Dr. Richard Moulange
speaker
1,784 appearances
1 recordings
1 series
first heard Mar 2026
last heard 31 Mar
Dr. Richard Moulange’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Mar 2026 with 1.
Appearances
Though especially in biology, I'm not sure this risk from open weight versus closed weight is as useful when we think about those narrow tools.
For LLMs, there is this huge difference.
It's gone on for a long time.
It's discussed in the International AI Safety Report.
But at least for those narrow biological tools,
Most of them are open weights.
In the CLTR RAND Global Risk Index, we found that more than half of the highest risk narrow tools that we assessed were fully open in the sense they had open data, open code, and open weights.
So there isn't really this question of this dichotomy.
There are very few closed weight biological tools.
We haven't even got to a place yet that we should be talking about, oh, which is worse, because there's mostly just one type of
Yeah.
Yes.
So when we're talking about LLMs, generally open, useful for less well-resourced, less sophisticated, closed, more useful.
But you're right.
You've hit on an important point.
The more complicated and multi-step the sort of activity that even a highly sophisticated actor might want to carry out, the harder it is to use closed models.
Because there's a chance that at some point you get refusal.
Even if you can jailbreak it once, it might be that with enough context, wow, sure seems like the user is talking about bioweapons a lot, even though they were talking about fairy tales a few moments ago.
But I think this might be bad.
Oh, no, I'm not going to reply.
Showing 781–800 of 1,784 · page 40 of 90
← Previous
Next →