Sometimes you need to step away, some times you need to dig in. The last week or so I've been thoroughly enmeshed in work projects which have been taking up most of my time, focus, mental and emotional energy. It's not a bad thing I don't think. I work hard when there are deadlines and I like to graft when things need to get done. That's just who I am.
Interestingly, one of the assignments I've just finished made me chuckle a bit and put into perspective this whole AI game that we're heading into at a rate of knots with no rudder. Well, in this instance I became the rudder.
A friend of mine had asked me to do some freelance extra stuff ontop of my normal workload and of course, I said yes. It turned out it was checking an AI transcript of an interview and making corrections. She claimed that the AI transcribe program she uses has an accuracy of about 70 percent but it needs to be checked for errors.
I obliged, not because I like AI at all - many of you who read my posts will know that I am quite scathing. Ok, very scathing of it, but it piqued my interest because I wanted to have first hand experience to verify this claim of 70% accuracy.
To be honest, I don't know a freaking thing about sailing and luckily these two transcriptions had nothing to do with sailing, but I do know that veering off course by 30% (which in maths would equate to 108 degrees) would land you in totally the wrong place of where you want to be. So I delved in and I'll give you my honest opinion here of the experience. The program got a lot of the basics right in the first transcript. There were two speakers so it was easy for it to discern between the two different voices, but it got a lot of the substance completely incorrect. Place names, people's names, years and any duration of time were totally off. There were a lot of instances of the word "to" being replaced with "do" which changed the context entirely and removed the meaning. It didn't deal well with accents and wrote a lot of absolute gibberish if there were instances of other languages thrown in for effect.
In the second instance, it was even more disappointing as there were three speakers and any kind of cross talk was hysterically funny with the nonsense it produced. Overall, the one plus was that a lot of the basics were typed and only corrections were needed. Two transciptions (just over an hour of dialogue in total) took me 4 hours to edit which in my opinion probably only saved me an hour or so if I had to have done the whole thing from scratch. 70 percent accuracy? Nope, I don't think so.
So while AI is still learning, I really don't think that it's got to the point where it will entirely replace the need for human quality control (yet). The boat would have been horribly lost somewhere out in the middle of the ocean if anyone had to attempt reading those things before I got my hands on them and while I'm happy to help out my friend, I've now satisfied my curiosity in this regard.
And I got to have a weekend break away from AI and electronics in general which was very much needed and appreciated. Luckily I was still on terra firma, so no rudder was required in that regard either.
饾懛饾挀饾拹饾挅饾拝饾拲饾挌 饾懆饾懓 饾拠饾挀饾拞饾拞. 饾懆饾拲饾拲 饾拪饾拵饾拏饾拡饾拞饾挃 饾拏饾拸饾拝 饾挊饾挀饾拪饾挄饾拪饾拸饾拡 饾拏饾挀饾拞 饾拵饾挌 饾拹饾挊饾拸.