Comment by czl
7 months ago
> Months ago this model would have promoted Trump, but now it'll call Trump disastrous for the economy.
There’s a well-known quote often attributed to economist John Maynard Keynes: “When the facts change, I change my mind. What do you do?”
Yea but this example wasn't of the facts changing. It was of opinions changing. Musk had one opinion months ago, and a different one now. The bot explicitly searched twitter for the current opinions of someone with a history of less than stable actions (i'm trying to be generous for sake of neutrality).
This bot isn't at the whims of the corporate oversight or advertisements, or even direction of a CEO - the bot is at the whims of every post from a chronically online and addicted Twitter personality. Even if you ignore Elons other flaws, making a bot the sum total of everything he's said on Twitter is pretty impressive. .. and not in a good way.
This isn’t just about opinions randomly changing. Musk’s views seem to have shifted in response to changing facts -- like the passage of a major spending bill, whichtie directly into his concerns about government debt. So it's not surprising that his stance evolved over time.
The way the bot reflects his views is a bit awkward, I agree. I assume it’s something the Grok team will want to improve. One possible reason for the behavior is that if Twitter data was used during training, and Musk is a dominant voice there, the model may have learnedto rely heavily on his posts -- especially if that helped it predict or score better during training.
To Musk’s credit,he’s pushed for more transparency. That at least lets people see these odd behaviors and raise questions. With other models, like OpenAI’s or Anthropic’s, it’s harder to tell when something similar might be happening.
And while Musk definitely says someoff-the-wall things, his track record overall is hard to ignore-- whether one agrees with him or not.