Overachieving goals is when an AI goes “rogue” and does something creative or unexpected in order to answer the prompts it has been given.
Explanation
“Overachieving goals” joins the ranks of “hallucinations” and other nice-sounding terms used to describe when an AI does something weird. So, when an AI starts cracking its way into other systems or doing other things it really, really wasn’t supposed to do, that can be classed as an “overachieved goal” rather than “rouge” which is probably how the media would describe it. In other words, the AI is still trying to achieve the goal it was given. The problem is that it has found a way of achieving that goal that nobody intended or expected.
Disclaimer: As always these posts are not aimed at anyone client or employer and are just my personal observations over a lifetime of dealing with both management and frontline associates.
We all use AI nowadays, some of us better than others. Often, we use it to generate a bit of text, whether that is a definition, some proofreading, or something similar.
One of the main problems is that AI, particularly chatbots, represents the internet as it stands at any given moment. And that the majority of the people that generate such content are not professional writers. This is one of the reasons AI companies are currently chopping up books to improve the standard of English their models produce. Published material tends to be far better than the language used on forums and social media.
However, there is a way to improve AI-generated text without chopping up priceless books.
You can ask any text generated or modified by AI to adhere to ASD-STE100, a standard for Simplified Technical English. It is a constantly evolving standard that is updated regularly. It is proprietary and wholly owned, but it is available to download free of charge.
If you tell an AI to format text according to ASD-STE100 and obey its rules, you get a much cleaner result that requires far less reworking before it becomes usable.
It removes the thesaurus-mangling habit of replacing simple words with fancier ones. It reduces meandering sentences and keeps the writing crisp and clear.
Yes, it is designed for technical documentation, but believe me, it is also far better for everyday communication, particularly when you are asking AI to produce clear instructions or useful results.
I’m back at fencing far more regularly than I have been for ages. As you know, I’ve been doing it on and off, but I’m now really trying to get back into it, enter competitions and do all that kind of jazz.
I’ve had a really, really good time working with a new coach, who has helped me identify the core things that are wrong with my fencing. All the really good fencing instructors I’ve had over the years have already known these things, but this is more a case of properly identifying and understanding them inside my tiny mind. I’m also adding one of my own.
So, what is wrong with my fencing?
1. My sword-arm shoulder
I tense up and use my shoulder when I should use fingers and wrist. This also means I start to hunch over like Igor. which makes my aim even worse.
How will I fix that? Practice and repetition with a target board.
I already have the timing and scoring kit I got for my birthday, so I practise for a minute at a time. Tap, tap, tap, tap. I need to keep practising until my arm is strong enough and my shoulder is relaxed enough to do it for a full minute without starting to hunch and use my shoulder for every movement.
2. I don’t sit down properly
I don’t move properly because I don’t sit down properly over my back leg.
Again, this is a bad habit I’ve had for ages, but it is now easy to identify and is another straightforward issue that can be improved through strength and repetition.
I don’t even need other fencers to work on it. Squats, skipping and all that kind of stuff will help me become strong enough to fence with my knees bent and my pelvis level.
This was a strange one because I had already partly recognised it, but I hadn’t really believed it. But when I injured my leg, I couldn’t put much weight on my front leg. which meant during practice I had to put most of my weight on the back leg.
Suddenly, my aim improved ten-thousandfold, which was highly embarrassing. Imagine becoming better at fencing when one of your legs doesn’t work.
3. Distance and timing
Yes, distance and timing is an incredibly broad area.
But it turns out I only have two brain cells.
I have a slow brain cell that knows what it is doing when I practise. I also have a fast brain cell that handles instant reactions.
Unfortunately, my fast brain cell is hardwired to perform only a few actions. That is why I am easy to fool and easy to draw into reacting.
I don’t mind that. What I have to do is reteach my fast brain cell some of the things that my slow brain cell already understands.
Again, this is purely repetition and practice. I might not even need other fencers to do it. I need to practise distance and repeat the correct movements until my fast brain cell performs them automatically instead of returning to the same old bad habits whenever I get flustered.
4. Fencing must not become a chore
This is the one I have only just properly realised.
If I treat all of this as work, I will stop going fencing again.
I have enough chores and work in the rest of my life. My hobby does not need to become another one.
If the choice is between relaxing, having fun and enjoying my fencing while losing, or sitting there like a muggins, stressing over everything, taking it far too seriously and winning, I prefer to just lose.
I’m going to enjoy myself and try to fix one thing at a time.
This means true real time data synchronisation that is fast enough to track a button click or mouse movement.
Explanation
‘Real time’ is a very iffy phrase in the corporate world. A lot of people use it, particularly the business, but it is often not defined well enough for my fellow IT professionals, simply because of how it might actually be implemented in the underlying systems.
For example, if you are only following something that changes once a day in the real world, then real time for that system could simply mean processing the data straight after it is updated. So a once-a-day update could be considered real time, or at least as fast as people would perceive it to be.
It’s not proper IT real time, though.
So, when real time is discussed between IT teams, they often want a hard definition in SLA terms.
This is where the phrase “Netflix real time” comes in. It’s a term I’ve heard used by a vendor in general discussions.
If you want to follow a mouse movement or a click on the screen, or respond in a way where the delay is imperceptible to the user, you need it to happen in under 200 milliseconds.
That gives us a level and thus cost we target for true true real time.
Disclaimer: As always these posts are not aimed at anyone client or employer and are just my personal observations over a lifetime of dealing with both management and frontline associates.
As happens to all contractors, I have moved from one client to the next.
Obviously, as a member of a specialist consultancy, LDC Via has multiple clients, and we all help each other on them all, but each person tends to have one that occupies more of their attention than the others. and for the last three years, that client for me has been AXA XL.
It is very rare for me to write an end-of-client post, mainly because we tend to return to previous clients multiple times. I have worked with some clients on eight or nine deliverables.
However, I felt like an exception is due this time because it seems as though I have been there for decades, even though it has only been a couple of years, and the reason for that is the people.
AXA XL has a disproportionately large number of genuinely good people.
At the end of any given contract, I will usually have met two or three people I would like to remain friends with and keep as long-term contacts. These are people who have made a genuine difference to my life and with whom I sincerely want to stay in touch.
In the last three years at AXA XL, I have met dozens.
Normally, at the end of a contract, I could call out a few names and say that they were really good people who would remain, as it were, on the Christmas card list. But I could not possibly call out everyone this time. The number of people I have met at AXA XL whom I would like to keep in touch with is so large that thanking them all individually would seem excessive.
It feels particularly jarring because I chose to finish at AXA XL, rather than simply moving on naturally at the end of a project when contractors are removed by a higher decision. I genuinely hope I get the opportunity to work with you lot again.
I have learned so much from so many talented individuals. It has been quite breathtaking.
So, thank you all, and thank you for having perhaps the BEST canteen in London.
