Rahia Mashoodh
@rmash.bsky.social
UCL Research Fellow in epigenetics, inheritance, and behaviour. Lover of R, tinkering and food. Sometimes I hurt myself bicycling. 🪲🇨🇦🚲
Reposted by Rahia Mashoodh
And that's why I think this is an example of endemic problems in our field. So much of the field thinks of stats as an afterthought. A technicality. The thing you use to tell your story. But if you use faulty heuristics to evaluate the quality of your model, your story could be complete bullshit.
October 30, 2025 at 6:59 PM
And that's why I think this is an example of endemic problems in our field. So much of the field thinks of stats as an afterthought. A technicality. The thing you use to tell your story. But if you use faulty heuristics to evaluate the quality of your model, your story could be complete bullshit.
Reposted by Rahia Mashoodh
Mine are mostly cons (if we're talking generic LLMs like ChatGPT and not bespoke ML models developed&trained in-house)
1. I don't trust them. I can't trust them bec they do not/cannot have a sense of truth or care for the integrity of the work. They bullshit randomly and don't know the difference.
1. I don't trust them. I can't trust them bec they do not/cannot have a sense of truth or care for the integrity of the work. They bullshit randomly and don't know the difference.
September 4, 2025 at 2:41 PM
Mine are mostly cons (if we're talking generic LLMs like ChatGPT and not bespoke ML models developed&trained in-house)
1. I don't trust them. I can't trust them bec they do not/cannot have a sense of truth or care for the integrity of the work. They bullshit randomly and don't know the difference.
1. I don't trust them. I can't trust them bec they do not/cannot have a sense of truth or care for the integrity of the work. They bullshit randomly and don't know the difference.