It's fun to think about, but if you want to really reign in the C-suite, you need to start setting up systems within companies that allow for the rank-and-file to replace bad CEOs.<p>We've spent the last 80 years (ostensibly) talking about how great representative government is, then we let people own a majority of voting shares so that they can run companies like dictatorships. I've worked at a company that could have remained more competitive (and done a better job of maintaining software that is critical to human life) if there had been more uptake of ideas from workers instead of simply trying to make shareholders happy.
My dad had a manager recently that he started calling "VPGPT" behind his back.<p>He felt like the guy primarily spoke in truisms and with long-winded statements that ultimately said very little but gave the illusion of having depth. He wasn't a huge fan.
Unfortunately our CEO at a 5000 person company that heavily uses Slack, still hasn't activated their Slack account, so I don't have much to finetune the bot on.
I'm saddened to tell you fellow hackers here that the real world will be more like:
"Replace your knowledge worker with an LLM" ordered by your CEO.
In my past I was the technical co-founder for a successful startup and eventual exit. I had a background in marketing automation and analytics.<p>At different points I acted as CTO, CEO and COO and ended up at a CTO but with a cutout that I ran the "renevue" team.<p>Except for glad handing VCs in the early stages the first thing that should be automated away is most roles in the C-Suite.<p>You'll still need a flesh and blood figurehead for external meetings as well.<p>Many software engineers seem to be extremely blind to practical concerns around positioning, market fit, pricing, etc. and may bring certain idealogical biases and "purities" into their decision making that sabotage their success, just put yourself in the hands of AI and trust it.<p>The simple fact is most C-suite teams don't perform better than the macro environment they exist in, so getting "decent" AI decisions in place and then energetically executing you'll beat most teams, and in early stage startups.<p>If some people say we can replace the bottom 20% of devs right now, we can probably safely replace the bottom 40% of c-suite roles.
> All the components—scraping, fine-tuning, inference and slack event handlers run on Modal, and the code itself is open-source and available here.<p>Does anyone know how much it would cost to put this together? How much does it cost to fine tune Lama 3.1 on someone's Slack messages?<p>I'm guessing it depends on the number of messages, but roughly for 10k-100k messages, at 200 characters each (guessing about 50 tokens each, so 500k-5M tokens)?
I'm mostly curious about the decision to do fine-tuning here instead of synthesizing a good system message, maybe a RAG setup with access to the slack database, etc. I tried fine-tuning on a content generation case when GPT 4 first came out and we had much better results without it. Also, you don't get the benefit of upgrading to new models when they are released.
I think they're burying the lede here, they released a DoppleBot you can train on Linus's messages from the lkml as the ultimate arbitrator of coding disputes.
Yet again HN gets weird with highly contradictory popular posts trending simultaneously<p><a href="https://news.ycombinator.com/item?id=42923870">https://news.ycombinator.com/item?id=42923870</a> "A computer can never be held accountable"