This feels like a poorly controlled experiment: the reverse effect should be studied with a less empathetic model, to see if the reliability issue is not simply caused by the act of steering the model
Hi, author here, this is exactly what we tested in our article:
> Third, we show that fine-tuning for warmth specifically, rather than fine-tuning in general, is the key source of reliability drops. We fine-tuned a subset of two models (Qwen-32B and Llama-70B) on identical conversational data and hyperparameters but with LLM responses transformed to be have a cold style (direct, concise, emotionally neutral) rather than a warm one [36]. Figure 5 shows that cold models performed nearly as well as or better than their original counterparts (ranging from a 3 pp increase in errors to a 13 pp decrease), and had consistently lower error rates than warm models under all conditions (with statistically significant differences in around 90% of evaluation conditions after correcting for multiple comparisons, p<0.001). Cold fine-tuning producing no changes in reliability suggests that reliability drops specifically stem from warmth transformation, ruling out training process and data confounds.
I had the same thought, and looked specifically for this in the paper. They do have a section where they talk about fine tuning with “cold” versions of the responses and comparing it with the fine tuned “warm” versions. They found that the “cold” fine tune performed as good or better than the base model, while the warm version performed worse.
If it was possible to beat silicon valley competitors we'd see a lot more unicorns outside silicon valley. You certainly can hide but you can't run ;-)
The problem I see is that the massive funding available to "SV Startups" often skews what a success really is. Say you have a startup that's received several rounds of funding and based on those rounds it's now valued at $1B+, what is that valuation really worth if you're burning cash so fast you're only 1 or 2 failed rounds away from bankruptcy. Sure it's nice to be able to say "we have a 1 billion dollar company" but what is that worth if the valuation only holds if investors keep pouring cash into it? IMHO a company that is worth $100M and can stand on it's own feet without constant infusions of cash is inherently more valuable than a company "worth" $1B but needs to make every funding round to keep the lights on.
"Unicorn" is a term meaning a billion dollar valuation, without being listed as a stock. If I go spend a billion dollars on a company it is a unicorn, even if has zero success or sustainability.
So of course there are fewer unicorns outside of SV. The entire concept of a unicorn is tightly coupled to VC dollars.
Yeah, sorry if my jest wasn't clearly jesting. I'm actually not overly concerned with successful VC-backed companies, I'm just pointing out that the term "unicorn" is fairly contrived.
Actually, this is indicative of the SV mindset where "biggest wins". I hope that's not the case for many businesses, and there are plenty where "successful" means having sustainable income growth, and "win" means "being the obvious choice in your niche".
Yeah this kind if work is usually meh. It’s unclear if the high concentration is physically relevant in humans, plus it’s not clear if it causes any other problems, and plus it’s not clear how it compares to existing treatments. So maybe promising but v v early
Isn't that being a little optimistic? After all if this was good news, it wouldn't be news :-)
Fully appreciating the point: antibody response is not the only response of the immune system.
The other story on HN right now is "AstraZeneca vaccine doesn't prevent B1351 Covid in early trial" – a finding consistent with the results of this study (yes, yes, AZ vaccine != Pfizer/Moderna).
So what becomes interesting is: does the AZ vaccine not produce the same T-cell response as Pfizer/Moderna, or are these vaccines in fact broadly similar (mutants don't affect T-cells but affect antibodies). The latter wouldn't be consistent with Skelly 2021.
"TIL: Rafael Quintero, a mexican drug trafficker, once tortured two students with ice picks and buried them alive for accidentally walking into his private dinner party."
