z
0 likes, 0 repeats
@simon Theoretically, someone could create a special-purpose LLM that was actually a skills amplifier regardless of skill level. For example, I can completely imagine a correctly programmed and trained LLM being able to take a good outline and generate a good talk proposal from it.But not any of the LLMs available to the public, currently. Part of the problem is that he majority of the talk proposals published on the web are bad.
(DIR) Post #Aa8NGAGWWv5yaV1SuO by peter@thepit.social
2023-09-25T17:28:25Z
0 likes, 0 repeats
@simon i'm always amazed that tech people like yourself don't understand how this is also going to affect you. if it's easier for anyone to leverage these skills, and a ton of knowledge workers are retraining because they can't earn a living wage in their chosen field, it stands to reason the labor supply of CS/IT professionals is going to go up, which is going to push wages down, which is the whole reason the capitalists are excited about this technology (as cool as it might be, and it IS cool)
(DIR) Post #Aa8NGB6HQUUnB1Ena4 by simon@fedi.simonwillison.net
2023-09-25T18:33:21Z
0 likes, 0 repeats
@peter Don't assume I don't think about that a lot - I've spent a couple of decades building up my own experience at writing code, and now it turns out writing code is one of the things LLMs are best at - coding is a whole lot easier for them than writing in human languages, because the grammar rules of Python/JavaScript etc are trivial in comparison to Spanish or Chinese or English
(DIR) Post #Aa8NRmS6tOAuCvysvg by fuzzychef@m6n.io
2023-09-25T18:33:44Z
0 likes, 0 repeats
@simon Right, and is that more effort than just learning to do the thing in the first place? There is a level of expertise in many tasks where an LLM, no matter how good, could not improve on it, or even make it faster.
(DIR) Post #Aa8Nh5rmzdawlOIID2 by simon@fedi.simonwillison.net
2023-09-25T18:35:30Z
0 likes, 1 repeats
@fuzzychef My currenty intuition, based on my own experience and signals I've been picking up from various other places, is that LLMs really can provide a substantial productivity boost beyond just "learning to do the thing" for an increasingly wide range of tasksI was already a very productive programmer, but I've seen a 2-3x performance boost on the time I spend typing-code-into-a-computer (which is admittedly only ~10% of my work, but still material)
(DIR) Post #Aa8NhA9J4Rz42sWL4K by simon@fedi.simonwillison.net
2023-09-25T18:36:45Z
0 likes, 0 repeats
@fuzzychef But I'm also very much an LLM power user - I've been using these things on a daily basis for a solid year now, since a few months before ChatGPT even came out (I was a heavy user of GPT-3 via the Playground tool)The research I'm most interested in concerns how new users pick up and use these tools
(DIR) Post #Aa8OIu7n679tLnoAa0 by recker5@mastodon.social
2023-09-25T18:40:06Z
0 likes, 0 repeats
@simon Only if the market is infinite.If a, or better, all newspapers through AI can cut their editors down to 20% and produce the same product, they won't invest the saved money in 80% new editors.Many things will be produced with much less humans when IA really takes off. But it will be a good thing over all, not the worldwide poverty with an unusable workforce.In some hundred years there will be 5 or 10 hour work weeks. Thanx to AI and robots.
(DIR) Post #Aa8OTOWOwB3yWMK0BM by fuzzychef@m6n.io
2023-09-25T18:37:23Z
0 likes, 0 repeats
@simon There's also a productivity factor that the LLM advocates are studiously not examining. If the LLM is faster 90% of the time, but 10% of the time makes a critical error, it is NOT a net gain. Workers will spend far more time dealing with the 10% errors than they would have doing the work in the first place.When I did data analysis, we used to have a working rule that if more than 8% of data was untrustworthy, then you throw out the set and start over.
(DIR) Post #Aa8OTPIc2vcyvssVKS by simon@fedi.simonwillison.net
2023-09-25T18:40:15Z
0 likes, 0 repeats
@fuzzychef one of my main areas of investigation right now involves how LLMs can be applied in the field of data journalism - which is a fascinating challenge, because hallucinated mistakes are a DISASTER if you're a journalist trying to tell truthful stories about the world
(DIR) Post #Aa8Oddd5caieGzKoGe by simon@fedi.simonwillison.net
2023-09-25T18:41:14Z
0 likes, 0 repeats
@recker5 "only if the market is infinite" is a very astute observation, thanks for that
(DIR) Post #Aa8OpAyPzPcWsBoRHM by fuzzychef@m6n.io
2023-09-25T18:45:50Z
0 likes, 0 repeats
@simon yeah, that would be interesting. Like, the users I deal with are NOT power users, which is probably why I have the experience of LLMs doing more harm than good.
(DIR) Post #Aa9V1iFurmqiSDJnEW by janl@narrativ.es
2023-09-26T07:36:35Z
0 likes, 0 repeats
@simon a friend of mine works in the industry and was on a podcast the other week (in German): so far any form of automation is used to push wages and salaries down because a “translation reviewer” supposedly does “less work” than a proper translator, but it’s the same people, skill and time required behind either task.The automations the people doing the work ask for that would support them on their work is barely a thing.