If this is true, I am so disappointed in Wordpress (where I used to work). They were the leader in customer support, keeping it all in-house and requiring all employees to spend time doing support. If they really just hooked a LLM bot up to their email support, they're ruining their own reputation, and I've lost my last example of a company doing support right. https://social.panic.com/@cabel/111692900717018267
Oh you don't know half of it. Since LLMs are supposed to take over their tasks, now all the support folks in WordPress.com have a monthly sales goal instead. The fist support level is a bot. The second, someone who is being forced to try to sell you a more expensive plan is they want to keep their jobs.
@fraying this unfortunately tracks since leadership there has seemed VERY onboard with how much “AI” + WordPress is a fantastic idea and the future of everything, including a nod of support to the “Techno Optimist” fascist manifesto.
@sara I got in so many fights (well, let's call them "spirited conversations") with the data scientists there about what we could feasibly do with LLMs.
I'm not anti-LLM as a rule, but making it customer-facing is asking it to do something it is NOT capable of and by doing so you're demonstrating that you don't give a shit about your customers or your community.
@fraying Yeah, like, I could see ways you could have that built into the support service as a capitalist speed bump (if you're clear about it) the way some places have a "Does this answer your question? If not, hit this button to get help from a person" but not like this, not at all like this.
@jessamyn Just make it a backend assistant tool to the real humans who know when it's wrong BEFORE it goes out to the user. Use the tool as a TOOL, not a replacement. (Everything goes back to why we still need librarians, as usual.)
@fraying I often compare LLMs to search engines since Google has been approaching LLM-like responses for years and often search results don't quite return relevant results. Expecting it to as accurate as a calculator is not reasonable. With that in mind, exposing LLMs directly is like exposing massive search indices without filtering out garbage data is unhelpful.
I believe the fatal flaw which remains with LLMs is the inability to trace back and cite sources from training data. Once this is overcome and LLM responses actually generate traffic to sources I hope it is regarded as acceptable as search engines have been for years. When it comes to news sources or artwork, I’d like some kind of royalty system to be in place so there is a benefit to contributing to LLMs.
Add comment