HN user

m3m3tic

62 karma
Posts2
Comments7
View on HN

temp 0 means that there will be no randomness injected into the response, and that for any given input you will get the exact same output, assuming the context window is also the same. Part of what makes an LLM more of a "thinking machine" than purely a "calculation machine" is that it will occasionally choose a less-probable next token than the statistically most likely token as a way of making the response more "flavorful" (or at least that's my understanding of why), and the likelihood of the response diverging from its most probable outcome is influenced by the temperature.

There's luck involved, but if you listen or read some of Marc Andreesen's ideas there's no doubt he's highly intelligent and thoughtful. Ben Horowitz even said he was the smartest person he'd ever met in his book. While he was obviously working on the right thing at the right time, his fortune is not what I would consider blind luck.

This is already possible with search engines, there is enough information on the internet that you can substantiate just about any claim regardless of how much evidence there is to the contrary. (see flat-earth, plenty of plausible sounding claims with real, albeit, cherry picked evidence).

"chronic stress aside, that decline in BMR mirrors almost perfectly the curves of PUFA consumption rates in the general population. Namely, as the BMR curve has steadily declined over the last 100 years the PUFA consumption rate curve has steadily moved upwards over time. Unless this trend of ever-increasing PUFA consumption is interrupted, I don’t see the decline of BMR flattening (let alone reversing) any time soon." eating massive quantities of highly processed seed oils that are ubiquitous in our food supply because they are much cheaper than "real" food is surely a contributor.

http://haidut.me/?p=2195