Security Writer
I’ve had a week of Microsoft Copilot integration testing across this organisation. Most features being assessed from user productivity to Azure admin.
And let me tell you, I’m coining it now, we’re going to be left with years of ‘Accuracy Debt’ when we see more organisations adopt it.
My very favourite example of just how fundamental these problems are were with user email synopsis and meeting time suggestions.
Security Writer
On Wednesday, next Wednesday, a week on Wednesday, in two Wednesday’s time, three Wednesday’s on, or any other temporal variant, and including regional dialects are all totally alien concepts to it. And god forbid you discuss more than one timescale or event in one email.
What I do fear is that for those that do heavily invest, they’ll be required by the likes of Microsoft to begin to speak in ways that Copilot can parse and ‘understand’. Which in larger organisations will shape the way that they communicate with others, and it’ll spread. In the same way that American corporate speak, or Agile terminology has infected the rest of the world.
Our interactions, the things that make us most human, will be modified for the sake of corporate benefit over the coming years.
Security Writer
Language is a wonderful thing. It deeply saddens me that this isn’t some fringe theory, but an inevitability.
Context and nuance is going to be erased wholesale from captured information… which will also be filled to the brim with inaccuracies.
pettter
@SecurityWriter nothing social is inevitable
Andrew Clews
@SecurityWriter I still don’t understand why any company wants to nuke its productivity by staff having to take more time to check and fix whatever is spouted out.
Or worse, the company and employees DON’T check and those mistakes filter into their work/products/services.
Felichs
@CrackedWindscreen @SecurityWriter
Any company that forces me to use a tool that introduces mistakes to my daily work flows GREATLY overestimates how much time I'm going to spend correcting those mistakes.
David Chisnall (*Now with 50% more sarcasm!*)
What I do fear is that for those that do heavily invest, they’ll be required by the likes of Microsoft to begin to speak in ways that Copilot can parse and ‘understand’.
We have those things, they’re called programming languages and they have well-specified syntax (and, often, semantics). Unfortunately, the shape of what an LLM can parse depends on its architecture and training data, and fairly small changes to either can have a big impact. Copilot seems to be a roll in release thing, where the underlying models can be changed at will. Just because you have created a shape that Copilot can parse today does not mean that it will work tomorrow.
We are going to see many organisations rediscovering ontology drift. My favourite thing about ontology drift is that it’s recursive: everyone who rediscovers it comes up with incompatible terminology for it and cannot share their findings in a useful way.
Watts Martin
@SecurityWriter I remember documenting the date and time library for Bixby, Samsung’s unloved voice assistant, and handling natural language for date expressions is wildly difficult—sometimes the library would return up to three possible candidates and the application would have to figure it out from context. I can imagine handing it off to an LLM without sanity checking—or maybe even with sanity checking—would create a fascinatingly unholy mess. :)
eLearningTechie
@SecurityWriter I have recently had the same trial 'privilege' where I work. The vast majority of Teams meeting summaries Copilot spat back at me were such gobbledegook, I had to watch the recordings to make sense of them. In at least a couple of cases Copilot contradicted what had actually been said.
I s'pose it's no accident that FAIL contains an AI...
Haelwenn /элвэн/ :triskell:
Haelwenn /элвэн/ :triskell:
Security Writer
@lanodan @chipotle Yeah, and it’s just a small part of the thousands of micro-contexts we provide every day in even casual conversation.
I’ll be a while til this improves I think.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.