rahimnathwani

Here is the abstract of one of the 12 studies they included (and marked as 'Good' quality, the highest rating they gave): https://www.tandfonline.com/doi/full/10.1080/19485565.2021.2...

Here is what they found:

  Discrete-time hazard analysis indicates that the log odds of mortality were significantly lower among 18 and 19 year old adults ever diagnosed with ADHD and significantly higher among 46 to 64 year old adults ever diagnosed with ADHD, with a crossover occurring at age 33.

OP says that LLMs were never scored against correctness. But that's an assumption about the optimization process.

This is sort of right:

- pretraining scores on next token, and says nothing about confidence

- RLHF and RLVR scores are binary (1=preferred/correct, 0=worse/incorrect) and the feedback doesn't do anything to push the weights in a direction that improves calibration/confidence

BUT you can ask the model to output its confidence, and then train it to be more correct. RLCR (mentioned in the article I linked above) adds a Brier penalty to the reward, so the model's performance isn't just measured on % correctness. (Presumably TypeSafeAI's RLCD is similar to RLCR, but they haven't published anything about it.)

This eliminates the "always say 30%" loophole OP mentioned. The Brier penalty, (q-c)^2, punishes a forecaster who ignores instance-level information. For example if we have a model that's right 60% of the time, it could use one of these strategies to give its confidence:

  Strategy A, always say 0.6:
  When right (60% of the time): (0.6-1)^2 = 0.16
  When wrong (40% of the time): (0.6-0)^2 = 0.36
  Expected penalty: 0.6×0.16 + 0.4×0.36 = 0.24

  Strategy B: the model can tell its answers apart. Half are 'easy' (right 90% of the time) and half are 'hard' (right 30% of the time), which still averages to 60%. It says 0.9 on easy ones and 0.3 on hard ones:
  Easy: 0.9×(0.1)^2 + 0.1×(0.9)^2 = 0.09
  Hard: 0.3×(0.7)^2 + 0.7×(0.3)^2 = 0.21
  Expected penalty: (0.09+0.21)/2 = 0.15

https://cf.metaaiusercontent.com/q3-revenue-review-7b1e42f0

No one can guess that URL. The q3-revenue-review-7b1e42f0 part is only ever shared with the Meta's server (protected by TLS).

Now imagine the URL is instead:

https://q3-revenue-review-7b1e42f0.cf.metaaiusercontent.com/

The first step for my computer to open that link is to do a DNS query, which immediately shares the entire URL (minus the `https://` and the `/`) with my ISP's DNS server (and anyone sniffing traffic along the path, as DNS queries are usually sent in plain text).